mirror of
https://github.com/GreptimeTeam/greptimedb.git
synced 2025-12-25 23:49:58 +00:00
Compare commits
36 Commits
v0.14.0-ni
...
v0.14.0
| Author | SHA1 | Date | |
|---|---|---|---|
|
66e2242e46 | ||
|
489b16ae30 | ||
|
85d564b0fb | ||
|
d5026f3491 | ||
|
e30753fc31 | ||
|
b476584f56 | ||
|
|
ff3a46b1d0 | ||
|
|
a533ac2555 | ||
|
|
cc5629b4a1 | ||
|
|
f3d000f6ec | ||
|
|
9557b76224 | ||
|
|
a0900f5b90 | ||
|
45a05fb08c | ||
|
71db79c8d6 | ||
|
|
79ed7bbc44 | ||
|
02e9a66d7a | ||
|
|
55cadcd2c0 | ||
|
8c4796734a | ||
|
919956999b | ||
|
7e5f6cbeae | ||
|
5c07f0dec7 | ||
|
|
9fb0487e67 | ||
|
|
6e407ae4b9 | ||
|
|
bcefc6b83f | ||
|
|
0f77135ef9 | ||
|
|
0a4594c9e2 | ||
|
|
d9437c6da7 | ||
|
|
35f4fa3c3e | ||
|
60e4607b64 | ||
|
|
3b8c6d5ce3 | ||
|
|
7a8e1bc3f9 | ||
|
|
ee07b9bfa8 | ||
|
90ffaa8a62 | ||
|
|
56f319a707 | ||
|
|
9df493988b | ||
|
|
ad1b77ab04 |
@@ -1,15 +0,0 @@
|
||||
# yaml-language-server: $schema=https://coderabbit.ai/integrations/schema.v2.json
|
||||
language: "en-US"
|
||||
early_access: false
|
||||
reviews:
|
||||
profile: "chill"
|
||||
request_changes_workflow: false
|
||||
high_level_summary: true
|
||||
poem: true
|
||||
review_status: true
|
||||
collapse_walkthrough: false
|
||||
auto_review:
|
||||
enabled: false
|
||||
drafts: false
|
||||
chat:
|
||||
auto_reply: true
|
||||
@@ -7,7 +7,8 @@ meta:
|
||||
provider = "kafka"
|
||||
broker_endpoints = ["kafka.kafka-cluster.svc.cluster.local:9092"]
|
||||
num_topics = 3
|
||||
auto_prune_topic_records = true
|
||||
auto_prune_interval = "30s"
|
||||
trigger_flush_threshold = 100
|
||||
|
||||
[datanode]
|
||||
[datanode.client]
|
||||
@@ -22,6 +23,7 @@ datanode:
|
||||
provider = "kafka"
|
||||
broker_endpoints = ["kafka.kafka-cluster.svc.cluster.local:9092"]
|
||||
linger = "2ms"
|
||||
overwrite_entry_start_id = true
|
||||
frontend:
|
||||
configData: |-
|
||||
[runtime]
|
||||
|
||||
30
.github/workflows/grafana.yml
vendored
30
.github/workflows/grafana.yml
vendored
@@ -21,32 +21,6 @@ jobs:
|
||||
run: sudo apt-get install -y jq
|
||||
|
||||
# Make the check.sh script executable
|
||||
- name: Make check.sh executable
|
||||
run: chmod +x grafana/check.sh
|
||||
|
||||
# Run the check.sh script
|
||||
- name: Run check.sh
|
||||
run: ./grafana/check.sh
|
||||
|
||||
# Only run summary.sh for pull_request events (not for merge queues or final pushes)
|
||||
- name: Check if this is a pull request
|
||||
id: check-pr
|
||||
- name: Check grafana dashboards
|
||||
run: |
|
||||
if [[ "${{ github.event_name }}" == "pull_request" ]]; then
|
||||
echo "is_pull_request=true" >> $GITHUB_OUTPUT
|
||||
else
|
||||
echo "is_pull_request=false" >> $GITHUB_OUTPUT
|
||||
fi
|
||||
|
||||
# Make the summary.sh script executable
|
||||
- name: Make summary.sh executable
|
||||
if: steps.check-pr.outputs.is_pull_request == 'true'
|
||||
run: chmod +x grafana/summary.sh
|
||||
|
||||
# Run the summary.sh script and add its output to the GitHub Job Summary
|
||||
- name: Run summary.sh and add to Job Summary
|
||||
if: steps.check-pr.outputs.is_pull_request == 'true'
|
||||
run: |
|
||||
SUMMARY=$(./grafana/summary.sh)
|
||||
echo "### Summary of Grafana Panels" >> $GITHUB_STEP_SUMMARY
|
||||
echo "$SUMMARY" >> $GITHUB_STEP_SUMMARY
|
||||
make check-dashboards
|
||||
|
||||
594
Cargo.lock
generated
594
Cargo.lock
generated
File diff suppressed because it is too large
Load Diff
25
Cargo.toml
25
Cargo.toml
@@ -112,15 +112,15 @@ clap = { version = "4.4", features = ["derive"] }
|
||||
config = "0.13.0"
|
||||
crossbeam-utils = "0.8"
|
||||
dashmap = "6.1"
|
||||
datafusion = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "5bbedc6704162afb03478f56ffb629405a4e1220" }
|
||||
datafusion-common = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "5bbedc6704162afb03478f56ffb629405a4e1220" }
|
||||
datafusion-expr = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "5bbedc6704162afb03478f56ffb629405a4e1220" }
|
||||
datafusion-functions = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "5bbedc6704162afb03478f56ffb629405a4e1220" }
|
||||
datafusion-optimizer = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "5bbedc6704162afb03478f56ffb629405a4e1220" }
|
||||
datafusion-physical-expr = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "5bbedc6704162afb03478f56ffb629405a4e1220" }
|
||||
datafusion-physical-plan = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "5bbedc6704162afb03478f56ffb629405a4e1220" }
|
||||
datafusion-sql = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "5bbedc6704162afb03478f56ffb629405a4e1220" }
|
||||
datafusion-substrait = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "5bbedc6704162afb03478f56ffb629405a4e1220" }
|
||||
datafusion = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "e104c7cf62b11dd5fe41461b82514978234326b4" }
|
||||
datafusion-common = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "e104c7cf62b11dd5fe41461b82514978234326b4" }
|
||||
datafusion-expr = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "e104c7cf62b11dd5fe41461b82514978234326b4" }
|
||||
datafusion-functions = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "e104c7cf62b11dd5fe41461b82514978234326b4" }
|
||||
datafusion-optimizer = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "e104c7cf62b11dd5fe41461b82514978234326b4" }
|
||||
datafusion-physical-expr = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "e104c7cf62b11dd5fe41461b82514978234326b4" }
|
||||
datafusion-physical-plan = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "e104c7cf62b11dd5fe41461b82514978234326b4" }
|
||||
datafusion-sql = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "e104c7cf62b11dd5fe41461b82514978234326b4" }
|
||||
datafusion-substrait = { git = "https://github.com/waynexia/arrow-datafusion.git", rev = "e104c7cf62b11dd5fe41461b82514978234326b4" }
|
||||
deadpool = "0.12"
|
||||
deadpool-postgres = "0.14"
|
||||
derive_builder = "0.20"
|
||||
@@ -129,7 +129,7 @@ etcd-client = "0.14"
|
||||
fst = "0.4.7"
|
||||
futures = "0.3"
|
||||
futures-util = "0.3"
|
||||
greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "b6d9cffd43c4e6358805a798f17e03e232994b82" }
|
||||
greptime-proto = { git = "https://github.com/GreptimeTeam/greptime-proto.git", rev = "e82b0158cd38d4021edb4e4c0ae77f999051e62f" }
|
||||
hex = "0.4"
|
||||
http = "1"
|
||||
humantime = "2.1"
|
||||
@@ -191,7 +191,7 @@ simd-json = "0.15"
|
||||
similar-asserts = "1.6.0"
|
||||
smallvec = { version = "1", features = ["serde"] }
|
||||
snafu = "0.8"
|
||||
sqlparser = { git = "https://github.com/GreptimeTeam/sqlparser-rs.git", rev = "e98e6b322426a9d397a71efef17075966223c089", features = [
|
||||
sqlparser = { git = "https://github.com/GreptimeTeam/sqlparser-rs.git", rev = "0cf6c04490d59435ee965edd2078e8855bd8471e", features = [
|
||||
"visitor",
|
||||
"serde",
|
||||
] } # branch = "v0.54.x"
|
||||
@@ -269,6 +269,9 @@ metric-engine = { path = "src/metric-engine" }
|
||||
mito2 = { path = "src/mito2" }
|
||||
object-store = { path = "src/object-store" }
|
||||
operator = { path = "src/operator" }
|
||||
otel-arrow-rust = { git = "https://github.com/open-telemetry/otel-arrow", rev = "5d551412d2a12e689cde4d84c14ef29e36784e51", features = [
|
||||
"server",
|
||||
] }
|
||||
partition = { path = "src/partition" }
|
||||
pipeline = { path = "src/pipeline" }
|
||||
plugins = { path = "src/plugins" }
|
||||
|
||||
10
Makefile
10
Makefile
@@ -222,6 +222,16 @@ start-cluster: ## Start the greptimedb cluster with etcd by using docker compose
|
||||
stop-cluster: ## Stop the greptimedb cluster that created by docker compose.
|
||||
docker compose -f ./docker/docker-compose/cluster-with-etcd.yaml stop
|
||||
|
||||
##@ Grafana
|
||||
|
||||
.PHONY: check-dashboards
|
||||
check-dashboards: ## Check the Grafana dashboards.
|
||||
@./grafana/scripts/check.sh
|
||||
|
||||
.PHONY: dashboards
|
||||
dashboards: ## Generate the Grafana dashboards for standalone mode and intermediate dashboards.
|
||||
@./grafana/scripts/gen-dashboards.sh
|
||||
|
||||
##@ Docs
|
||||
config-docs: ## Generate configuration documentation from toml files.
|
||||
docker run --rm \
|
||||
|
||||
@@ -6,7 +6,7 @@
|
||||
</picture>
|
||||
</p>
|
||||
|
||||
<h2 align="center">Unified & Cost-Effective Observability Database for Metrics, Logs, and Events</h2>
|
||||
<h2 align="center">Real-Time & Cloud-Native Observability Database<br/>for metrics, logs, and traces</h2>
|
||||
|
||||
<div align="center">
|
||||
<h3 align="center">
|
||||
|
||||
@@ -319,6 +319,7 @@
|
||||
| `selector` | String | `round_robin` | Datanode selector type.<br/>- `round_robin` (default value)<br/>- `lease_based`<br/>- `load_based`<br/>For details, please see "https://docs.greptime.com/developer-guide/metasrv/selector". |
|
||||
| `use_memory_store` | Bool | `false` | Store data in memory. |
|
||||
| `enable_region_failover` | Bool | `false` | Whether to enable region failover.<br/>This feature is only available on GreptimeDB running on cluster mode and<br/>- Using Remote WAL<br/>- Using shared storage (e.g., s3). |
|
||||
| `allow_region_failover_on_local_wal` | Bool | `false` | Whether to allow region failover on local WAL.<br/>**This option is not recommended to be set to true, because it may lead to data loss during failover.** |
|
||||
| `node_max_idle_time` | String | `24hours` | Max allowed idle time before removing node info from metasrv memory. |
|
||||
| `enable_telemetry` | Bool | `true` | Whether to enable greptimedb telemetry. Enabled by default. |
|
||||
| `runtime` | -- | -- | The runtime options. |
|
||||
|
||||
@@ -50,6 +50,10 @@ use_memory_store = false
|
||||
## - Using shared storage (e.g., s3).
|
||||
enable_region_failover = false
|
||||
|
||||
## Whether to allow region failover on local WAL.
|
||||
## **This option is not recommended to be set to true, because it may lead to data loss during failover.**
|
||||
allow_region_failover_on_local_wal = false
|
||||
|
||||
## Max allowed idle time before removing node info from metasrv memory.
|
||||
node_max_idle_time = "24hours"
|
||||
|
||||
|
||||
@@ -1,61 +1,89 @@
|
||||
Grafana dashboard for GreptimeDB
|
||||
--------------------------------
|
||||
# Grafana dashboards for GreptimeDB
|
||||
|
||||
GreptimeDB's official Grafana dashboard.
|
||||
## Overview
|
||||
|
||||
Status notify: we are still working on this config. It's expected to change frequently in the recent days. Please feel free to submit your feedback and/or contribution to this dashboard 🤗
|
||||
This repository maintains the Grafana dashboards for GreptimeDB. It has two types of dashboards:
|
||||
|
||||
If you use Helm [chart](https://github.com/GreptimeTeam/helm-charts) to deploy GreptimeDB cluster, you can enable self-monitoring by setting the following values in your Helm chart:
|
||||
- `cluster/dashboard.json`: The Grafana dashboard for the GreptimeDB cluster. Read the [dashboard.md](./dashboards/cluster/dashboard.md) for more details.
|
||||
- `standalone/dashboard.json`: The Grafana dashboard for the standalone GreptimeDB instance. **It's generated from the `cluster/dashboard.json` by removing the instance filter through the `make dashboards` command**. Read the [dashboard.md](./dashboards/standalone/dashboard.md) for more details.
|
||||
|
||||
As the rapid development of GreptimeDB, the metrics may be changed, and please feel free to submit your feedback and/or contribution to this dashboard 🤗
|
||||
|
||||
**NOTE**:
|
||||
|
||||
- The Grafana version should be greater than 9.0.
|
||||
|
||||
- If you want to modify the dashboards, you only need to modify the `cluster/dashboard.json` and run the `make dashboards` command to generate the `standalone/dashboard.json` and other related files.
|
||||
|
||||
To maintain the dashboards easily, we use the [`dac`](https://github.com/zyy17/dac) tool to generate the intermediate dashboards and markdown documents:
|
||||
|
||||
- `cluster/dashboard.yaml`: The intermediate dashboard for the GreptimeDB cluster.
|
||||
- `standalone/dashboard.yaml`: The intermediate dashboard for the standalone GreptimeDB instance.
|
||||
|
||||
## Data Sources
|
||||
|
||||
There are two data sources for the dashboards to fetch the metrics:
|
||||
|
||||
- **Prometheus**: Expose the metrics of GreptimeDB.
|
||||
- **Information Schema**: It is the MySQL port of the current monitored instance. The `overview` dashboard will use this datasource to show the information schema of the current instance.
|
||||
|
||||
## Instance Filters
|
||||
|
||||
To deploy the dashboards for multiple scenarios (K8s, bare metal, etc.), we prefer to use the `instance` label when filtering instances.
|
||||
|
||||
Additionally, we recommend including the `pod` label in the legend to make it easier to identify each instance, even though this field will be empty in bare metal scenarios.
|
||||
|
||||
For example, the following query is recommended:
|
||||
|
||||
```promql
|
||||
sum(process_resident_memory_bytes{instance=~"$datanode"}) by (instance, pod)
|
||||
```
|
||||
|
||||
And the legend will be like: `[{{instance}}]-[{{ pod }}]`.
|
||||
|
||||
## Deployment
|
||||
|
||||
### Helm
|
||||
|
||||
If you use the Helm [chart](https://github.com/GreptimeTeam/helm-charts) to deploy a GreptimeDB cluster, you can enable self-monitoring by setting the following values in your Helm chart:
|
||||
|
||||
- `monitoring.enabled=true`: Deploys a standalone GreptimeDB instance dedicated to monitoring the cluster;
|
||||
- `grafana.enabled=true`: Deploys Grafana and automatically imports the monitoring dashboard;
|
||||
|
||||
The standalone GreptimeDB instance will collect metrics from your cluster and the dashboard will be available in the Grafana UI. For detailed deployment instructions, please refer to our [Kubernetes deployment guide](https://docs.greptime.com/nightly/user-guide/deployments/deploy-on-kubernetes/getting-started).
|
||||
The standalone GreptimeDB instance will collect metrics from your cluster, and the dashboard will be available in the Grafana UI. For detailed deployment instructions, please refer to our [Kubernetes deployment guide](https://docs.greptime.com/nightly/user-guide/deployments/deploy-on-kubernetes/getting-started).
|
||||
|
||||
# How to use
|
||||
### Self-host Prometheus and import dashboards manually
|
||||
|
||||
## `greptimedb.json`
|
||||
1. **Configure Prometheus to scrape the cluster**
|
||||
|
||||
Open Grafana Dashboard page, choose `New` -> `Import`. And upload `greptimedb.json` file.
|
||||
The following is an example configuration(**Please modify it according to your actual situation**):
|
||||
|
||||
## `greptimedb-cluster.json`
|
||||
```yml
|
||||
# example config
|
||||
# only to indicate how to assign labels to each target
|
||||
# modify yours accordingly
|
||||
scrape_configs:
|
||||
- job_name: metasrv
|
||||
static_configs:
|
||||
- targets: ['<metasrv-ip>:<port>']
|
||||
|
||||
This cluster dashboard provides a comprehensive view of incoming requests, response statuses, and internal activities such as flush and compaction, with a layered structure from frontend to datanode. Designed with a focus on alert functionality, its primary aim is to highlight any anomalies in metrics, allowing users to quickly pinpoint the cause of errors.
|
||||
- job_name: datanode
|
||||
static_configs:
|
||||
- targets: ['<datanode0-ip>:<port>', '<datanode1-ip>:<port>', '<datanode2-ip>:<port>']
|
||||
|
||||
We use Prometheus to scrape off metrics from nodes in GreptimeDB cluster, Grafana to visualize the diagram. Any compatible stack should work too.
|
||||
- job_name: frontend
|
||||
static_configs:
|
||||
- targets: ['<frontend-ip>:<port>']
|
||||
```
|
||||
|
||||
__Note__: This dashboard is still in an early stage of development. Any issue or advice on improvement is welcomed.
|
||||
2. **Configure the data sources in Grafana**
|
||||
|
||||
### Configuration
|
||||
You need to add two data sources in Grafana:
|
||||
|
||||
Please ensure the following configuration before importing the dashboard into Grafana.
|
||||
- Prometheus: It is the Prometheus instance that scrapes the GreptimeDB metrics.
|
||||
- Information Schema: It is the MySQL port of the current monitored instance. The dashboard will use this datasource to show the information schema of the current instance.
|
||||
|
||||
__1. Prometheus scrape config__
|
||||
3. **Import the dashboards based on your deployment scenario**
|
||||
|
||||
Configure Prometheus to scrape the cluster.
|
||||
|
||||
```yml
|
||||
# example config
|
||||
# only to indicate how to assign labels to each target
|
||||
# modify yours accordingly
|
||||
scrape_configs:
|
||||
- job_name: metasrv
|
||||
static_configs:
|
||||
- targets: ['<metasrv-ip>:<port>']
|
||||
|
||||
- job_name: datanode
|
||||
static_configs:
|
||||
- targets: ['<datanode0-ip>:<port>', '<datanode1-ip>:<port>', '<datanode2-ip>:<port>']
|
||||
|
||||
- job_name: frontend
|
||||
static_configs:
|
||||
- targets: ['<frontend-ip>:<port>']
|
||||
```
|
||||
|
||||
__2. Grafana config__
|
||||
|
||||
Create a Prometheus data source in Grafana before using this dashboard. We use `datasource` as a variable in Grafana dashboard so that multiple environments are supported.
|
||||
|
||||
### Usage
|
||||
|
||||
Use `datasource` or `instance` on the upper-left corner to filter data from certain node.
|
||||
- **Cluster**: Import the `cluster/dashboard.json` dashboard.
|
||||
- **Standalone**: Import the `standalone/dashboard.json` dashboard.
|
||||
|
||||
@@ -1,19 +0,0 @@
|
||||
#!/usr/bin/env bash
|
||||
|
||||
BASEDIR=$(dirname "$0")
|
||||
|
||||
# Use jq to check for panels with empty or missing descriptions
|
||||
invalid_panels=$(cat $BASEDIR/greptimedb-cluster.json | jq -r '
|
||||
.panels[]
|
||||
| select((.type == "stats" or .type == "timeseries") and (.description == "" or .description == null))
|
||||
')
|
||||
|
||||
# Check if any invalid panels were found
|
||||
if [[ -n "$invalid_panels" ]]; then
|
||||
echo "Error: The following panels have empty or missing descriptions:"
|
||||
echo "$invalid_panels"
|
||||
exit 1
|
||||
else
|
||||
echo "All panels with type 'stats' or 'timeseries' have valid descriptions."
|
||||
exit 0
|
||||
fi
|
||||
7193
grafana/dashboards/cluster/dashboard.json
Normal file
7193
grafana/dashboards/cluster/dashboard.json
Normal file
File diff suppressed because it is too large
Load Diff
97
grafana/dashboards/cluster/dashboard.md
Normal file
97
grafana/dashboards/cluster/dashboard.md
Normal file
@@ -0,0 +1,97 @@
|
||||
# Overview
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Uptime | `time() - process_start_time_seconds` | `stat` | The start time of GreptimeDB. | `prometheus` | `s` | `__auto` |
|
||||
| Version | `SELECT pkg_version FROM information_schema.build_info` | `stat` | GreptimeDB version. | `mysql` | -- | -- |
|
||||
| Total Ingestion Rate | `sum(rate(greptime_table_operator_ingest_rows[$__rate_interval]))` | `stat` | Total ingestion rate. | `prometheus` | `rowsps` | `__auto` |
|
||||
| Total Storage Size | `select SUM(disk_size) from information_schema.region_statistics;` | `stat` | Total number of data file size. | `mysql` | `decbytes` | -- |
|
||||
| Total Rows | `select SUM(region_rows) from information_schema.region_statistics;` | `stat` | Total number of data rows in the cluster. Calculated by sum of rows from each region. | `mysql` | `sishort` | -- |
|
||||
| Deployment | `SELECT count(*) as datanode FROM information_schema.cluster_info WHERE peer_type = 'DATANODE';`<br/>`SELECT count(*) as frontend FROM information_schema.cluster_info WHERE peer_type = 'FRONTEND';`<br/>`SELECT count(*) as metasrv FROM information_schema.cluster_info WHERE peer_type = 'METASRV';`<br/>`SELECT count(*) as flownode FROM information_schema.cluster_info WHERE peer_type = 'FLOWNODE';` | `stat` | The deployment topology of GreptimeDB. | `mysql` | -- | -- |
|
||||
| Database Resources | `SELECT COUNT(*) as databases FROM information_schema.schemata WHERE schema_name NOT IN ('greptime_private', 'information_schema')`<br/>`SELECT COUNT(*) as tables FROM information_schema.tables WHERE table_schema != 'information_schema'`<br/>`SELECT COUNT(region_id) as regions FROM information_schema.region_peers`<br/>`SELECT COUNT(*) as flows FROM information_schema.flows` | `stat` | The number of the key resources in GreptimeDB. | `mysql` | -- | -- |
|
||||
| Data Size | `SELECT SUM(memtable_size) * 0.42825 as WAL FROM information_schema.region_statistics;`<br/>`SELECT SUM(index_size) as index FROM information_schema.region_statistics;`<br/>`SELECT SUM(manifest_size) as manifest FROM information_schema.region_statistics;` | `stat` | The data size of wal/index/manifest in the GreptimeDB. | `mysql` | `decbytes` | -- |
|
||||
# Ingestion
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Total Ingestion Rate | `sum(rate(greptime_table_operator_ingest_rows{instance=~"$frontend"}[$__rate_interval]))` | `timeseries` | Total ingestion rate.<br/><br/>Here we listed 3 primary protocols:<br/><br/>- Prometheus remote write<br/>- Greptime's gRPC API (when using our ingest SDK)<br/>- Log ingestion http API<br/> | `prometheus` | `rowsps` | `ingestion` |
|
||||
| Ingestion Rate by Type | `sum(rate(greptime_servers_http_logs_ingestion_counter[$__rate_interval]))`<br/>`sum(rate(greptime_servers_prometheus_remote_write_samples[$__rate_interval]))` | `timeseries` | Total ingestion rate.<br/><br/>Here we listed 3 primary protocols:<br/><br/>- Prometheus remote write<br/>- Greptime's gRPC API (when using our ingest SDK)<br/>- Log ingestion http API<br/> | `prometheus` | `rowsps` | `http-logs` |
|
||||
# Queries
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Total Query Rate | `sum (rate(greptime_servers_mysql_query_elapsed_count{instance=~"$frontend"}[$__rate_interval]))`<br/>`sum (rate(greptime_servers_postgres_query_elapsed_count{instance=~"$frontend"}[$__rate_interval]))`<br/>`sum (rate(greptime_servers_http_promql_elapsed_counte{instance=~"$frontend"}[$__rate_interval]))` | `timeseries` | Total rate of query API calls by protocol. This metric is collected from frontends.<br/><br/>Here we listed 3 main protocols:<br/>- MySQL<br/>- Postgres<br/>- Prometheus API<br/><br/>Note that there are some other minor query APIs like /sql are not included | `prometheus` | `reqps` | `mysql` |
|
||||
# Resources
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Datanode Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$datanode"}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{instance}}]-[{{ pod }}]` |
|
||||
| Datanode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$datanode"}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
|
||||
| Frontend Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$frontend"}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]` |
|
||||
| Frontend CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$frontend"}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]-cpu` |
|
||||
| Metasrv Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$metasrv"}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]-resident` |
|
||||
| Metasrv CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$metasrv"}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
|
||||
| Flownode Memory per Instance | `sum(process_resident_memory_bytes{instance=~"$flownode"}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]` |
|
||||
| Flownode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{instance=~"$flownode"}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
|
||||
# Frontend Requests
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| HTTP QPS per Instance | `sum by(instance, pod, path, method, code) (rate(greptime_servers_http_requests_elapsed_count{instance=~"$frontend",path!~"/health\|/metrics"}[$__rate_interval]))` | `timeseries` | HTTP QPS per Instance. | `prometheus` | `reqps` | `[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]` |
|
||||
| HTTP P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, path, method, code) (rate(greptime_servers_http_requests_elapsed_bucket{instance=~"$frontend",path!~"/health\|/metrics"}[$__rate_interval])))` | `timeseries` | HTTP P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]-p99` |
|
||||
| gRPC QPS per Instance | `sum by(instance, pod, path, code) (rate(greptime_servers_grpc_requests_elapsed_count{instance=~"$frontend"}[$__rate_interval]))` | `timeseries` | gRPC QPS per Instance. | `prometheus` | `reqps` | `[{{instance}}]-[{{pod}}]-[{{path}}]-[{{code}}]` |
|
||||
| gRPC P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, path, code) (rate(greptime_servers_grpc_requests_elapsed_bucket{instance=~"$frontend"}[$__rate_interval])))` | `timeseries` | gRPC P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]-p99` |
|
||||
| MySQL QPS per Instance | `sum by(pod, instance)(rate(greptime_servers_mysql_query_elapsed_count{instance=~"$frontend"}[$__rate_interval]))` | `timeseries` | MySQL QPS per Instance. | `prometheus` | `reqps` | `[{{instance}}]-[{{pod}}]` |
|
||||
| MySQL P99 per Instance | `histogram_quantile(0.99, sum by(pod, instance, le) (rate(greptime_servers_mysql_query_elapsed_bucket{instance=~"$frontend"}[$__rate_interval])))` | `timeseries` | MySQL P99 per Instance. | `prometheus` | `s` | `[{{ instance }}]-[{{ pod }}]-p99` |
|
||||
| PostgreSQL QPS per Instance | `sum by(pod, instance)(rate(greptime_servers_postgres_query_elapsed_count{instance=~"$frontend"}[$__rate_interval]))` | `timeseries` | PostgreSQL QPS per Instance. | `prometheus` | `reqps` | `[{{instance}}]-[{{pod}}]` |
|
||||
| PostgreSQL P99 per Instance | `histogram_quantile(0.99, sum by(pod,instance,le) (rate(greptime_servers_postgres_query_elapsed_bucket{instance=~"$frontend"}[$__rate_interval])))` | `timeseries` | PostgreSQL P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-p99` |
|
||||
# Frontend to Datanode
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Ingest Rows per Instance | `sum by(instance, pod)(rate(greptime_table_operator_ingest_rows{instance=~"$frontend"}[$__rate_interval]))` | `timeseries` | Ingestion rate by row as in each frontend | `prometheus` | `rowsps` | `[{{instance}}]-[{{pod}}]` |
|
||||
| Region Call QPS per Instance | `sum by(instance, pod, request_type) (rate(greptime_grpc_region_request_count{instance=~"$frontend"}[$__rate_interval]))` | `timeseries` | Region Call QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{request_type}}]` |
|
||||
| Region Call P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, request_type) (rate(greptime_grpc_region_request_bucket{instance=~"$frontend"}[$__rate_interval])))` | `timeseries` | Region Call P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{request_type}}]` |
|
||||
# Mito Engine
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Request OPS per Instance | `sum by(instance, pod, type) (rate(greptime_mito_handle_request_elapsed_count{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | Request QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{type}}]` |
|
||||
| Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, type) (rate(greptime_mito_handle_request_elapsed_bucket{instance=~"$datanode"}[$__rate_interval])))` | `timeseries` | Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{type}}]` |
|
||||
| Write Buffer per Instance | `greptime_mito_write_buffer_bytes{instance=~"$datanode"}` | `timeseries` | Write Buffer per Instance. | `prometheus` | `decbytes` | `[{{instance}}]-[{{pod}}]` |
|
||||
| Write Rows per Instance | `sum by (instance, pod) (rate(greptime_mito_write_rows_total{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | Ingestion size by row counts. | `prometheus` | `rowsps` | `[{{instance}}]-[{{pod}}]` |
|
||||
| Flush OPS per Instance | `sum by(instance, pod, reason) (rate(greptime_mito_flush_requests_total{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | Flush QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{reason}}]` |
|
||||
| Write Stall per Instance | `sum by(instance, pod) (greptime_mito_write_stall_total{instance=~"$datanode"})` | `timeseries` | Write Stall per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]` |
|
||||
| Read Stage OPS per Instance | `sum by(instance, pod) (rate(greptime_mito_read_stage_elapsed_count{instance=~"$datanode", stage="total"}[$__rate_interval]))` | `timeseries` | Read Stage OPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]` |
|
||||
| Read Stage P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_read_stage_elapsed_bucket{instance=~"$datanode"}[$__rate_interval])))` | `timeseries` | Read Stage P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]` |
|
||||
| Write Stage P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_write_stage_elapsed_bucket{instance=~"$datanode"}[$__rate_interval])))` | `timeseries` | Write Stage P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]` |
|
||||
| Compaction OPS per Instance | `sum by(instance, pod) (rate(greptime_mito_compaction_total_elapsed_count{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | Compaction OPS per Instance. | `prometheus` | `ops` | `[{{ instance }}]-[{{pod}}]` |
|
||||
| Compaction P99 per Instance by Stage | `histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_compaction_stage_elapsed_bucket{instance=~"$datanode"}[$__rate_interval])))` | `timeseries` | Compaction latency by stage | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]-p99` |
|
||||
| Compaction P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le,stage) (rate(greptime_mito_compaction_total_elapsed_bucket{instance=~"$datanode"}[$__rate_interval])))` | `timeseries` | Compaction P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]-compaction` |
|
||||
| WAL write size | `histogram_quantile(0.95, sum by(le,instance, pod) (rate(raft_engine_write_size_bucket[$__rate_interval])))`<br/>`histogram_quantile(0.99, sum by(le,instance,pod) (rate(raft_engine_write_size_bucket[$__rate_interval])))`<br/>`sum by (instance, pod)(rate(raft_engine_write_size_sum[$__rate_interval]))` | `timeseries` | Write-ahead logs write size as bytes. This chart includes stats of p95 and p99 size by instance, total WAL write rate. | `prometheus` | `bytes` | `[{{instance}}]-[{{pod}}]-req-size-p95` |
|
||||
| Cached Bytes per Instance | `greptime_mito_cache_bytes{instance=~"$datanode"}` | `timeseries` | Cached Bytes per Instance. | `prometheus` | `decbytes` | `[{{instance}}]-[{{pod}}]-[{{type}}]` |
|
||||
| Inflight Compaction | `greptime_mito_inflight_compaction_count` | `timeseries` | Ongoing compaction task count | `prometheus` | `none` | `[{{instance}}]-[{{pod}}]` |
|
||||
| WAL sync duration seconds | `histogram_quantile(0.99, sum by(le, type, node, instance, pod) (rate(raft_engine_sync_log_duration_seconds_bucket[$__rate_interval])))` | `timeseries` | Raft engine (local disk) log store sync latency, p99 | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-p99` |
|
||||
| Log Store op duration seconds | `histogram_quantile(0.99, sum by(le,logstore,optype,instance, pod) (rate(greptime_logstore_op_elapsed_bucket[$__rate_interval])))` | `timeseries` | Write-ahead log operations latency at p99 | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{logstore}}]-[{{optype}}]-p99` |
|
||||
| Inflight Flush | `greptime_mito_inflight_flush_count` | `timeseries` | Ongoing flush task count | `prometheus` | `none` | `[{{instance}}]-[{{pod}}]` |
|
||||
# OpenDAL
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
|
||||
| Read QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="read"}[$__rate_interval]))` | `timeseries` | Read QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
|
||||
| Read P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode",operation="read"}[$__rate_interval])))` | `timeseries` | Read P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
|
||||
| Write QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="write"}[$__rate_interval]))` | `timeseries` | Write QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
|
||||
| Write P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation="write"}[$__rate_interval])))` | `timeseries` | Write P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
|
||||
| List QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="list"}[$__rate_interval]))` | `timeseries` | List QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
|
||||
| List P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation="list"}[$__rate_interval])))` | `timeseries` | List P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
|
||||
| Other Requests per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode",operation!~"read\|write\|list\|stat"}[$__rate_interval]))` | `timeseries` | Other Requests per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
|
||||
| Other Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation!~"read\|write\|list"}[$__rate_interval])))` | `timeseries` | Other Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
|
||||
| Opendal traffic | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_bytes_sum{instance=~"$datanode"}[$__rate_interval]))` | `timeseries` | Total traffic as in bytes by instance and operation | `prometheus` | `decbytes` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
|
||||
| OpenDAL errors per Instance | `sum by(instance, pod, scheme, operation, error) (rate(opendal_operation_errors_total{instance=~"$datanode", error!="NotFound"}[$__rate_interval]))` | `timeseries` | OpenDAL error counts per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]-[{{error}}]` |
|
||||
# Metasrv
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Region migration datanode | `greptime_meta_region_migration_stat{datanode_type="src"}`<br/>`greptime_meta_region_migration_stat{datanode_type="desc"}` | `state-timeline` | Counter of region migration by source and destination | `prometheus` | `none` | `from-datanode-{{datanode_id}}` |
|
||||
| Region migration error | `greptime_meta_region_migration_error` | `timeseries` | Counter of region migration error | `prometheus` | `none` | `__auto` |
|
||||
| Datanode load | `greptime_datanode_load` | `timeseries` | Gauge of load information of each datanode, collected via heartbeat between datanode and metasrv. This information is for metasrv to schedule workloads. | `prometheus` | `none` | `__auto` |
|
||||
# Flownode
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Flow Ingest / Output Rate | `sum by(instance, pod, direction) (rate(greptime_flow_processed_rows[$__rate_interval]))` | `timeseries` | Flow Ingest / Output Rate. | `prometheus` | -- | `[{{pod}}]-[{{instance}}]-[{{direction}}]` |
|
||||
| Flow Ingest Latency | `histogram_quantile(0.95, sum(rate(greptime_flow_insert_elapsed_bucket[$__rate_interval])) by (le, instance, pod))`<br/>`histogram_quantile(0.99, sum(rate(greptime_flow_insert_elapsed_bucket[$__rate_interval])) by (le, instance, pod))` | `timeseries` | Flow Ingest Latency. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-p95` |
|
||||
| Flow Operation Latency | `histogram_quantile(0.95, sum(rate(greptime_flow_processing_time_bucket[$__rate_interval])) by (le,instance,pod,type))`<br/>`histogram_quantile(0.99, sum(rate(greptime_flow_processing_time_bucket[$__rate_interval])) by (le,instance,pod,type))` | `timeseries` | Flow Operation Latency. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{type}}]-p95` |
|
||||
| Flow Buffer Size per Instance | `greptime_flow_input_buf_size` | `timeseries` | Flow Buffer Size per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}]` |
|
||||
| Flow Processing Error per Instance | `sum by(instance,pod,code) (rate(greptime_flow_errors[$__rate_interval]))` | `timeseries` | Flow Processing Error per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{code}}]` |
|
||||
769
grafana/dashboards/cluster/dashboard.yaml
Normal file
769
grafana/dashboards/cluster/dashboard.yaml
Normal file
@@ -0,0 +1,769 @@
|
||||
groups:
|
||||
- title: Overview
|
||||
panels:
|
||||
- title: Uptime
|
||||
type: stat
|
||||
description: The start time of GreptimeDB.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: time() - process_start_time_seconds
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: __auto
|
||||
- title: Version
|
||||
type: stat
|
||||
description: GreptimeDB version.
|
||||
queries:
|
||||
- expr: SELECT pkg_version FROM information_schema.build_info
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Total Ingestion Rate
|
||||
type: stat
|
||||
description: Total ingestion rate.
|
||||
unit: rowsps
|
||||
queries:
|
||||
- expr: sum(rate(greptime_table_operator_ingest_rows[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: __auto
|
||||
- title: Total Storage Size
|
||||
type: stat
|
||||
description: Total number of data file size.
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: select SUM(disk_size) from information_schema.region_statistics;
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Total Rows
|
||||
type: stat
|
||||
description: Total number of data rows in the cluster. Calculated by sum of rows from each region.
|
||||
unit: sishort
|
||||
queries:
|
||||
- expr: select SUM(region_rows) from information_schema.region_statistics;
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Deployment
|
||||
type: stat
|
||||
description: The deployment topology of GreptimeDB.
|
||||
queries:
|
||||
- expr: SELECT count(*) as datanode FROM information_schema.cluster_info WHERE peer_type = 'DATANODE';
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT count(*) as frontend FROM information_schema.cluster_info WHERE peer_type = 'FRONTEND';
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT count(*) as metasrv FROM information_schema.cluster_info WHERE peer_type = 'METASRV';
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT count(*) as flownode FROM information_schema.cluster_info WHERE peer_type = 'FLOWNODE';
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Database Resources
|
||||
type: stat
|
||||
description: The number of the key resources in GreptimeDB.
|
||||
queries:
|
||||
- expr: SELECT COUNT(*) as databases FROM information_schema.schemata WHERE schema_name NOT IN ('greptime_private', 'information_schema')
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT COUNT(*) as tables FROM information_schema.tables WHERE table_schema != 'information_schema'
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT COUNT(region_id) as regions FROM information_schema.region_peers
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT COUNT(*) as flows FROM information_schema.flows
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Data Size
|
||||
type: stat
|
||||
description: The data size of wal/index/manifest in the GreptimeDB.
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: SELECT SUM(memtable_size) * 0.42825 as WAL FROM information_schema.region_statistics;
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT SUM(index_size) as index FROM information_schema.region_statistics;
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT SUM(manifest_size) as manifest FROM information_schema.region_statistics;
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Ingestion
|
||||
panels:
|
||||
- title: Total Ingestion Rate
|
||||
type: timeseries
|
||||
description: |
|
||||
Total ingestion rate.
|
||||
|
||||
Here we listed 3 primary protocols:
|
||||
|
||||
- Prometheus remote write
|
||||
- Greptime's gRPC API (when using our ingest SDK)
|
||||
- Log ingestion http API
|
||||
unit: rowsps
|
||||
queries:
|
||||
- expr: sum(rate(greptime_table_operator_ingest_rows{instance=~"$frontend"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: ingestion
|
||||
- title: Ingestion Rate by Type
|
||||
type: timeseries
|
||||
description: |
|
||||
Total ingestion rate.
|
||||
|
||||
Here we listed 3 primary protocols:
|
||||
|
||||
- Prometheus remote write
|
||||
- Greptime's gRPC API (when using our ingest SDK)
|
||||
- Log ingestion http API
|
||||
unit: rowsps
|
||||
queries:
|
||||
- expr: sum(rate(greptime_servers_http_logs_ingestion_counter[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: http-logs
|
||||
- expr: sum(rate(greptime_servers_prometheus_remote_write_samples[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: prometheus-remote-write
|
||||
- title: Queries
|
||||
panels:
|
||||
- title: Total Query Rate
|
||||
type: timeseries
|
||||
description: |-
|
||||
Total rate of query API calls by protocol. This metric is collected from frontends.
|
||||
|
||||
Here we listed 3 main protocols:
|
||||
- MySQL
|
||||
- Postgres
|
||||
- Prometheus API
|
||||
|
||||
Note that there are some other minor query APIs like /sql are not included
|
||||
unit: reqps
|
||||
queries:
|
||||
- expr: sum (rate(greptime_servers_mysql_query_elapsed_count{instance=~"$frontend"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: mysql
|
||||
- expr: sum (rate(greptime_servers_postgres_query_elapsed_count{instance=~"$frontend"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: pg
|
||||
- expr: sum (rate(greptime_servers_http_promql_elapsed_counte{instance=~"$frontend"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: promql
|
||||
- title: Resources
|
||||
panels:
|
||||
- title: Datanode Memory per Instance
|
||||
type: timeseries
|
||||
description: Current memory usage by instance
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: sum(process_resident_memory_bytes{instance=~"$datanode"}) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{ pod }}]'
|
||||
- title: Datanode CPU Usage per Instance
|
||||
type: timeseries
|
||||
description: Current cpu usage by instance
|
||||
unit: none
|
||||
queries:
|
||||
- expr: sum(rate(process_cpu_seconds_total{instance=~"$datanode"}[$__rate_interval]) * 1000) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]'
|
||||
- title: Frontend Memory per Instance
|
||||
type: timeseries
|
||||
description: Current memory usage by instance
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: sum(process_resident_memory_bytes{instance=~"$frontend"}) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]'
|
||||
- title: Frontend CPU Usage per Instance
|
||||
type: timeseries
|
||||
description: Current cpu usage by instance
|
||||
unit: none
|
||||
queries:
|
||||
- expr: sum(rate(process_cpu_seconds_total{instance=~"$frontend"}[$__rate_interval]) * 1000) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]-cpu'
|
||||
- title: Metasrv Memory per Instance
|
||||
type: timeseries
|
||||
description: Current memory usage by instance
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: sum(process_resident_memory_bytes{instance=~"$metasrv"}) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]-resident'
|
||||
- title: Metasrv CPU Usage per Instance
|
||||
type: timeseries
|
||||
description: Current cpu usage by instance
|
||||
unit: none
|
||||
queries:
|
||||
- expr: sum(rate(process_cpu_seconds_total{instance=~"$metasrv"}[$__rate_interval]) * 1000) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]'
|
||||
- title: Flownode Memory per Instance
|
||||
type: timeseries
|
||||
description: Current memory usage by instance
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: sum(process_resident_memory_bytes{instance=~"$flownode"}) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]'
|
||||
- title: Flownode CPU Usage per Instance
|
||||
type: timeseries
|
||||
description: Current cpu usage by instance
|
||||
unit: none
|
||||
queries:
|
||||
- expr: sum(rate(process_cpu_seconds_total{instance=~"$flownode"}[$__rate_interval]) * 1000) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]'
|
||||
- title: Frontend Requests
|
||||
panels:
|
||||
- title: HTTP QPS per Instance
|
||||
type: timeseries
|
||||
description: HTTP QPS per Instance.
|
||||
unit: reqps
|
||||
queries:
|
||||
- expr: sum by(instance, pod, path, method, code) (rate(greptime_servers_http_requests_elapsed_count{instance=~"$frontend",path!~"/health|/metrics"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]'
|
||||
- title: HTTP P99 per Instance
|
||||
type: timeseries
|
||||
description: HTTP P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, path, method, code) (rate(greptime_servers_http_requests_elapsed_bucket{instance=~"$frontend",path!~"/health|/metrics"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]-p99'
|
||||
- title: gRPC QPS per Instance
|
||||
type: timeseries
|
||||
description: gRPC QPS per Instance.
|
||||
unit: reqps
|
||||
queries:
|
||||
- expr: sum by(instance, pod, path, code) (rate(greptime_servers_grpc_requests_elapsed_count{instance=~"$frontend"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{path}}]-[{{code}}]'
|
||||
- title: gRPC P99 per Instance
|
||||
type: timeseries
|
||||
description: gRPC P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, path, code) (rate(greptime_servers_grpc_requests_elapsed_bucket{instance=~"$frontend"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]-p99'
|
||||
- title: MySQL QPS per Instance
|
||||
type: timeseries
|
||||
description: MySQL QPS per Instance.
|
||||
unit: reqps
|
||||
queries:
|
||||
- expr: sum by(pod, instance)(rate(greptime_servers_mysql_query_elapsed_count{instance=~"$frontend"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: MySQL P99 per Instance
|
||||
type: timeseries
|
||||
description: MySQL P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(pod, instance, le) (rate(greptime_servers_mysql_query_elapsed_bucket{instance=~"$frontend"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]-p99'
|
||||
- title: PostgreSQL QPS per Instance
|
||||
type: timeseries
|
||||
description: PostgreSQL QPS per Instance.
|
||||
unit: reqps
|
||||
queries:
|
||||
- expr: sum by(pod, instance)(rate(greptime_servers_postgres_query_elapsed_count{instance=~"$frontend"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: PostgreSQL P99 per Instance
|
||||
type: timeseries
|
||||
description: PostgreSQL P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(pod,instance,le) (rate(greptime_servers_postgres_query_elapsed_bucket{instance=~"$frontend"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-p99'
|
||||
- title: Frontend to Datanode
|
||||
panels:
|
||||
- title: Ingest Rows per Instance
|
||||
type: timeseries
|
||||
description: Ingestion rate by row as in each frontend
|
||||
unit: rowsps
|
||||
queries:
|
||||
- expr: sum by(instance, pod)(rate(greptime_table_operator_ingest_rows{instance=~"$frontend"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: Region Call QPS per Instance
|
||||
type: timeseries
|
||||
description: Region Call QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, request_type) (rate(greptime_grpc_region_request_count{instance=~"$frontend"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{request_type}}]'
|
||||
- title: Region Call P99 per Instance
|
||||
type: timeseries
|
||||
description: Region Call P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, request_type) (rate(greptime_grpc_region_request_bucket{instance=~"$frontend"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{request_type}}]'
|
||||
- title: Mito Engine
|
||||
panels:
|
||||
- title: Request OPS per Instance
|
||||
type: timeseries
|
||||
description: Request QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, type) (rate(greptime_mito_handle_request_elapsed_count{instance=~"$datanode"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]'
|
||||
- title: Request P99 per Instance
|
||||
type: timeseries
|
||||
description: Request P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, type) (rate(greptime_mito_handle_request_elapsed_bucket{instance=~"$datanode"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]'
|
||||
- title: Write Buffer per Instance
|
||||
type: timeseries
|
||||
description: Write Buffer per Instance.
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: greptime_mito_write_buffer_bytes{instance=~"$datanode"}
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: Write Rows per Instance
|
||||
type: timeseries
|
||||
description: Ingestion size by row counts.
|
||||
unit: rowsps
|
||||
queries:
|
||||
- expr: sum by (instance, pod) (rate(greptime_mito_write_rows_total{instance=~"$datanode"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: Flush OPS per Instance
|
||||
type: timeseries
|
||||
description: Flush QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, reason) (rate(greptime_mito_flush_requests_total{instance=~"$datanode"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{reason}}]'
|
||||
- title: Write Stall per Instance
|
||||
type: timeseries
|
||||
description: Write Stall per Instance.
|
||||
queries:
|
||||
- expr: sum by(instance, pod) (greptime_mito_write_stall_total{instance=~"$datanode"})
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: Read Stage OPS per Instance
|
||||
type: timeseries
|
||||
description: Read Stage OPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod) (rate(greptime_mito_read_stage_elapsed_count{instance=~"$datanode", stage="total"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: Read Stage P99 per Instance
|
||||
type: timeseries
|
||||
description: Read Stage P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_read_stage_elapsed_bucket{instance=~"$datanode"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{stage}}]'
|
||||
- title: Write Stage P99 per Instance
|
||||
type: timeseries
|
||||
description: Write Stage P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_write_stage_elapsed_bucket{instance=~"$datanode"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{stage}}]'
|
||||
- title: Compaction OPS per Instance
|
||||
type: timeseries
|
||||
description: Compaction OPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod) (rate(greptime_mito_compaction_total_elapsed_count{instance=~"$datanode"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{pod}}]'
|
||||
- title: Compaction P99 per Instance by Stage
|
||||
type: timeseries
|
||||
description: Compaction latency by stage
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_compaction_stage_elapsed_bucket{instance=~"$datanode"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{stage}}]-p99'
|
||||
- title: Compaction P99 per Instance
|
||||
type: timeseries
|
||||
description: Compaction P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le,stage) (rate(greptime_mito_compaction_total_elapsed_bucket{instance=~"$datanode"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{stage}}]-compaction'
|
||||
- title: WAL write size
|
||||
type: timeseries
|
||||
description: Write-ahead logs write size as bytes. This chart includes stats of p95 and p99 size by instance, total WAL write rate.
|
||||
unit: bytes
|
||||
queries:
|
||||
- expr: histogram_quantile(0.95, sum by(le,instance, pod) (rate(raft_engine_write_size_bucket[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-req-size-p95'
|
||||
- expr: histogram_quantile(0.99, sum by(le,instance,pod) (rate(raft_engine_write_size_bucket[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-req-size-p99'
|
||||
- expr: sum by (instance, pod)(rate(raft_engine_write_size_sum[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-throughput'
|
||||
- title: Cached Bytes per Instance
|
||||
type: timeseries
|
||||
description: Cached Bytes per Instance.
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: greptime_mito_cache_bytes{instance=~"$datanode"}
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]'
|
||||
- title: Inflight Compaction
|
||||
type: timeseries
|
||||
description: Ongoing compaction task count
|
||||
unit: none
|
||||
queries:
|
||||
- expr: greptime_mito_inflight_compaction_count
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: WAL sync duration seconds
|
||||
type: timeseries
|
||||
description: Raft engine (local disk) log store sync latency, p99
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(le, type, node, instance, pod) (rate(raft_engine_sync_log_duration_seconds_bucket[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-p99'
|
||||
- title: Log Store op duration seconds
|
||||
type: timeseries
|
||||
description: Write-ahead log operations latency at p99
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(le,logstore,optype,instance, pod) (rate(greptime_logstore_op_elapsed_bucket[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{logstore}}]-[{{optype}}]-p99'
|
||||
- title: Inflight Flush
|
||||
type: timeseries
|
||||
description: Ongoing flush task count
|
||||
unit: none
|
||||
queries:
|
||||
- expr: greptime_mito_inflight_flush_count
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: OpenDAL
|
||||
panels:
|
||||
- title: QPS per Instance
|
||||
type: timeseries
|
||||
description: QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
|
||||
- title: Read QPS per Instance
|
||||
type: timeseries
|
||||
description: Read QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="read"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
|
||||
- title: Read P99 per Instance
|
||||
type: timeseries
|
||||
description: Read P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode",operation="read"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
|
||||
- title: Write QPS per Instance
|
||||
type: timeseries
|
||||
description: Write QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="write"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
|
||||
- title: Write P99 per Instance
|
||||
type: timeseries
|
||||
description: Write P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation="write"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
|
||||
- title: List QPS per Instance
|
||||
type: timeseries
|
||||
description: List QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode", operation="list"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
|
||||
- title: List P99 per Instance
|
||||
type: timeseries
|
||||
description: List P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation="list"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
|
||||
- title: Other Requests per Instance
|
||||
type: timeseries
|
||||
description: Other Requests per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{instance=~"$datanode",operation!~"read|write|list|stat"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
|
||||
- title: Other Request P99 per Instance
|
||||
type: timeseries
|
||||
description: Other Request P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{instance=~"$datanode", operation!~"read|write|list"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
|
||||
- title: Opendal traffic
|
||||
type: timeseries
|
||||
description: Total traffic as in bytes by instance and operation
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_bytes_sum{instance=~"$datanode"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
|
||||
- title: OpenDAL errors per Instance
|
||||
type: timeseries
|
||||
description: OpenDAL error counts per Instance.
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme, operation, error) (rate(opendal_operation_errors_total{instance=~"$datanode", error!="NotFound"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]-[{{error}}]'
|
||||
- title: Metasrv
|
||||
panels:
|
||||
- title: Region migration datanode
|
||||
type: state-timeline
|
||||
description: Counter of region migration by source and destination
|
||||
unit: none
|
||||
queries:
|
||||
- expr: greptime_meta_region_migration_stat{datanode_type="src"}
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: from-datanode-{{datanode_id}}
|
||||
- expr: greptime_meta_region_migration_stat{datanode_type="desc"}
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: to-datanode-{{datanode_id}}
|
||||
- title: Region migration error
|
||||
type: timeseries
|
||||
description: Counter of region migration error
|
||||
unit: none
|
||||
queries:
|
||||
- expr: greptime_meta_region_migration_error
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: __auto
|
||||
- title: Datanode load
|
||||
type: timeseries
|
||||
description: Gauge of load information of each datanode, collected via heartbeat between datanode and metasrv. This information is for metasrv to schedule workloads.
|
||||
unit: none
|
||||
queries:
|
||||
- expr: greptime_datanode_load
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: __auto
|
||||
- title: Flownode
|
||||
panels:
|
||||
- title: Flow Ingest / Output Rate
|
||||
type: timeseries
|
||||
description: Flow Ingest / Output Rate.
|
||||
queries:
|
||||
- expr: sum by(instance, pod, direction) (rate(greptime_flow_processed_rows[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{pod}}]-[{{instance}}]-[{{direction}}]'
|
||||
- title: Flow Ingest Latency
|
||||
type: timeseries
|
||||
description: Flow Ingest Latency.
|
||||
queries:
|
||||
- expr: histogram_quantile(0.95, sum(rate(greptime_flow_insert_elapsed_bucket[$__rate_interval])) by (le, instance, pod))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-p95'
|
||||
- expr: histogram_quantile(0.99, sum(rate(greptime_flow_insert_elapsed_bucket[$__rate_interval])) by (le, instance, pod))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-p99'
|
||||
- title: Flow Operation Latency
|
||||
type: timeseries
|
||||
description: Flow Operation Latency.
|
||||
queries:
|
||||
- expr: histogram_quantile(0.95, sum(rate(greptime_flow_processing_time_bucket[$__rate_interval])) by (le,instance,pod,type))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]-p95'
|
||||
- expr: histogram_quantile(0.99, sum(rate(greptime_flow_processing_time_bucket[$__rate_interval])) by (le,instance,pod,type))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]-p99'
|
||||
- title: Flow Buffer Size per Instance
|
||||
type: timeseries
|
||||
description: Flow Buffer Size per Instance.
|
||||
queries:
|
||||
- expr: greptime_flow_input_buf_size
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}]'
|
||||
- title: Flow Processing Error per Instance
|
||||
type: timeseries
|
||||
description: Flow Processing Error per Instance.
|
||||
queries:
|
||||
- expr: sum by(instance,pod,code) (rate(greptime_flow_errors[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{code}}]'
|
||||
7193
grafana/dashboards/standalone/dashboard.json
Normal file
7193
grafana/dashboards/standalone/dashboard.json
Normal file
File diff suppressed because it is too large
Load Diff
97
grafana/dashboards/standalone/dashboard.md
Normal file
97
grafana/dashboards/standalone/dashboard.md
Normal file
@@ -0,0 +1,97 @@
|
||||
# Overview
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Uptime | `time() - process_start_time_seconds` | `stat` | The start time of GreptimeDB. | `prometheus` | `s` | `__auto` |
|
||||
| Version | `SELECT pkg_version FROM information_schema.build_info` | `stat` | GreptimeDB version. | `mysql` | -- | -- |
|
||||
| Total Ingestion Rate | `sum(rate(greptime_table_operator_ingest_rows[$__rate_interval]))` | `stat` | Total ingestion rate. | `prometheus` | `rowsps` | `__auto` |
|
||||
| Total Storage Size | `select SUM(disk_size) from information_schema.region_statistics;` | `stat` | Total number of data file size. | `mysql` | `decbytes` | -- |
|
||||
| Total Rows | `select SUM(region_rows) from information_schema.region_statistics;` | `stat` | Total number of data rows in the cluster. Calculated by sum of rows from each region. | `mysql` | `sishort` | -- |
|
||||
| Deployment | `SELECT count(*) as datanode FROM information_schema.cluster_info WHERE peer_type = 'DATANODE';`<br/>`SELECT count(*) as frontend FROM information_schema.cluster_info WHERE peer_type = 'FRONTEND';`<br/>`SELECT count(*) as metasrv FROM information_schema.cluster_info WHERE peer_type = 'METASRV';`<br/>`SELECT count(*) as flownode FROM information_schema.cluster_info WHERE peer_type = 'FLOWNODE';` | `stat` | The deployment topology of GreptimeDB. | `mysql` | -- | -- |
|
||||
| Database Resources | `SELECT COUNT(*) as databases FROM information_schema.schemata WHERE schema_name NOT IN ('greptime_private', 'information_schema')`<br/>`SELECT COUNT(*) as tables FROM information_schema.tables WHERE table_schema != 'information_schema'`<br/>`SELECT COUNT(region_id) as regions FROM information_schema.region_peers`<br/>`SELECT COUNT(*) as flows FROM information_schema.flows` | `stat` | The number of the key resources in GreptimeDB. | `mysql` | -- | -- |
|
||||
| Data Size | `SELECT SUM(memtable_size) * 0.42825 as WAL FROM information_schema.region_statistics;`<br/>`SELECT SUM(index_size) as index FROM information_schema.region_statistics;`<br/>`SELECT SUM(manifest_size) as manifest FROM information_schema.region_statistics;` | `stat` | The data size of wal/index/manifest in the GreptimeDB. | `mysql` | `decbytes` | -- |
|
||||
# Ingestion
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Total Ingestion Rate | `sum(rate(greptime_table_operator_ingest_rows{}[$__rate_interval]))` | `timeseries` | Total ingestion rate.<br/><br/>Here we listed 3 primary protocols:<br/><br/>- Prometheus remote write<br/>- Greptime's gRPC API (when using our ingest SDK)<br/>- Log ingestion http API<br/> | `prometheus` | `rowsps` | `ingestion` |
|
||||
| Ingestion Rate by Type | `sum(rate(greptime_servers_http_logs_ingestion_counter[$__rate_interval]))`<br/>`sum(rate(greptime_servers_prometheus_remote_write_samples[$__rate_interval]))` | `timeseries` | Total ingestion rate.<br/><br/>Here we listed 3 primary protocols:<br/><br/>- Prometheus remote write<br/>- Greptime's gRPC API (when using our ingest SDK)<br/>- Log ingestion http API<br/> | `prometheus` | `rowsps` | `http-logs` |
|
||||
# Queries
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Total Query Rate | `sum (rate(greptime_servers_mysql_query_elapsed_count{}[$__rate_interval]))`<br/>`sum (rate(greptime_servers_postgres_query_elapsed_count{}[$__rate_interval]))`<br/>`sum (rate(greptime_servers_http_promql_elapsed_counte{}[$__rate_interval]))` | `timeseries` | Total rate of query API calls by protocol. This metric is collected from frontends.<br/><br/>Here we listed 3 main protocols:<br/>- MySQL<br/>- Postgres<br/>- Prometheus API<br/><br/>Note that there are some other minor query APIs like /sql are not included | `prometheus` | `reqps` | `mysql` |
|
||||
# Resources
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Datanode Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{instance}}]-[{{ pod }}]` |
|
||||
| Datanode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
|
||||
| Frontend Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]` |
|
||||
| Frontend CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]-cpu` |
|
||||
| Metasrv Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]-resident` |
|
||||
| Metasrv CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
|
||||
| Flownode Memory per Instance | `sum(process_resident_memory_bytes{}) by (instance, pod)` | `timeseries` | Current memory usage by instance | `prometheus` | `decbytes` | `[{{ instance }}]-[{{ pod }}]` |
|
||||
| Flownode CPU Usage per Instance | `sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)` | `timeseries` | Current cpu usage by instance | `prometheus` | `none` | `[{{ instance }}]-[{{ pod }}]` |
|
||||
# Frontend Requests
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| HTTP QPS per Instance | `sum by(instance, pod, path, method, code) (rate(greptime_servers_http_requests_elapsed_count{path!~"/health\|/metrics"}[$__rate_interval]))` | `timeseries` | HTTP QPS per Instance. | `prometheus` | `reqps` | `[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]` |
|
||||
| HTTP P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, path, method, code) (rate(greptime_servers_http_requests_elapsed_bucket{path!~"/health\|/metrics"}[$__rate_interval])))` | `timeseries` | HTTP P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]-p99` |
|
||||
| gRPC QPS per Instance | `sum by(instance, pod, path, code) (rate(greptime_servers_grpc_requests_elapsed_count{}[$__rate_interval]))` | `timeseries` | gRPC QPS per Instance. | `prometheus` | `reqps` | `[{{instance}}]-[{{pod}}]-[{{path}}]-[{{code}}]` |
|
||||
| gRPC P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, path, code) (rate(greptime_servers_grpc_requests_elapsed_bucket{}[$__rate_interval])))` | `timeseries` | gRPC P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]-p99` |
|
||||
| MySQL QPS per Instance | `sum by(pod, instance)(rate(greptime_servers_mysql_query_elapsed_count{}[$__rate_interval]))` | `timeseries` | MySQL QPS per Instance. | `prometheus` | `reqps` | `[{{instance}}]-[{{pod}}]` |
|
||||
| MySQL P99 per Instance | `histogram_quantile(0.99, sum by(pod, instance, le) (rate(greptime_servers_mysql_query_elapsed_bucket{}[$__rate_interval])))` | `timeseries` | MySQL P99 per Instance. | `prometheus` | `s` | `[{{ instance }}]-[{{ pod }}]-p99` |
|
||||
| PostgreSQL QPS per Instance | `sum by(pod, instance)(rate(greptime_servers_postgres_query_elapsed_count{}[$__rate_interval]))` | `timeseries` | PostgreSQL QPS per Instance. | `prometheus` | `reqps` | `[{{instance}}]-[{{pod}}]` |
|
||||
| PostgreSQL P99 per Instance | `histogram_quantile(0.99, sum by(pod,instance,le) (rate(greptime_servers_postgres_query_elapsed_bucket{}[$__rate_interval])))` | `timeseries` | PostgreSQL P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-p99` |
|
||||
# Frontend to Datanode
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Ingest Rows per Instance | `sum by(instance, pod)(rate(greptime_table_operator_ingest_rows{}[$__rate_interval]))` | `timeseries` | Ingestion rate by row as in each frontend | `prometheus` | `rowsps` | `[{{instance}}]-[{{pod}}]` |
|
||||
| Region Call QPS per Instance | `sum by(instance, pod, request_type) (rate(greptime_grpc_region_request_count{}[$__rate_interval]))` | `timeseries` | Region Call QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{request_type}}]` |
|
||||
| Region Call P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, request_type) (rate(greptime_grpc_region_request_bucket{}[$__rate_interval])))` | `timeseries` | Region Call P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{request_type}}]` |
|
||||
# Mito Engine
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Request OPS per Instance | `sum by(instance, pod, type) (rate(greptime_mito_handle_request_elapsed_count{}[$__rate_interval]))` | `timeseries` | Request QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{type}}]` |
|
||||
| Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, type) (rate(greptime_mito_handle_request_elapsed_bucket{}[$__rate_interval])))` | `timeseries` | Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{type}}]` |
|
||||
| Write Buffer per Instance | `greptime_mito_write_buffer_bytes{}` | `timeseries` | Write Buffer per Instance. | `prometheus` | `decbytes` | `[{{instance}}]-[{{pod}}]` |
|
||||
| Write Rows per Instance | `sum by (instance, pod) (rate(greptime_mito_write_rows_total{}[$__rate_interval]))` | `timeseries` | Ingestion size by row counts. | `prometheus` | `rowsps` | `[{{instance}}]-[{{pod}}]` |
|
||||
| Flush OPS per Instance | `sum by(instance, pod, reason) (rate(greptime_mito_flush_requests_total{}[$__rate_interval]))` | `timeseries` | Flush QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{reason}}]` |
|
||||
| Write Stall per Instance | `sum by(instance, pod) (greptime_mito_write_stall_total{})` | `timeseries` | Write Stall per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]` |
|
||||
| Read Stage OPS per Instance | `sum by(instance, pod) (rate(greptime_mito_read_stage_elapsed_count{ stage="total"}[$__rate_interval]))` | `timeseries` | Read Stage OPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]` |
|
||||
| Read Stage P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_read_stage_elapsed_bucket{}[$__rate_interval])))` | `timeseries` | Read Stage P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]` |
|
||||
| Write Stage P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_write_stage_elapsed_bucket{}[$__rate_interval])))` | `timeseries` | Write Stage P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]` |
|
||||
| Compaction OPS per Instance | `sum by(instance, pod) (rate(greptime_mito_compaction_total_elapsed_count{}[$__rate_interval]))` | `timeseries` | Compaction OPS per Instance. | `prometheus` | `ops` | `[{{ instance }}]-[{{pod}}]` |
|
||||
| Compaction P99 per Instance by Stage | `histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_compaction_stage_elapsed_bucket{}[$__rate_interval])))` | `timeseries` | Compaction latency by stage | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]-p99` |
|
||||
| Compaction P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le,stage) (rate(greptime_mito_compaction_total_elapsed_bucket{}[$__rate_interval])))` | `timeseries` | Compaction P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{stage}}]-compaction` |
|
||||
| WAL write size | `histogram_quantile(0.95, sum by(le,instance, pod) (rate(raft_engine_write_size_bucket[$__rate_interval])))`<br/>`histogram_quantile(0.99, sum by(le,instance,pod) (rate(raft_engine_write_size_bucket[$__rate_interval])))`<br/>`sum by (instance, pod)(rate(raft_engine_write_size_sum[$__rate_interval]))` | `timeseries` | Write-ahead logs write size as bytes. This chart includes stats of p95 and p99 size by instance, total WAL write rate. | `prometheus` | `bytes` | `[{{instance}}]-[{{pod}}]-req-size-p95` |
|
||||
| Cached Bytes per Instance | `greptime_mito_cache_bytes{}` | `timeseries` | Cached Bytes per Instance. | `prometheus` | `decbytes` | `[{{instance}}]-[{{pod}}]-[{{type}}]` |
|
||||
| Inflight Compaction | `greptime_mito_inflight_compaction_count` | `timeseries` | Ongoing compaction task count | `prometheus` | `none` | `[{{instance}}]-[{{pod}}]` |
|
||||
| WAL sync duration seconds | `histogram_quantile(0.99, sum by(le, type, node, instance, pod) (rate(raft_engine_sync_log_duration_seconds_bucket[$__rate_interval])))` | `timeseries` | Raft engine (local disk) log store sync latency, p99 | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-p99` |
|
||||
| Log Store op duration seconds | `histogram_quantile(0.99, sum by(le,logstore,optype,instance, pod) (rate(greptime_logstore_op_elapsed_bucket[$__rate_interval])))` | `timeseries` | Write-ahead log operations latency at p99 | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{logstore}}]-[{{optype}}]-p99` |
|
||||
| Inflight Flush | `greptime_mito_inflight_flush_count` | `timeseries` | Ongoing flush task count | `prometheus` | `none` | `[{{instance}}]-[{{pod}}]` |
|
||||
# OpenDAL
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| QPS per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{}[$__rate_interval]))` | `timeseries` | QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
|
||||
| Read QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="read"}[$__rate_interval]))` | `timeseries` | Read QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
|
||||
| Read P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{operation="read"}[$__rate_interval])))` | `timeseries` | Read P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
|
||||
| Write QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="write"}[$__rate_interval]))` | `timeseries` | Write QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-{{scheme}}` |
|
||||
| Write P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation="write"}[$__rate_interval])))` | `timeseries` | Write P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
|
||||
| List QPS per Instance | `sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="list"}[$__rate_interval]))` | `timeseries` | List QPS per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
|
||||
| List P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation="list"}[$__rate_interval])))` | `timeseries` | List P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]` |
|
||||
| Other Requests per Instance | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{operation!~"read\|write\|list\|stat"}[$__rate_interval]))` | `timeseries` | Other Requests per Instance. | `prometheus` | `ops` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
|
||||
| Other Request P99 per Instance | `histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~"read\|write\|list"}[$__rate_interval])))` | `timeseries` | Other Request P99 per Instance. | `prometheus` | `s` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
|
||||
| Opendal traffic | `sum by(instance, pod, scheme, operation) (rate(opendal_operation_bytes_sum{}[$__rate_interval]))` | `timeseries` | Total traffic as in bytes by instance and operation | `prometheus` | `decbytes` | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]` |
|
||||
| OpenDAL errors per Instance | `sum by(instance, pod, scheme, operation, error) (rate(opendal_operation_errors_total{ error!="NotFound"}[$__rate_interval]))` | `timeseries` | OpenDAL error counts per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]-[{{error}}]` |
|
||||
# Metasrv
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Region migration datanode | `greptime_meta_region_migration_stat{datanode_type="src"}`<br/>`greptime_meta_region_migration_stat{datanode_type="desc"}` | `state-timeline` | Counter of region migration by source and destination | `prometheus` | `none` | `from-datanode-{{datanode_id}}` |
|
||||
| Region migration error | `greptime_meta_region_migration_error` | `timeseries` | Counter of region migration error | `prometheus` | `none` | `__auto` |
|
||||
| Datanode load | `greptime_datanode_load` | `timeseries` | Gauge of load information of each datanode, collected via heartbeat between datanode and metasrv. This information is for metasrv to schedule workloads. | `prometheus` | `none` | `__auto` |
|
||||
# Flownode
|
||||
| Title | Query | Type | Description | Datasource | Unit | Legend Format |
|
||||
| --- | --- | --- | --- | --- | --- | --- |
|
||||
| Flow Ingest / Output Rate | `sum by(instance, pod, direction) (rate(greptime_flow_processed_rows[$__rate_interval]))` | `timeseries` | Flow Ingest / Output Rate. | `prometheus` | -- | `[{{pod}}]-[{{instance}}]-[{{direction}}]` |
|
||||
| Flow Ingest Latency | `histogram_quantile(0.95, sum(rate(greptime_flow_insert_elapsed_bucket[$__rate_interval])) by (le, instance, pod))`<br/>`histogram_quantile(0.99, sum(rate(greptime_flow_insert_elapsed_bucket[$__rate_interval])) by (le, instance, pod))` | `timeseries` | Flow Ingest Latency. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-p95` |
|
||||
| Flow Operation Latency | `histogram_quantile(0.95, sum(rate(greptime_flow_processing_time_bucket[$__rate_interval])) by (le,instance,pod,type))`<br/>`histogram_quantile(0.99, sum(rate(greptime_flow_processing_time_bucket[$__rate_interval])) by (le,instance,pod,type))` | `timeseries` | Flow Operation Latency. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{type}}]-p95` |
|
||||
| Flow Buffer Size per Instance | `greptime_flow_input_buf_size` | `timeseries` | Flow Buffer Size per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}]` |
|
||||
| Flow Processing Error per Instance | `sum by(instance,pod,code) (rate(greptime_flow_errors[$__rate_interval]))` | `timeseries` | Flow Processing Error per Instance. | `prometheus` | -- | `[{{instance}}]-[{{pod}}]-[{{code}}]` |
|
||||
769
grafana/dashboards/standalone/dashboard.yaml
Normal file
769
grafana/dashboards/standalone/dashboard.yaml
Normal file
@@ -0,0 +1,769 @@
|
||||
groups:
|
||||
- title: Overview
|
||||
panels:
|
||||
- title: Uptime
|
||||
type: stat
|
||||
description: The start time of GreptimeDB.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: time() - process_start_time_seconds
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: __auto
|
||||
- title: Version
|
||||
type: stat
|
||||
description: GreptimeDB version.
|
||||
queries:
|
||||
- expr: SELECT pkg_version FROM information_schema.build_info
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Total Ingestion Rate
|
||||
type: stat
|
||||
description: Total ingestion rate.
|
||||
unit: rowsps
|
||||
queries:
|
||||
- expr: sum(rate(greptime_table_operator_ingest_rows[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: __auto
|
||||
- title: Total Storage Size
|
||||
type: stat
|
||||
description: Total number of data file size.
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: select SUM(disk_size) from information_schema.region_statistics;
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Total Rows
|
||||
type: stat
|
||||
description: Total number of data rows in the cluster. Calculated by sum of rows from each region.
|
||||
unit: sishort
|
||||
queries:
|
||||
- expr: select SUM(region_rows) from information_schema.region_statistics;
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Deployment
|
||||
type: stat
|
||||
description: The deployment topology of GreptimeDB.
|
||||
queries:
|
||||
- expr: SELECT count(*) as datanode FROM information_schema.cluster_info WHERE peer_type = 'DATANODE';
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT count(*) as frontend FROM information_schema.cluster_info WHERE peer_type = 'FRONTEND';
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT count(*) as metasrv FROM information_schema.cluster_info WHERE peer_type = 'METASRV';
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT count(*) as flownode FROM information_schema.cluster_info WHERE peer_type = 'FLOWNODE';
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Database Resources
|
||||
type: stat
|
||||
description: The number of the key resources in GreptimeDB.
|
||||
queries:
|
||||
- expr: SELECT COUNT(*) as databases FROM information_schema.schemata WHERE schema_name NOT IN ('greptime_private', 'information_schema')
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT COUNT(*) as tables FROM information_schema.tables WHERE table_schema != 'information_schema'
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT COUNT(region_id) as regions FROM information_schema.region_peers
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT COUNT(*) as flows FROM information_schema.flows
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Data Size
|
||||
type: stat
|
||||
description: The data size of wal/index/manifest in the GreptimeDB.
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: SELECT SUM(memtable_size) * 0.42825 as WAL FROM information_schema.region_statistics;
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT SUM(index_size) as index FROM information_schema.region_statistics;
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- expr: SELECT SUM(manifest_size) as manifest FROM information_schema.region_statistics;
|
||||
datasource:
|
||||
type: mysql
|
||||
uid: ${information_schema}
|
||||
- title: Ingestion
|
||||
panels:
|
||||
- title: Total Ingestion Rate
|
||||
type: timeseries
|
||||
description: |
|
||||
Total ingestion rate.
|
||||
|
||||
Here we listed 3 primary protocols:
|
||||
|
||||
- Prometheus remote write
|
||||
- Greptime's gRPC API (when using our ingest SDK)
|
||||
- Log ingestion http API
|
||||
unit: rowsps
|
||||
queries:
|
||||
- expr: sum(rate(greptime_table_operator_ingest_rows{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: ingestion
|
||||
- title: Ingestion Rate by Type
|
||||
type: timeseries
|
||||
description: |
|
||||
Total ingestion rate.
|
||||
|
||||
Here we listed 3 primary protocols:
|
||||
|
||||
- Prometheus remote write
|
||||
- Greptime's gRPC API (when using our ingest SDK)
|
||||
- Log ingestion http API
|
||||
unit: rowsps
|
||||
queries:
|
||||
- expr: sum(rate(greptime_servers_http_logs_ingestion_counter[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: http-logs
|
||||
- expr: sum(rate(greptime_servers_prometheus_remote_write_samples[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: prometheus-remote-write
|
||||
- title: Queries
|
||||
panels:
|
||||
- title: Total Query Rate
|
||||
type: timeseries
|
||||
description: |-
|
||||
Total rate of query API calls by protocol. This metric is collected from frontends.
|
||||
|
||||
Here we listed 3 main protocols:
|
||||
- MySQL
|
||||
- Postgres
|
||||
- Prometheus API
|
||||
|
||||
Note that there are some other minor query APIs like /sql are not included
|
||||
unit: reqps
|
||||
queries:
|
||||
- expr: sum (rate(greptime_servers_mysql_query_elapsed_count{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: mysql
|
||||
- expr: sum (rate(greptime_servers_postgres_query_elapsed_count{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: pg
|
||||
- expr: sum (rate(greptime_servers_http_promql_elapsed_counte{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: promql
|
||||
- title: Resources
|
||||
panels:
|
||||
- title: Datanode Memory per Instance
|
||||
type: timeseries
|
||||
description: Current memory usage by instance
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: sum(process_resident_memory_bytes{}) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{ pod }}]'
|
||||
- title: Datanode CPU Usage per Instance
|
||||
type: timeseries
|
||||
description: Current cpu usage by instance
|
||||
unit: none
|
||||
queries:
|
||||
- expr: sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]'
|
||||
- title: Frontend Memory per Instance
|
||||
type: timeseries
|
||||
description: Current memory usage by instance
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: sum(process_resident_memory_bytes{}) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]'
|
||||
- title: Frontend CPU Usage per Instance
|
||||
type: timeseries
|
||||
description: Current cpu usage by instance
|
||||
unit: none
|
||||
queries:
|
||||
- expr: sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]-cpu'
|
||||
- title: Metasrv Memory per Instance
|
||||
type: timeseries
|
||||
description: Current memory usage by instance
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: sum(process_resident_memory_bytes{}) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]-resident'
|
||||
- title: Metasrv CPU Usage per Instance
|
||||
type: timeseries
|
||||
description: Current cpu usage by instance
|
||||
unit: none
|
||||
queries:
|
||||
- expr: sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]'
|
||||
- title: Flownode Memory per Instance
|
||||
type: timeseries
|
||||
description: Current memory usage by instance
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: sum(process_resident_memory_bytes{}) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]'
|
||||
- title: Flownode CPU Usage per Instance
|
||||
type: timeseries
|
||||
description: Current cpu usage by instance
|
||||
unit: none
|
||||
queries:
|
||||
- expr: sum(rate(process_cpu_seconds_total{}[$__rate_interval]) * 1000) by (instance, pod)
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]'
|
||||
- title: Frontend Requests
|
||||
panels:
|
||||
- title: HTTP QPS per Instance
|
||||
type: timeseries
|
||||
description: HTTP QPS per Instance.
|
||||
unit: reqps
|
||||
queries:
|
||||
- expr: sum by(instance, pod, path, method, code) (rate(greptime_servers_http_requests_elapsed_count{path!~"/health|/metrics"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]'
|
||||
- title: HTTP P99 per Instance
|
||||
type: timeseries
|
||||
description: HTTP P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, path, method, code) (rate(greptime_servers_http_requests_elapsed_bucket{path!~"/health|/metrics"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]-p99'
|
||||
- title: gRPC QPS per Instance
|
||||
type: timeseries
|
||||
description: gRPC QPS per Instance.
|
||||
unit: reqps
|
||||
queries:
|
||||
- expr: sum by(instance, pod, path, code) (rate(greptime_servers_grpc_requests_elapsed_count{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{path}}]-[{{code}}]'
|
||||
- title: gRPC P99 per Instance
|
||||
type: timeseries
|
||||
description: gRPC P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, path, code) (rate(greptime_servers_grpc_requests_elapsed_bucket{}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{path}}]-[{{method}}]-[{{code}}]-p99'
|
||||
- title: MySQL QPS per Instance
|
||||
type: timeseries
|
||||
description: MySQL QPS per Instance.
|
||||
unit: reqps
|
||||
queries:
|
||||
- expr: sum by(pod, instance)(rate(greptime_servers_mysql_query_elapsed_count{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: MySQL P99 per Instance
|
||||
type: timeseries
|
||||
description: MySQL P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(pod, instance, le) (rate(greptime_servers_mysql_query_elapsed_bucket{}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{ pod }}]-p99'
|
||||
- title: PostgreSQL QPS per Instance
|
||||
type: timeseries
|
||||
description: PostgreSQL QPS per Instance.
|
||||
unit: reqps
|
||||
queries:
|
||||
- expr: sum by(pod, instance)(rate(greptime_servers_postgres_query_elapsed_count{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: PostgreSQL P99 per Instance
|
||||
type: timeseries
|
||||
description: PostgreSQL P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(pod,instance,le) (rate(greptime_servers_postgres_query_elapsed_bucket{}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-p99'
|
||||
- title: Frontend to Datanode
|
||||
panels:
|
||||
- title: Ingest Rows per Instance
|
||||
type: timeseries
|
||||
description: Ingestion rate by row as in each frontend
|
||||
unit: rowsps
|
||||
queries:
|
||||
- expr: sum by(instance, pod)(rate(greptime_table_operator_ingest_rows{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: Region Call QPS per Instance
|
||||
type: timeseries
|
||||
description: Region Call QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, request_type) (rate(greptime_grpc_region_request_count{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{request_type}}]'
|
||||
- title: Region Call P99 per Instance
|
||||
type: timeseries
|
||||
description: Region Call P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, request_type) (rate(greptime_grpc_region_request_bucket{}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{request_type}}]'
|
||||
- title: Mito Engine
|
||||
panels:
|
||||
- title: Request OPS per Instance
|
||||
type: timeseries
|
||||
description: Request QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, type) (rate(greptime_mito_handle_request_elapsed_count{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]'
|
||||
- title: Request P99 per Instance
|
||||
type: timeseries
|
||||
description: Request P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, type) (rate(greptime_mito_handle_request_elapsed_bucket{}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]'
|
||||
- title: Write Buffer per Instance
|
||||
type: timeseries
|
||||
description: Write Buffer per Instance.
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: greptime_mito_write_buffer_bytes{}
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: Write Rows per Instance
|
||||
type: timeseries
|
||||
description: Ingestion size by row counts.
|
||||
unit: rowsps
|
||||
queries:
|
||||
- expr: sum by (instance, pod) (rate(greptime_mito_write_rows_total{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: Flush OPS per Instance
|
||||
type: timeseries
|
||||
description: Flush QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, reason) (rate(greptime_mito_flush_requests_total{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{reason}}]'
|
||||
- title: Write Stall per Instance
|
||||
type: timeseries
|
||||
description: Write Stall per Instance.
|
||||
queries:
|
||||
- expr: sum by(instance, pod) (greptime_mito_write_stall_total{})
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: Read Stage OPS per Instance
|
||||
type: timeseries
|
||||
description: Read Stage OPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod) (rate(greptime_mito_read_stage_elapsed_count{ stage="total"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: Read Stage P99 per Instance
|
||||
type: timeseries
|
||||
description: Read Stage P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_read_stage_elapsed_bucket{}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{stage}}]'
|
||||
- title: Write Stage P99 per Instance
|
||||
type: timeseries
|
||||
description: Write Stage P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_write_stage_elapsed_bucket{}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{stage}}]'
|
||||
- title: Compaction OPS per Instance
|
||||
type: timeseries
|
||||
description: Compaction OPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod) (rate(greptime_mito_compaction_total_elapsed_count{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{ instance }}]-[{{pod}}]'
|
||||
- title: Compaction P99 per Instance by Stage
|
||||
type: timeseries
|
||||
description: Compaction latency by stage
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, stage) (rate(greptime_mito_compaction_stage_elapsed_bucket{}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{stage}}]-p99'
|
||||
- title: Compaction P99 per Instance
|
||||
type: timeseries
|
||||
description: Compaction P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le,stage) (rate(greptime_mito_compaction_total_elapsed_bucket{}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{stage}}]-compaction'
|
||||
- title: WAL write size
|
||||
type: timeseries
|
||||
description: Write-ahead logs write size as bytes. This chart includes stats of p95 and p99 size by instance, total WAL write rate.
|
||||
unit: bytes
|
||||
queries:
|
||||
- expr: histogram_quantile(0.95, sum by(le,instance, pod) (rate(raft_engine_write_size_bucket[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-req-size-p95'
|
||||
- expr: histogram_quantile(0.99, sum by(le,instance,pod) (rate(raft_engine_write_size_bucket[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-req-size-p99'
|
||||
- expr: sum by (instance, pod)(rate(raft_engine_write_size_sum[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-throughput'
|
||||
- title: Cached Bytes per Instance
|
||||
type: timeseries
|
||||
description: Cached Bytes per Instance.
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: greptime_mito_cache_bytes{}
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]'
|
||||
- title: Inflight Compaction
|
||||
type: timeseries
|
||||
description: Ongoing compaction task count
|
||||
unit: none
|
||||
queries:
|
||||
- expr: greptime_mito_inflight_compaction_count
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: WAL sync duration seconds
|
||||
type: timeseries
|
||||
description: Raft engine (local disk) log store sync latency, p99
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(le, type, node, instance, pod) (rate(raft_engine_sync_log_duration_seconds_bucket[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-p99'
|
||||
- title: Log Store op duration seconds
|
||||
type: timeseries
|
||||
description: Write-ahead log operations latency at p99
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(le,logstore,optype,instance, pod) (rate(greptime_logstore_op_elapsed_bucket[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{logstore}}]-[{{optype}}]-p99'
|
||||
- title: Inflight Flush
|
||||
type: timeseries
|
||||
description: Ongoing flush task count
|
||||
unit: none
|
||||
queries:
|
||||
- expr: greptime_mito_inflight_flush_count
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]'
|
||||
- title: OpenDAL
|
||||
panels:
|
||||
- title: QPS per Instance
|
||||
type: timeseries
|
||||
description: QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
|
||||
- title: Read QPS per Instance
|
||||
type: timeseries
|
||||
description: Read QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="read"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
|
||||
- title: Read P99 per Instance
|
||||
type: timeseries
|
||||
description: Read P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{operation="read"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
|
||||
- title: Write QPS per Instance
|
||||
type: timeseries
|
||||
description: Write QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="write"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-{{scheme}}'
|
||||
- title: Write P99 per Instance
|
||||
type: timeseries
|
||||
description: Write P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation="write"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
|
||||
- title: List QPS per Instance
|
||||
type: timeseries
|
||||
description: List QPS per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme) (rate(opendal_operation_duration_seconds_count{ operation="list"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
|
||||
- title: List P99 per Instance
|
||||
type: timeseries
|
||||
description: List P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme) (rate(opendal_operation_duration_seconds_bucket{ operation="list"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]'
|
||||
- title: Other Requests per Instance
|
||||
type: timeseries
|
||||
description: Other Requests per Instance.
|
||||
unit: ops
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_duration_seconds_count{operation!~"read|write|list|stat"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
|
||||
- title: Other Request P99 per Instance
|
||||
type: timeseries
|
||||
description: Other Request P99 per Instance.
|
||||
unit: s
|
||||
queries:
|
||||
- expr: histogram_quantile(0.99, sum by(instance, pod, le, scheme, operation) (rate(opendal_operation_duration_seconds_bucket{ operation!~"read|write|list"}[$__rate_interval])))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
|
||||
- title: Opendal traffic
|
||||
type: timeseries
|
||||
description: Total traffic as in bytes by instance and operation
|
||||
unit: decbytes
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme, operation) (rate(opendal_operation_bytes_sum{}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]'
|
||||
- title: OpenDAL errors per Instance
|
||||
type: timeseries
|
||||
description: OpenDAL error counts per Instance.
|
||||
queries:
|
||||
- expr: sum by(instance, pod, scheme, operation, error) (rate(opendal_operation_errors_total{ error!="NotFound"}[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{scheme}}]-[{{operation}}]-[{{error}}]'
|
||||
- title: Metasrv
|
||||
panels:
|
||||
- title: Region migration datanode
|
||||
type: state-timeline
|
||||
description: Counter of region migration by source and destination
|
||||
unit: none
|
||||
queries:
|
||||
- expr: greptime_meta_region_migration_stat{datanode_type="src"}
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: from-datanode-{{datanode_id}}
|
||||
- expr: greptime_meta_region_migration_stat{datanode_type="desc"}
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: to-datanode-{{datanode_id}}
|
||||
- title: Region migration error
|
||||
type: timeseries
|
||||
description: Counter of region migration error
|
||||
unit: none
|
||||
queries:
|
||||
- expr: greptime_meta_region_migration_error
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: __auto
|
||||
- title: Datanode load
|
||||
type: timeseries
|
||||
description: Gauge of load information of each datanode, collected via heartbeat between datanode and metasrv. This information is for metasrv to schedule workloads.
|
||||
unit: none
|
||||
queries:
|
||||
- expr: greptime_datanode_load
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: __auto
|
||||
- title: Flownode
|
||||
panels:
|
||||
- title: Flow Ingest / Output Rate
|
||||
type: timeseries
|
||||
description: Flow Ingest / Output Rate.
|
||||
queries:
|
||||
- expr: sum by(instance, pod, direction) (rate(greptime_flow_processed_rows[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{pod}}]-[{{instance}}]-[{{direction}}]'
|
||||
- title: Flow Ingest Latency
|
||||
type: timeseries
|
||||
description: Flow Ingest Latency.
|
||||
queries:
|
||||
- expr: histogram_quantile(0.95, sum(rate(greptime_flow_insert_elapsed_bucket[$__rate_interval])) by (le, instance, pod))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-p95'
|
||||
- expr: histogram_quantile(0.99, sum(rate(greptime_flow_insert_elapsed_bucket[$__rate_interval])) by (le, instance, pod))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-p99'
|
||||
- title: Flow Operation Latency
|
||||
type: timeseries
|
||||
description: Flow Operation Latency.
|
||||
queries:
|
||||
- expr: histogram_quantile(0.95, sum(rate(greptime_flow_processing_time_bucket[$__rate_interval])) by (le,instance,pod,type))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]-p95'
|
||||
- expr: histogram_quantile(0.99, sum(rate(greptime_flow_processing_time_bucket[$__rate_interval])) by (le,instance,pod,type))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{type}}]-p99'
|
||||
- title: Flow Buffer Size per Instance
|
||||
type: timeseries
|
||||
description: Flow Buffer Size per Instance.
|
||||
queries:
|
||||
- expr: greptime_flow_input_buf_size
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}]'
|
||||
- title: Flow Processing Error per Instance
|
||||
type: timeseries
|
||||
description: Flow Processing Error per Instance.
|
||||
queries:
|
||||
- expr: sum by(instance,pod,code) (rate(greptime_flow_errors[$__rate_interval]))
|
||||
datasource:
|
||||
type: prometheus
|
||||
uid: ${metrics}
|
||||
legendFormat: '[{{instance}}]-[{{pod}}]-[{{code}}]'
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
54
grafana/scripts/check.sh
Executable file
54
grafana/scripts/check.sh
Executable file
@@ -0,0 +1,54 @@
|
||||
#!/usr/bin/env bash
|
||||
|
||||
DASHBOARD_DIR=${1:-grafana/dashboards}
|
||||
|
||||
check_dashboard_description() {
|
||||
for dashboard in $(find $DASHBOARD_DIR -name "*.json"); do
|
||||
echo "Checking $dashboard description"
|
||||
|
||||
# Use jq to check for panels with empty or missing descriptions
|
||||
invalid_panels=$(cat $dashboard | jq -r '
|
||||
.panels[]
|
||||
| select((.type == "stats" or .type == "timeseries") and (.description == "" or .description == null))')
|
||||
|
||||
# Check if any invalid panels were found
|
||||
if [[ -n "$invalid_panels" ]]; then
|
||||
echo "Error: The following panels have empty or missing descriptions:"
|
||||
echo "$invalid_panels"
|
||||
exit 1
|
||||
else
|
||||
echo "All panels with type 'stats' or 'timeseries' have valid descriptions."
|
||||
fi
|
||||
done
|
||||
}
|
||||
|
||||
check_dashboards_generation() {
|
||||
./grafana/scripts/gen-dashboards.sh
|
||||
|
||||
if [[ -n "$(git diff --name-only grafana/dashboards)" ]]; then
|
||||
echo "Error: The dashboards are not generated correctly. You should execute the `make dashboards` command."
|
||||
exit 1
|
||||
fi
|
||||
}
|
||||
|
||||
check_datasource() {
|
||||
for dashboard in $(find $DASHBOARD_DIR -name "*.json"); do
|
||||
echo "Checking $dashboard datasource"
|
||||
jq -r '.panels[] | select(.type != "row") | .targets[] | [.datasource.type, .datasource.uid] | @tsv' $dashboard | while read -r type uid; do
|
||||
# if the datasource is prometheus, check if the uid is ${metrics}
|
||||
if [[ "$type" == "prometheus" && "$uid" != "\${metrics}" ]]; then
|
||||
echo "Error: The datasource uid of $dashboard is not valid. It should be \${metrics}, got $uid"
|
||||
exit 1
|
||||
fi
|
||||
# if the datasource is mysql, check if the uid is ${information_schema}
|
||||
if [[ "$type" == "mysql" && "$uid" != "\${information_schema}" ]]; then
|
||||
echo "Error: The datasource uid of $dashboard is not valid. It should be \${information_schema}, got $uid"
|
||||
exit 1
|
||||
fi
|
||||
done
|
||||
done
|
||||
}
|
||||
|
||||
check_dashboards_generation
|
||||
check_dashboard_description
|
||||
check_datasource
|
||||
25
grafana/scripts/gen-dashboards.sh
Executable file
25
grafana/scripts/gen-dashboards.sh
Executable file
@@ -0,0 +1,25 @@
|
||||
#! /usr/bin/env bash
|
||||
|
||||
CLUSTER_DASHBOARD_DIR=${1:-grafana/dashboards/cluster}
|
||||
STANDALONE_DASHBOARD_DIR=${2:-grafana/dashboards/standalone}
|
||||
DAC_IMAGE=ghcr.io/zyy17/dac:20250423-522bd35
|
||||
|
||||
remove_instance_filters() {
|
||||
# Remove the instance filters for the standalone dashboards.
|
||||
sed 's/instance=~\\"$datanode\\",//; s/instance=~\\"$datanode\\"//; s/instance=~\\"$frontend\\",//; s/instance=~\\"$frontend\\"//; s/instance=~\\"$metasrv\\",//; s/instance=~\\"$metasrv\\"//; s/instance=~\\"$flownode\\",//; s/instance=~\\"$flownode\\"//;' $CLUSTER_DASHBOARD_DIR/dashboard.json > $STANDALONE_DASHBOARD_DIR/dashboard.json
|
||||
}
|
||||
|
||||
generate_intermediate_dashboards_and_docs() {
|
||||
docker run -v ${PWD}:/greptimedb --rm ${DAC_IMAGE} \
|
||||
-i /greptimedb/$CLUSTER_DASHBOARD_DIR/dashboard.json \
|
||||
-o /greptimedb/$CLUSTER_DASHBOARD_DIR/dashboard.yaml \
|
||||
-m /greptimedb/$CLUSTER_DASHBOARD_DIR/dashboard.md
|
||||
|
||||
docker run -v ${PWD}:/greptimedb --rm ${DAC_IMAGE} \
|
||||
-i /greptimedb/$STANDALONE_DASHBOARD_DIR/dashboard.json \
|
||||
-o /greptimedb/$STANDALONE_DASHBOARD_DIR/dashboard.yaml \
|
||||
-m /greptimedb/$STANDALONE_DASHBOARD_DIR/dashboard.md
|
||||
}
|
||||
|
||||
remove_instance_filters
|
||||
generate_intermediate_dashboards_and_docs
|
||||
@@ -1,11 +0,0 @@
|
||||
#!/usr/bin/env bash
|
||||
|
||||
BASEDIR=$(dirname "$0")
|
||||
echo '| Title | Description | Expressions |
|
||||
|---|---|---|'
|
||||
|
||||
cat $BASEDIR/greptimedb-cluster.json | jq -r '
|
||||
.panels |
|
||||
map(select(.type == "stat" or .type == "timeseries")) |
|
||||
.[] | "| \(.title) | \(.description | gsub("\n"; "<br>")) | \(.targets | map(.expr // .rawSql | "`\(.|gsub("\n"; "<br>"))`") | join("<br>")) |"
|
||||
'
|
||||
@@ -514,6 +514,7 @@ fn query_request_type(request: &QueryRequest) -> &'static str {
|
||||
Some(Query::Sql(_)) => "query.sql",
|
||||
Some(Query::LogicalPlan(_)) => "query.logical_plan",
|
||||
Some(Query::PromRangeQuery(_)) => "query.prom_range",
|
||||
Some(Query::InsertIntoPlan(_)) => "query.insert_into_plan",
|
||||
None => "query.empty",
|
||||
}
|
||||
}
|
||||
|
||||
@@ -27,7 +27,7 @@ use session::context::QueryContextRef;
|
||||
use snafu::{ensure, OptionExt, ResultExt};
|
||||
use table::metadata::TableType;
|
||||
use table::table::adapter::DfTableProviderAdapter;
|
||||
mod dummy_catalog;
|
||||
pub mod dummy_catalog;
|
||||
use dummy_catalog::DummyCatalogList;
|
||||
use table::TableRef;
|
||||
|
||||
|
||||
@@ -345,7 +345,7 @@ impl StartCommand {
|
||||
let client = Arc::new(NodeClients::new(channel_config));
|
||||
|
||||
let invoker = FrontendInvoker::build_from(
|
||||
flownode.flow_worker_manager().clone(),
|
||||
flownode.flow_engine().streaming_engine(),
|
||||
catalog_manager.clone(),
|
||||
cached_meta_backend.clone(),
|
||||
layered_cache_registry.clone(),
|
||||
@@ -355,7 +355,9 @@ impl StartCommand {
|
||||
.await
|
||||
.context(StartFlownodeSnafu)?;
|
||||
flownode
|
||||
.flow_worker_manager()
|
||||
.flow_engine()
|
||||
.streaming_engine()
|
||||
// TODO(discord9): refactor and avoid circular reference
|
||||
.set_frontend_invoker(invoker)
|
||||
.await;
|
||||
|
||||
|
||||
@@ -132,7 +132,7 @@ impl SubCommand {
|
||||
}
|
||||
|
||||
#[derive(Debug, Default, Parser)]
|
||||
struct StartCommand {
|
||||
pub struct StartCommand {
|
||||
/// The address to bind the gRPC server.
|
||||
#[clap(long, alias = "bind-addr")]
|
||||
rpc_bind_addr: Option<String>,
|
||||
@@ -172,7 +172,7 @@ struct StartCommand {
|
||||
}
|
||||
|
||||
impl StartCommand {
|
||||
fn load_options(&self, global_options: &GlobalOptions) -> Result<MetasrvOptions> {
|
||||
pub fn load_options(&self, global_options: &GlobalOptions) -> Result<MetasrvOptions> {
|
||||
let mut opts = MetasrvOptions::load_layered_options(
|
||||
self.config_file.as_deref(),
|
||||
self.env_prefix.as_ref(),
|
||||
@@ -261,7 +261,7 @@ impl StartCommand {
|
||||
Ok(())
|
||||
}
|
||||
|
||||
async fn build(&self, opts: MetasrvOptions) -> Result<Instance> {
|
||||
pub async fn build(&self, opts: MetasrvOptions) -> Result<Instance> {
|
||||
common_runtime::init_global_runtimes(&opts.runtime);
|
||||
|
||||
let guard = common_telemetry::init_global_logging(
|
||||
|
||||
@@ -56,8 +56,8 @@ use datanode::datanode::{Datanode, DatanodeBuilder};
|
||||
use datanode::region_server::RegionServer;
|
||||
use file_engine::config::EngineConfig as FileEngineConfig;
|
||||
use flow::{
|
||||
FlowConfig, FlowWorkerManager, FlownodeBuilder, FlownodeInstance, FlownodeOptions,
|
||||
FrontendClient, FrontendInvoker,
|
||||
FlowConfig, FlownodeBuilder, FlownodeInstance, FlownodeOptions, FrontendClient,
|
||||
FrontendInvoker, GrpcQueryHandlerWithBoxedError, StreamingEngine,
|
||||
};
|
||||
use frontend::frontend::{Frontend, FrontendOptions};
|
||||
use frontend::instance::builder::FrontendBuilder;
|
||||
@@ -524,17 +524,17 @@ impl StartCommand {
|
||||
..Default::default()
|
||||
};
|
||||
|
||||
// TODO(discord9): for standalone not use grpc, but just somehow get a handler to frontend grpc client without
|
||||
// for standalone not use grpc, but get a handler to frontend grpc client without
|
||||
// actually make a connection
|
||||
let fe_server_addr = fe_opts.grpc.bind_addr.clone();
|
||||
let frontend_client = FrontendClient::from_static_grpc_addr(fe_server_addr);
|
||||
let (frontend_client, frontend_instance_handler) =
|
||||
FrontendClient::from_empty_grpc_handler();
|
||||
let flow_builder = FlownodeBuilder::new(
|
||||
flownode_options,
|
||||
plugins.clone(),
|
||||
table_metadata_manager.clone(),
|
||||
catalog_manager.clone(),
|
||||
flow_metadata_manager.clone(),
|
||||
Arc::new(frontend_client),
|
||||
Arc::new(frontend_client.clone()),
|
||||
);
|
||||
let flownode = flow_builder
|
||||
.build()
|
||||
@@ -544,15 +544,15 @@ impl StartCommand {
|
||||
|
||||
// set the ref to query for the local flow state
|
||||
{
|
||||
let flow_worker_manager = flownode.flow_worker_manager();
|
||||
let flow_streaming_engine = flownode.flow_engine().streaming_engine();
|
||||
information_extension
|
||||
.set_flow_worker_manager(flow_worker_manager.clone())
|
||||
.set_flow_streaming_engine(flow_streaming_engine)
|
||||
.await;
|
||||
}
|
||||
|
||||
let node_manager = Arc::new(StandaloneDatanodeManager {
|
||||
region_server: datanode.region_server(),
|
||||
flow_server: flownode.flow_worker_manager(),
|
||||
flow_server: flownode.flow_engine(),
|
||||
});
|
||||
|
||||
let table_id_sequence = Arc::new(
|
||||
@@ -606,10 +606,19 @@ impl StartCommand {
|
||||
.context(error::StartFrontendSnafu)?;
|
||||
let fe_instance = Arc::new(fe_instance);
|
||||
|
||||
let flow_worker_manager = flownode.flow_worker_manager();
|
||||
// set the frontend client for flownode
|
||||
let grpc_handler = fe_instance.clone() as Arc<dyn GrpcQueryHandlerWithBoxedError>;
|
||||
let weak_grpc_handler = Arc::downgrade(&grpc_handler);
|
||||
frontend_instance_handler
|
||||
.lock()
|
||||
.unwrap()
|
||||
.replace(weak_grpc_handler);
|
||||
|
||||
// set the frontend invoker for flownode
|
||||
let flow_streaming_engine = flownode.flow_engine().streaming_engine();
|
||||
// flow server need to be able to use frontend to write insert requests back
|
||||
let invoker = FrontendInvoker::build_from(
|
||||
flow_worker_manager.clone(),
|
||||
flow_streaming_engine.clone(),
|
||||
catalog_manager.clone(),
|
||||
kv_backend.clone(),
|
||||
layered_cache_registry.clone(),
|
||||
@@ -618,7 +627,7 @@ impl StartCommand {
|
||||
)
|
||||
.await
|
||||
.context(error::StartFlownodeSnafu)?;
|
||||
flow_worker_manager.set_frontend_invoker(invoker).await;
|
||||
flow_streaming_engine.set_frontend_invoker(invoker).await;
|
||||
|
||||
let export_metrics_task = ExportMetricsTask::try_new(&opts.export_metrics, Some(&plugins))
|
||||
.context(error::ServersSnafu)?;
|
||||
@@ -694,7 +703,7 @@ pub struct StandaloneInformationExtension {
|
||||
region_server: RegionServer,
|
||||
procedure_manager: ProcedureManagerRef,
|
||||
start_time_ms: u64,
|
||||
flow_worker_manager: RwLock<Option<Arc<FlowWorkerManager>>>,
|
||||
flow_streaming_engine: RwLock<Option<Arc<StreamingEngine>>>,
|
||||
}
|
||||
|
||||
impl StandaloneInformationExtension {
|
||||
@@ -703,14 +712,14 @@ impl StandaloneInformationExtension {
|
||||
region_server,
|
||||
procedure_manager,
|
||||
start_time_ms: common_time::util::current_time_millis() as u64,
|
||||
flow_worker_manager: RwLock::new(None),
|
||||
flow_streaming_engine: RwLock::new(None),
|
||||
}
|
||||
}
|
||||
|
||||
/// Set the flow worker manager for the standalone instance.
|
||||
pub async fn set_flow_worker_manager(&self, flow_worker_manager: Arc<FlowWorkerManager>) {
|
||||
let mut guard = self.flow_worker_manager.write().await;
|
||||
*guard = Some(flow_worker_manager);
|
||||
/// Set the flow streaming engine for the standalone instance.
|
||||
pub async fn set_flow_streaming_engine(&self, flow_streaming_engine: Arc<StreamingEngine>) {
|
||||
let mut guard = self.flow_streaming_engine.write().await;
|
||||
*guard = Some(flow_streaming_engine);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -789,7 +798,7 @@ impl InformationExtension for StandaloneInformationExtension {
|
||||
|
||||
async fn flow_stats(&self) -> std::result::Result<Option<FlowStat>, Self::Error> {
|
||||
Ok(Some(
|
||||
self.flow_worker_manager
|
||||
self.flow_streaming_engine
|
||||
.read()
|
||||
.await
|
||||
.as_ref()
|
||||
|
||||
@@ -74,6 +74,7 @@ fn test_load_datanode_example_config() {
|
||||
RegionEngineConfig::File(FileEngineConfig {}),
|
||||
RegionEngineConfig::Metric(MetricEngineConfig {
|
||||
experimental_sparse_primary_key_encoding: false,
|
||||
flush_metadata_region_interval: Duration::from_secs(30),
|
||||
}),
|
||||
],
|
||||
logging: LoggingOptions {
|
||||
@@ -216,6 +217,7 @@ fn test_load_standalone_example_config() {
|
||||
RegionEngineConfig::File(FileEngineConfig {}),
|
||||
RegionEngineConfig::Metric(MetricEngineConfig {
|
||||
experimental_sparse_primary_key_encoding: false,
|
||||
flush_metadata_region_interval: Duration::from_secs(30),
|
||||
}),
|
||||
],
|
||||
storage: StorageConfig {
|
||||
|
||||
@@ -31,7 +31,8 @@ impl Plugins {
|
||||
}
|
||||
|
||||
pub fn insert<T: 'static + Send + Sync>(&self, value: T) {
|
||||
let _ = self.write().insert(value);
|
||||
let last = self.write().insert(value);
|
||||
assert!(last.is_none(), "each type of plugins must be one and only");
|
||||
}
|
||||
|
||||
pub fn get<T: 'static + Send + Sync + Clone>(&self) -> Option<T> {
|
||||
@@ -137,4 +138,12 @@ mod tests {
|
||||
assert_eq!(plugins.len(), 2);
|
||||
assert!(!plugins.is_empty());
|
||||
}
|
||||
|
||||
#[test]
|
||||
#[should_panic(expected = "each type of plugins must be one and only")]
|
||||
fn test_plugin_uniqueness() {
|
||||
let plugins = Plugins::new();
|
||||
plugins.insert(1i32);
|
||||
plugins.insert(2i32);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -115,6 +115,13 @@ impl Function for UddSketchCalcFunction {
|
||||
}
|
||||
};
|
||||
|
||||
// Check if the sketch is empty, if so, return null
|
||||
// This is important to avoid panics when calling estimate_quantile on an empty sketch
|
||||
// In practice, this will happen if input is all null
|
||||
if sketch.bucket_iter().count() == 0 {
|
||||
builder.push_null();
|
||||
continue;
|
||||
}
|
||||
// Compute the estimated quantile from the sketch
|
||||
let result = sketch.estimate_quantile(perc);
|
||||
builder.push(Some(result));
|
||||
|
||||
@@ -187,6 +187,7 @@ mod tests {
|
||||
},
|
||||
flownode_ids: BTreeMap::from([(0, 1), (1, 2), (2, 3)]),
|
||||
catalog_name: DEFAULT_CATALOG_NAME.to_string(),
|
||||
query_context: None,
|
||||
flow_name: "my_flow".to_string(),
|
||||
raw_sql: "sql".to_string(),
|
||||
expire_after: Some(300),
|
||||
|
||||
@@ -38,7 +38,7 @@ use table::metadata::TableId;
|
||||
use crate::cache_invalidator::Context;
|
||||
use crate::ddl::utils::{add_peer_context_if_needed, handle_retry_error};
|
||||
use crate::ddl::DdlContext;
|
||||
use crate::error::{self, Result};
|
||||
use crate::error::{self, Result, UnexpectedSnafu};
|
||||
use crate::instruction::{CacheIdent, CreateFlow};
|
||||
use crate::key::flow::flow_info::FlowInfoValue;
|
||||
use crate::key::flow::flow_route::FlowRouteValue;
|
||||
@@ -171,7 +171,7 @@ impl CreateFlowProcedure {
|
||||
}
|
||||
self.data.state = CreateFlowState::CreateFlows;
|
||||
// determine flow type
|
||||
self.data.flow_type = Some(determine_flow_type(&self.data.task));
|
||||
self.data.flow_type = Some(get_flow_type_from_options(&self.data.task)?);
|
||||
|
||||
Ok(Status::executing(true))
|
||||
}
|
||||
@@ -196,8 +196,8 @@ impl CreateFlowProcedure {
|
||||
});
|
||||
}
|
||||
info!(
|
||||
"Creating flow({:?}) on flownodes with peers={:?}",
|
||||
self.data.flow_id, self.data.peers
|
||||
"Creating flow({:?}, type={:?}) on flownodes with peers={:?}",
|
||||
self.data.flow_id, self.data.flow_type, self.data.peers
|
||||
);
|
||||
join_all(create_flow)
|
||||
.await
|
||||
@@ -306,8 +306,20 @@ impl Procedure for CreateFlowProcedure {
|
||||
}
|
||||
}
|
||||
|
||||
pub fn determine_flow_type(_flow_task: &CreateFlowTask) -> FlowType {
|
||||
FlowType::Batching
|
||||
pub fn get_flow_type_from_options(flow_task: &CreateFlowTask) -> Result<FlowType> {
|
||||
let flow_type = flow_task
|
||||
.flow_options
|
||||
.get(FlowType::FLOW_TYPE_KEY)
|
||||
.map(|s| s.as_str());
|
||||
match flow_type {
|
||||
Some(FlowType::BATCHING) => Ok(FlowType::Batching),
|
||||
Some(FlowType::STREAMING) => Ok(FlowType::Streaming),
|
||||
Some(unknown) => UnexpectedSnafu {
|
||||
err_msg: format!("Unknown flow type: {}", unknown),
|
||||
}
|
||||
.fail(),
|
||||
None => Ok(FlowType::Batching),
|
||||
}
|
||||
}
|
||||
|
||||
/// The state of [CreateFlowProcedure].
|
||||
@@ -437,6 +449,7 @@ impl From<&CreateFlowData> for (FlowInfoValue, Vec<(FlowPartitionId, FlowRouteVa
|
||||
sink_table_name,
|
||||
flownode_ids,
|
||||
catalog_name,
|
||||
query_context: Some(value.query_context.clone()),
|
||||
flow_name,
|
||||
raw_sql: sql,
|
||||
expire_after,
|
||||
|
||||
@@ -46,7 +46,7 @@ pub(crate) fn test_create_flow_task(
|
||||
create_if_not_exists,
|
||||
expire_after: Some(300),
|
||||
comment: "".to_string(),
|
||||
sql: "raw_sql".to_string(),
|
||||
sql: "select 1".to_string(),
|
||||
flow_options: Default::default(),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -401,6 +401,13 @@ pub enum Error {
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Invalid flow request body: {:?}", body))]
|
||||
InvalidFlowRequestBody {
|
||||
body: Box<Option<api::v1::flow::flow_request::Body>>,
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to get kv cache, err: {}", err_msg))]
|
||||
GetKvCache { err_msg: String },
|
||||
|
||||
@@ -783,6 +790,14 @@ pub enum Error {
|
||||
#[snafu(source)]
|
||||
source: common_procedure::error::Error,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to parse timezone"))]
|
||||
InvalidTimeZone {
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
#[snafu(source)]
|
||||
error: common_time::error::Error,
|
||||
},
|
||||
}
|
||||
|
||||
pub type Result<T> = std::result::Result<T, Error>;
|
||||
@@ -853,7 +868,9 @@ impl ErrorExt for Error {
|
||||
| TlsConfig { .. }
|
||||
| InvalidSetDatabaseOption { .. }
|
||||
| InvalidUnsetDatabaseOption { .. }
|
||||
| InvalidTopicNamePrefix { .. } => StatusCode::InvalidArguments,
|
||||
| InvalidTopicNamePrefix { .. }
|
||||
| InvalidTimeZone { .. } => StatusCode::InvalidArguments,
|
||||
InvalidFlowRequestBody { .. } => StatusCode::InvalidArguments,
|
||||
|
||||
FlowNotFound { .. } => StatusCode::FlowNotFound,
|
||||
FlowRouteNotFound { .. } => StatusCode::Unexpected,
|
||||
|
||||
@@ -452,6 +452,7 @@ mod tests {
|
||||
};
|
||||
FlowInfoValue {
|
||||
catalog_name: catalog_name.to_string(),
|
||||
query_context: None,
|
||||
flow_name: flow_name.to_string(),
|
||||
source_table_ids,
|
||||
sink_table_name,
|
||||
@@ -625,6 +626,7 @@ mod tests {
|
||||
};
|
||||
let flow_value = FlowInfoValue {
|
||||
catalog_name: "greptime".to_string(),
|
||||
query_context: None,
|
||||
flow_name: "flow".to_string(),
|
||||
source_table_ids: vec![1024, 1025, 1026],
|
||||
sink_table_name: another_sink_table_name,
|
||||
@@ -864,6 +866,7 @@ mod tests {
|
||||
};
|
||||
let flow_value = FlowInfoValue {
|
||||
catalog_name: "greptime".to_string(),
|
||||
query_context: None,
|
||||
flow_name: "flow".to_string(),
|
||||
source_table_ids: vec![1024, 1025, 1026],
|
||||
sink_table_name: another_sink_table_name,
|
||||
|
||||
@@ -121,6 +121,13 @@ pub struct FlowInfoValue {
|
||||
pub(crate) flownode_ids: BTreeMap<FlowPartitionId, FlownodeId>,
|
||||
/// The catalog name.
|
||||
pub(crate) catalog_name: String,
|
||||
/// The query context used when create flow.
|
||||
/// Although flow doesn't belong to any schema, this query_context is needed to remember
|
||||
/// the query context when `create_flow` is executed
|
||||
/// for recovering flow using the same sql&query_context after db restart.
|
||||
/// if none, should use default query context
|
||||
#[serde(default)]
|
||||
pub(crate) query_context: Option<crate::rpc::ddl::QueryContext>,
|
||||
/// The flow name.
|
||||
pub(crate) flow_name: String,
|
||||
/// The raw sql.
|
||||
@@ -155,6 +162,10 @@ impl FlowInfoValue {
|
||||
&self.catalog_name
|
||||
}
|
||||
|
||||
pub fn query_context(&self) -> &Option<crate::rpc::ddl::QueryContext> {
|
||||
&self.query_context
|
||||
}
|
||||
|
||||
pub fn flow_name(&self) -> &String {
|
||||
&self.flow_name
|
||||
}
|
||||
|
||||
@@ -113,8 +113,10 @@ impl LeaderRegionManifestInfo {
|
||||
pub fn prunable_entry_id(&self) -> u64 {
|
||||
match self {
|
||||
LeaderRegionManifestInfo::Mito {
|
||||
flushed_entry_id, ..
|
||||
} => *flushed_entry_id,
|
||||
flushed_entry_id,
|
||||
topic_latest_entry_id,
|
||||
..
|
||||
} => (*flushed_entry_id).max(*topic_latest_entry_id),
|
||||
LeaderRegionManifestInfo::Metric {
|
||||
data_flushed_entry_id,
|
||||
data_topic_latest_entry_id,
|
||||
|
||||
@@ -35,17 +35,20 @@ use api::v1::{
|
||||
};
|
||||
use base64::engine::general_purpose;
|
||||
use base64::Engine as _;
|
||||
use common_time::DatabaseTimeToLive;
|
||||
use common_time::{DatabaseTimeToLive, Timezone};
|
||||
use prost::Message;
|
||||
use serde::{Deserialize, Serialize};
|
||||
use serde_with::{serde_as, DefaultOnNull};
|
||||
use session::context::QueryContextRef;
|
||||
use session::context::{QueryContextBuilder, QueryContextRef};
|
||||
use snafu::{OptionExt, ResultExt};
|
||||
use table::metadata::{RawTableInfo, TableId};
|
||||
use table::table_name::TableName;
|
||||
use table::table_reference::TableReference;
|
||||
|
||||
use crate::error::{self, InvalidSetDatabaseOptionSnafu, InvalidUnsetDatabaseOptionSnafu, Result};
|
||||
use crate::error::{
|
||||
self, InvalidSetDatabaseOptionSnafu, InvalidTimeZoneSnafu, InvalidUnsetDatabaseOptionSnafu,
|
||||
Result,
|
||||
};
|
||||
use crate::key::FlowId;
|
||||
|
||||
/// DDL tasks
|
||||
@@ -1202,7 +1205,7 @@ impl From<DropFlowTask> for PbDropFlowTask {
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Debug, Clone, Serialize, Deserialize)]
|
||||
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq)]
|
||||
pub struct QueryContext {
|
||||
current_catalog: String,
|
||||
current_schema: String,
|
||||
@@ -1223,6 +1226,19 @@ impl From<QueryContextRef> for QueryContext {
|
||||
}
|
||||
}
|
||||
|
||||
impl TryFrom<QueryContext> for session::context::QueryContext {
|
||||
type Error = error::Error;
|
||||
fn try_from(value: QueryContext) -> std::result::Result<Self, Self::Error> {
|
||||
Ok(QueryContextBuilder::default()
|
||||
.current_catalog(value.current_catalog)
|
||||
.current_schema(value.current_schema)
|
||||
.timezone(Timezone::from_tz_string(&value.timezone).context(InvalidTimeZoneSnafu)?)
|
||||
.extensions(value.extensions)
|
||||
.channel((value.channel as u32).into())
|
||||
.build())
|
||||
}
|
||||
}
|
||||
|
||||
impl From<QueryContext> for PbQueryContext {
|
||||
fn from(
|
||||
QueryContext {
|
||||
|
||||
@@ -18,16 +18,19 @@ mod udaf;
|
||||
|
||||
use std::sync::Arc;
|
||||
|
||||
use api::v1::TableName;
|
||||
use datafusion::catalog::CatalogProviderList;
|
||||
use datafusion::error::Result as DatafusionResult;
|
||||
use datafusion::logical_expr::{LogicalPlan, LogicalPlanBuilder};
|
||||
use datafusion_common::Column;
|
||||
use datafusion_expr::col;
|
||||
use datafusion_common::{Column, TableReference};
|
||||
use datafusion_expr::dml::InsertOp;
|
||||
use datafusion_expr::{col, DmlStatement, WriteOp};
|
||||
pub use expr::{build_filter_from_timestamp, build_same_type_ts_filter};
|
||||
use snafu::ResultExt;
|
||||
|
||||
pub use self::accumulator::{Accumulator, AggregateFunctionCreator, AggregateFunctionCreatorRef};
|
||||
pub use self::udaf::AggregateFunction;
|
||||
use crate::error::Result;
|
||||
use crate::error::{GeneralDataFusionSnafu, Result};
|
||||
use crate::logical_plan::accumulator::*;
|
||||
use crate::signature::{Signature, Volatility};
|
||||
|
||||
@@ -79,6 +82,74 @@ pub fn rename_logical_plan_columns(
|
||||
LogicalPlanBuilder::from(plan).project(projection)?.build()
|
||||
}
|
||||
|
||||
/// Convert a insert into logical plan to an (table_name, logical_plan)
|
||||
/// where table_name is the name of the table to insert into.
|
||||
/// logical_plan is the plan to be executed.
|
||||
///
|
||||
/// if input logical plan is not `insert into table_name <input>`, return None
|
||||
///
|
||||
/// Returned TableName will use provided catalog and schema if not specified in the logical plan,
|
||||
/// if table scan in logical plan have full table name, will **NOT** override it.
|
||||
pub fn breakup_insert_plan(
|
||||
plan: &LogicalPlan,
|
||||
default_catalog: &str,
|
||||
default_schema: &str,
|
||||
) -> Option<(TableName, Arc<LogicalPlan>)> {
|
||||
if let LogicalPlan::Dml(dml) = plan {
|
||||
if dml.op != WriteOp::Insert(InsertOp::Append) {
|
||||
return None;
|
||||
}
|
||||
let table_name = &dml.table_name;
|
||||
let table_name = match table_name {
|
||||
TableReference::Bare { table } => TableName {
|
||||
catalog_name: default_catalog.to_string(),
|
||||
schema_name: default_schema.to_string(),
|
||||
table_name: table.to_string(),
|
||||
},
|
||||
TableReference::Partial { schema, table } => TableName {
|
||||
catalog_name: default_catalog.to_string(),
|
||||
schema_name: schema.to_string(),
|
||||
table_name: table.to_string(),
|
||||
},
|
||||
TableReference::Full {
|
||||
catalog,
|
||||
schema,
|
||||
table,
|
||||
} => TableName {
|
||||
catalog_name: catalog.to_string(),
|
||||
schema_name: schema.to_string(),
|
||||
table_name: table.to_string(),
|
||||
},
|
||||
};
|
||||
let logical_plan = dml.input.clone();
|
||||
Some((table_name, logical_plan))
|
||||
} else {
|
||||
None
|
||||
}
|
||||
}
|
||||
|
||||
/// create a `insert into table_name <input>` logical plan
|
||||
pub fn add_insert_to_logical_plan(
|
||||
table_name: TableName,
|
||||
table_schema: datafusion_common::DFSchemaRef,
|
||||
input: LogicalPlan,
|
||||
) -> Result<LogicalPlan> {
|
||||
let table_name = TableReference::Full {
|
||||
catalog: table_name.catalog_name.into(),
|
||||
schema: table_name.schema_name.into(),
|
||||
table: table_name.table_name.into(),
|
||||
};
|
||||
|
||||
let plan = LogicalPlan::Dml(DmlStatement::new(
|
||||
table_name,
|
||||
table_schema,
|
||||
WriteOp::Insert(InsertOp::Append),
|
||||
Arc::new(input),
|
||||
));
|
||||
let plan = plan.recompute_schema().context(GeneralDataFusionSnafu)?;
|
||||
Ok(plan)
|
||||
}
|
||||
|
||||
/// The datafusion `[LogicalPlan]` decoder.
|
||||
#[async_trait::async_trait]
|
||||
pub trait SubstraitPlanDecoder {
|
||||
|
||||
@@ -30,10 +30,10 @@ pub const DEFAULT_BACKOFF_CONFIG: BackoffConfig = BackoffConfig {
|
||||
deadline: Some(Duration::from_secs(120)),
|
||||
};
|
||||
|
||||
/// Default interval for active WAL pruning.
|
||||
pub const DEFAULT_ACTIVE_PRUNE_INTERVAL: Duration = Duration::ZERO;
|
||||
/// Default limit for concurrent active pruning tasks.
|
||||
pub const DEFAULT_ACTIVE_PRUNE_TASK_LIMIT: usize = 10;
|
||||
/// Default interval for auto WAL pruning.
|
||||
pub const DEFAULT_AUTO_PRUNE_INTERVAL: Duration = Duration::ZERO;
|
||||
/// Default limit for concurrent auto pruning tasks.
|
||||
pub const DEFAULT_AUTO_PRUNE_PARALLELISM: usize = 10;
|
||||
/// Default interval for sending flush request to regions when pruning remote WAL.
|
||||
pub const DEFAULT_TRIGGER_FLUSH_THRESHOLD: u64 = 0;
|
||||
|
||||
|
||||
@@ -18,8 +18,8 @@ use common_base::readable_size::ReadableSize;
|
||||
use serde::{Deserialize, Serialize};
|
||||
|
||||
use crate::config::kafka::common::{
|
||||
KafkaConnectionConfig, KafkaTopicConfig, DEFAULT_ACTIVE_PRUNE_INTERVAL,
|
||||
DEFAULT_ACTIVE_PRUNE_TASK_LIMIT, DEFAULT_TRIGGER_FLUSH_THRESHOLD,
|
||||
KafkaConnectionConfig, KafkaTopicConfig, DEFAULT_AUTO_PRUNE_INTERVAL,
|
||||
DEFAULT_AUTO_PRUNE_PARALLELISM, DEFAULT_TRIGGER_FLUSH_THRESHOLD,
|
||||
};
|
||||
|
||||
/// Kafka wal configurations for datanode.
|
||||
@@ -47,9 +47,8 @@ pub struct DatanodeKafkaConfig {
|
||||
pub dump_index_interval: Duration,
|
||||
/// Ignore missing entries during read WAL.
|
||||
pub overwrite_entry_start_id: bool,
|
||||
// Active WAL pruning.
|
||||
pub auto_prune_topic_records: bool,
|
||||
// Interval of WAL pruning.
|
||||
#[serde(with = "humantime_serde")]
|
||||
pub auto_prune_interval: Duration,
|
||||
// Threshold for sending flush request when pruning remote WAL.
|
||||
// `None` stands for never sending flush request.
|
||||
@@ -70,10 +69,9 @@ impl Default for DatanodeKafkaConfig {
|
||||
create_index: true,
|
||||
dump_index_interval: Duration::from_secs(60),
|
||||
overwrite_entry_start_id: false,
|
||||
auto_prune_topic_records: false,
|
||||
auto_prune_interval: DEFAULT_ACTIVE_PRUNE_INTERVAL,
|
||||
auto_prune_interval: DEFAULT_AUTO_PRUNE_INTERVAL,
|
||||
trigger_flush_threshold: DEFAULT_TRIGGER_FLUSH_THRESHOLD,
|
||||
auto_prune_parallelism: DEFAULT_ACTIVE_PRUNE_TASK_LIMIT,
|
||||
auto_prune_parallelism: DEFAULT_AUTO_PRUNE_PARALLELISM,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -17,8 +17,8 @@ use std::time::Duration;
|
||||
use serde::{Deserialize, Serialize};
|
||||
|
||||
use crate::config::kafka::common::{
|
||||
KafkaConnectionConfig, KafkaTopicConfig, DEFAULT_ACTIVE_PRUNE_INTERVAL,
|
||||
DEFAULT_ACTIVE_PRUNE_TASK_LIMIT, DEFAULT_TRIGGER_FLUSH_THRESHOLD,
|
||||
KafkaConnectionConfig, KafkaTopicConfig, DEFAULT_AUTO_PRUNE_INTERVAL,
|
||||
DEFAULT_AUTO_PRUNE_PARALLELISM, DEFAULT_TRIGGER_FLUSH_THRESHOLD,
|
||||
};
|
||||
|
||||
/// Kafka wal configurations for metasrv.
|
||||
@@ -34,6 +34,7 @@ pub struct MetasrvKafkaConfig {
|
||||
// Automatically create topics for WAL.
|
||||
pub auto_create_topics: bool,
|
||||
// Interval of WAL pruning.
|
||||
#[serde(with = "humantime_serde")]
|
||||
pub auto_prune_interval: Duration,
|
||||
// Threshold for sending flush request when pruning remote WAL.
|
||||
// `None` stands for never sending flush request.
|
||||
@@ -48,9 +49,9 @@ impl Default for MetasrvKafkaConfig {
|
||||
connection: Default::default(),
|
||||
kafka_topic: Default::default(),
|
||||
auto_create_topics: true,
|
||||
auto_prune_interval: DEFAULT_ACTIVE_PRUNE_INTERVAL,
|
||||
auto_prune_interval: DEFAULT_AUTO_PRUNE_INTERVAL,
|
||||
trigger_flush_threshold: DEFAULT_TRIGGER_FLUSH_THRESHOLD,
|
||||
auto_prune_parallelism: DEFAULT_ACTIVE_PRUNE_TASK_LIMIT,
|
||||
auto_prune_parallelism: DEFAULT_AUTO_PRUNE_PARALLELISM,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -57,9 +57,9 @@ use tokio::sync::Notify;
|
||||
|
||||
use crate::config::{DatanodeOptions, RegionEngineConfig, StorageConfig};
|
||||
use crate::error::{
|
||||
self, BuildMitoEngineSnafu, CreateDirSnafu, GetMetadataSnafu, MissingCacheSnafu,
|
||||
MissingKvBackendSnafu, MissingNodeIdSnafu, OpenLogStoreSnafu, Result, ShutdownInstanceSnafu,
|
||||
ShutdownServerSnafu, StartServerSnafu,
|
||||
self, BuildMetricEngineSnafu, BuildMitoEngineSnafu, CreateDirSnafu, GetMetadataSnafu,
|
||||
MissingCacheSnafu, MissingKvBackendSnafu, MissingNodeIdSnafu, OpenLogStoreSnafu, Result,
|
||||
ShutdownInstanceSnafu, ShutdownServerSnafu, StartServerSnafu,
|
||||
};
|
||||
use crate::event_listener::{
|
||||
new_region_server_event_channel, NoopRegionServerEventListener, RegionServerEventListenerRef,
|
||||
@@ -416,10 +416,11 @@ impl DatanodeBuilder {
|
||||
)
|
||||
.await?;
|
||||
|
||||
let metric_engine = MetricEngine::new(
|
||||
let metric_engine = MetricEngine::try_new(
|
||||
mito_engine.clone(),
|
||||
metric_engine_config.take().unwrap_or_default(),
|
||||
);
|
||||
)
|
||||
.context(BuildMetricEngineSnafu)?;
|
||||
engines.push(Arc::new(mito_engine) as _);
|
||||
engines.push(Arc::new(metric_engine) as _);
|
||||
}
|
||||
|
||||
@@ -336,6 +336,13 @@ pub enum Error {
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to build metric engine"))]
|
||||
BuildMetricEngine {
|
||||
source: metric_engine::error::Error,
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to serialize options to TOML"))]
|
||||
TomlFormat {
|
||||
#[snafu(implicit)]
|
||||
@@ -452,6 +459,7 @@ impl ErrorExt for Error {
|
||||
|
||||
FindLogicalRegions { source, .. } => source.status_code(),
|
||||
BuildMitoEngine { source, .. } => source.status_code(),
|
||||
BuildMetricEngine { source, .. } => source.status_code(),
|
||||
ConcurrentQueryLimiterClosed { .. } | ConcurrentQueryLimiterTimeout { .. } => {
|
||||
StatusCode::RegionBusy
|
||||
}
|
||||
|
||||
@@ -58,7 +58,7 @@ use crate::metrics::{METRIC_FLOW_INSERT_ELAPSED, METRIC_FLOW_ROWS, METRIC_FLOW_R
|
||||
use crate::repr::{self, DiffRow, RelationDesc, Row, BATCH_SIZE};
|
||||
use crate::{CreateFlowArgs, FlowId, TableName};
|
||||
|
||||
mod flownode_impl;
|
||||
pub(crate) mod flownode_impl;
|
||||
mod parse_expr;
|
||||
pub(crate) mod refill;
|
||||
mod stat;
|
||||
@@ -135,12 +135,13 @@ impl Configurable for FlownodeOptions {
|
||||
}
|
||||
|
||||
/// Arc-ed FlowNodeManager, cheaper to clone
|
||||
pub type FlowWorkerManagerRef = Arc<FlowWorkerManager>;
|
||||
pub type FlowStreamingEngineRef = Arc<StreamingEngine>;
|
||||
|
||||
/// FlowNodeManager manages the state of all tasks in the flow node, which should be run on the same thread
|
||||
///
|
||||
/// The choice of timestamp is just using current system timestamp for now
|
||||
pub struct FlowWorkerManager {
|
||||
///
|
||||
pub struct StreamingEngine {
|
||||
/// The handler to the worker that will run the dataflow
|
||||
/// which is `!Send` so a handle is used
|
||||
pub worker_handles: Vec<WorkerHandle>,
|
||||
@@ -158,7 +159,8 @@ pub struct FlowWorkerManager {
|
||||
flow_err_collectors: RwLock<BTreeMap<FlowId, ErrCollector>>,
|
||||
src_send_buf_lens: RwLock<BTreeMap<TableId, watch::Receiver<usize>>>,
|
||||
tick_manager: FlowTickManager,
|
||||
node_id: Option<u32>,
|
||||
/// This node id is only available in distributed mode, on standalone mode this is guaranteed to be `None`
|
||||
pub node_id: Option<u32>,
|
||||
/// Lock for flushing, will be `read` by `handle_inserts` and `write` by `flush_flow`
|
||||
///
|
||||
/// So that a series of event like `inserts -> flush` can be handled correctly
|
||||
@@ -168,7 +170,7 @@ pub struct FlowWorkerManager {
|
||||
}
|
||||
|
||||
/// Building FlownodeManager
|
||||
impl FlowWorkerManager {
|
||||
impl StreamingEngine {
|
||||
/// set frontend invoker
|
||||
pub async fn set_frontend_invoker(&self, frontend: FrontendInvoker) {
|
||||
*self.frontend_invoker.write().await = Some(frontend);
|
||||
@@ -187,7 +189,7 @@ impl FlowWorkerManager {
|
||||
let node_context = FlownodeContext::new(Box::new(srv_map.clone()) as _);
|
||||
let tick_manager = FlowTickManager::new();
|
||||
let worker_handles = Vec::new();
|
||||
FlowWorkerManager {
|
||||
StreamingEngine {
|
||||
worker_handles,
|
||||
worker_selector: Mutex::new(0),
|
||||
query_engine,
|
||||
@@ -263,7 +265,7 @@ pub fn batches_to_rows_req(batches: Vec<Batch>) -> Result<Vec<DiffRequest>, Erro
|
||||
}
|
||||
|
||||
/// This impl block contains methods to send writeback requests to frontend
|
||||
impl FlowWorkerManager {
|
||||
impl StreamingEngine {
|
||||
/// Return the number of requests it made
|
||||
pub async fn send_writeback_requests(&self) -> Result<usize, Error> {
|
||||
let all_reqs = self.generate_writeback_request().await?;
|
||||
@@ -534,7 +536,7 @@ impl FlowWorkerManager {
|
||||
}
|
||||
|
||||
/// Flow Runtime related methods
|
||||
impl FlowWorkerManager {
|
||||
impl StreamingEngine {
|
||||
/// Start state report handler, which will receive a sender from HeartbeatTask to send state size report back
|
||||
///
|
||||
/// if heartbeat task is shutdown, this future will exit too
|
||||
@@ -659,7 +661,7 @@ impl FlowWorkerManager {
|
||||
}
|
||||
// flow is now shutdown, drop frontend_invoker early so a ref cycle(in standalone mode) can be prevent:
|
||||
// FlowWorkerManager.frontend_invoker -> FrontendInvoker.inserter
|
||||
// -> Inserter.node_manager -> NodeManager.flownode -> Flownode.flow_worker_manager.frontend_invoker
|
||||
// -> Inserter.node_manager -> NodeManager.flownode -> Flownode.flow_streaming_engine.frontend_invoker
|
||||
self.frontend_invoker.write().await.take();
|
||||
}
|
||||
|
||||
@@ -728,7 +730,7 @@ impl FlowWorkerManager {
|
||||
}
|
||||
|
||||
/// Create&Remove flow
|
||||
impl FlowWorkerManager {
|
||||
impl StreamingEngine {
|
||||
/// remove a flow by it's id
|
||||
pub async fn remove_flow_inner(&self, flow_id: FlowId) -> Result<(), Error> {
|
||||
for handle in self.worker_handles.iter() {
|
||||
@@ -746,7 +748,6 @@ impl FlowWorkerManager {
|
||||
/// steps to create task:
|
||||
/// 1. parse query into typed plan(and optional parse expire_after expr)
|
||||
/// 2. render source/sink with output table id and used input table id
|
||||
#[allow(clippy::too_many_arguments)]
|
||||
pub async fn create_flow_inner(&self, args: CreateFlowArgs) -> Result<Option<FlowId>, Error> {
|
||||
let CreateFlowArgs {
|
||||
flow_id,
|
||||
|
||||
@@ -20,35 +20,394 @@ use api::v1::flow::{
|
||||
flow_request, CreateRequest, DropRequest, FlowRequest, FlowResponse, FlushFlow,
|
||||
};
|
||||
use api::v1::region::InsertRequests;
|
||||
use catalog::CatalogManager;
|
||||
use common_error::ext::BoxedError;
|
||||
use common_meta::ddl::create_flow::FlowType;
|
||||
use common_meta::error::{Result as MetaResult, UnexpectedSnafu};
|
||||
use common_meta::error::Result as MetaResult;
|
||||
use common_meta::key::flow::FlowMetadataManager;
|
||||
use common_runtime::JoinHandle;
|
||||
use common_telemetry::{trace, warn};
|
||||
use common_telemetry::{error, info, trace, warn};
|
||||
use datatypes::value::Value;
|
||||
use futures::TryStreamExt;
|
||||
use itertools::Itertools;
|
||||
use snafu::{IntoError, OptionExt, ResultExt};
|
||||
use session::context::QueryContextBuilder;
|
||||
use snafu::{ensure, IntoError, OptionExt, ResultExt};
|
||||
use store_api::storage::{RegionId, TableId};
|
||||
use tokio::sync::{Mutex, RwLock};
|
||||
|
||||
use crate::adapter::{CreateFlowArgs, FlowWorkerManager};
|
||||
use crate::adapter::{CreateFlowArgs, StreamingEngine};
|
||||
use crate::batching_mode::engine::BatchingEngine;
|
||||
use crate::engine::FlowEngine;
|
||||
use crate::error::{CreateFlowSnafu, FlowNotFoundSnafu, InsertIntoFlowSnafu, InternalSnafu};
|
||||
use crate::error::{
|
||||
CreateFlowSnafu, ExternalSnafu, FlowNotFoundSnafu, IllegalCheckTaskStateSnafu,
|
||||
InsertIntoFlowSnafu, InternalSnafu, JoinTaskSnafu, ListFlowsSnafu, SyncCheckTaskSnafu,
|
||||
UnexpectedSnafu,
|
||||
};
|
||||
use crate::metrics::METRIC_FLOW_TASK_COUNT;
|
||||
use crate::repr::{self, DiffRow};
|
||||
use crate::{Error, FlowId};
|
||||
|
||||
/// Ref to [`FlowDualEngine`]
|
||||
pub type FlowDualEngineRef = Arc<FlowDualEngine>;
|
||||
|
||||
/// Manage both streaming and batching mode engine
|
||||
///
|
||||
/// including create/drop/flush flow
|
||||
/// and redirect insert requests to the appropriate engine
|
||||
pub struct FlowDualEngine {
|
||||
streaming_engine: Arc<FlowWorkerManager>,
|
||||
streaming_engine: Arc<StreamingEngine>,
|
||||
batching_engine: Arc<BatchingEngine>,
|
||||
/// helper struct for faster query flow by table id or vice versa
|
||||
src_table2flow: std::sync::RwLock<SrcTableToFlow>,
|
||||
src_table2flow: RwLock<SrcTableToFlow>,
|
||||
flow_metadata_manager: Arc<FlowMetadataManager>,
|
||||
catalog_manager: Arc<dyn CatalogManager>,
|
||||
check_task: tokio::sync::Mutex<Option<ConsistentCheckTask>>,
|
||||
}
|
||||
|
||||
impl FlowDualEngine {
|
||||
pub fn new(
|
||||
streaming_engine: Arc<StreamingEngine>,
|
||||
batching_engine: Arc<BatchingEngine>,
|
||||
flow_metadata_manager: Arc<FlowMetadataManager>,
|
||||
catalog_manager: Arc<dyn CatalogManager>,
|
||||
) -> Self {
|
||||
Self {
|
||||
streaming_engine,
|
||||
batching_engine,
|
||||
src_table2flow: RwLock::new(SrcTableToFlow::default()),
|
||||
flow_metadata_manager,
|
||||
catalog_manager,
|
||||
check_task: Mutex::new(None),
|
||||
}
|
||||
}
|
||||
|
||||
pub fn streaming_engine(&self) -> Arc<StreamingEngine> {
|
||||
self.streaming_engine.clone()
|
||||
}
|
||||
|
||||
pub fn batching_engine(&self) -> Arc<BatchingEngine> {
|
||||
self.batching_engine.clone()
|
||||
}
|
||||
|
||||
/// Try to sync with check task, this is only used in drop flow&flush flow, so a flow id is required
|
||||
///
|
||||
/// the need to sync is to make sure flush flow actually get called
|
||||
async fn try_sync_with_check_task(
|
||||
&self,
|
||||
flow_id: FlowId,
|
||||
allow_drop: bool,
|
||||
) -> Result<(), Error> {
|
||||
// this function rarely get called so adding some log is helpful
|
||||
info!("Try to sync with check task for flow {}", flow_id);
|
||||
let mut retry = 0;
|
||||
let max_retry = 10;
|
||||
// keep trying to trigger consistent check
|
||||
while retry < max_retry {
|
||||
if let Some(task) = self.check_task.lock().await.as_ref() {
|
||||
task.trigger(false, allow_drop).await?;
|
||||
break;
|
||||
}
|
||||
retry += 1;
|
||||
tokio::time::sleep(std::time::Duration::from_millis(500)).await;
|
||||
}
|
||||
|
||||
if retry == max_retry {
|
||||
error!(
|
||||
"Can't sync with check task for flow {} with allow_drop={}",
|
||||
flow_id, allow_drop
|
||||
);
|
||||
return SyncCheckTaskSnafu {
|
||||
flow_id,
|
||||
allow_drop,
|
||||
}
|
||||
.fail();
|
||||
}
|
||||
info!("Successfully sync with check task for flow {}", flow_id);
|
||||
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Spawn a task to consistently check if all flow tasks in metasrv is created on flownode,
|
||||
/// so on startup, this will create all missing flow tasks, and constantly check at a interval
|
||||
async fn check_flow_consistent(
|
||||
&self,
|
||||
allow_create: bool,
|
||||
allow_drop: bool,
|
||||
) -> Result<(), Error> {
|
||||
// use nodeid to determine if this is standalone/distributed mode, and retrieve all flows in this node(in distributed mode)/or all flows(in standalone mode)
|
||||
let nodeid = self.streaming_engine.node_id;
|
||||
let should_exists: Vec<_> = if let Some(nodeid) = nodeid {
|
||||
// nodeid is available, so we only need to check flows on this node
|
||||
// which also means we are in distributed mode
|
||||
let to_be_recover = self
|
||||
.flow_metadata_manager
|
||||
.flownode_flow_manager()
|
||||
.flows(nodeid.into())
|
||||
.try_collect::<Vec<_>>()
|
||||
.await
|
||||
.context(ListFlowsSnafu {
|
||||
id: Some(nodeid.into()),
|
||||
})?;
|
||||
to_be_recover.into_iter().map(|(id, _)| id).collect()
|
||||
} else {
|
||||
// nodeid is not available, so we need to check all flows
|
||||
// which also means we are in standalone mode
|
||||
let all_catalogs = self
|
||||
.catalog_manager
|
||||
.catalog_names()
|
||||
.await
|
||||
.map_err(BoxedError::new)
|
||||
.context(ExternalSnafu)?;
|
||||
let mut all_flow_ids = vec![];
|
||||
for catalog in all_catalogs {
|
||||
let flows = self
|
||||
.flow_metadata_manager
|
||||
.flow_name_manager()
|
||||
.flow_names(&catalog)
|
||||
.await
|
||||
.try_collect::<Vec<_>>()
|
||||
.await
|
||||
.map_err(BoxedError::new)
|
||||
.context(ExternalSnafu)?;
|
||||
|
||||
all_flow_ids.extend(flows.into_iter().map(|(_, id)| id.flow_id()));
|
||||
}
|
||||
all_flow_ids
|
||||
};
|
||||
let should_exists = should_exists
|
||||
.into_iter()
|
||||
.map(|i| i as FlowId)
|
||||
.collect::<HashSet<_>>();
|
||||
let actual_exists = self.list_flows().await?.into_iter().collect::<HashSet<_>>();
|
||||
let to_be_created = should_exists
|
||||
.iter()
|
||||
.filter(|id| !actual_exists.contains(id))
|
||||
.collect::<Vec<_>>();
|
||||
let to_be_dropped = actual_exists
|
||||
.iter()
|
||||
.filter(|id| !should_exists.contains(id))
|
||||
.collect::<Vec<_>>();
|
||||
|
||||
if !to_be_created.is_empty() {
|
||||
if allow_create {
|
||||
info!(
|
||||
"Recovering {} flows: {:?}",
|
||||
to_be_created.len(),
|
||||
to_be_created
|
||||
);
|
||||
let mut errors = vec![];
|
||||
for flow_id in to_be_created {
|
||||
let flow_id = *flow_id;
|
||||
let info = self
|
||||
.flow_metadata_manager
|
||||
.flow_info_manager()
|
||||
.get(flow_id as u32)
|
||||
.await
|
||||
.map_err(BoxedError::new)
|
||||
.context(ExternalSnafu)?
|
||||
.context(FlowNotFoundSnafu { id: flow_id })?;
|
||||
|
||||
let sink_table_name = [
|
||||
info.sink_table_name().catalog_name.clone(),
|
||||
info.sink_table_name().schema_name.clone(),
|
||||
info.sink_table_name().table_name.clone(),
|
||||
];
|
||||
let args = CreateFlowArgs {
|
||||
flow_id,
|
||||
sink_table_name,
|
||||
source_table_ids: info.source_table_ids().to_vec(),
|
||||
// because recover should only happen on restart the `create_if_not_exists` and `or_replace` can be arbitrary value(since flow doesn't exist)
|
||||
// but for the sake of consistency and to make sure recover of flow actually happen, we set both to true
|
||||
// (which is also fine since checks for not allow both to be true is on metasrv and we already pass that)
|
||||
create_if_not_exists: true,
|
||||
or_replace: true,
|
||||
expire_after: info.expire_after(),
|
||||
comment: Some(info.comment().clone()),
|
||||
sql: info.raw_sql().clone(),
|
||||
flow_options: info.options().clone(),
|
||||
query_ctx: info
|
||||
.query_context()
|
||||
.clone()
|
||||
.map(|ctx| {
|
||||
ctx.try_into()
|
||||
.map_err(BoxedError::new)
|
||||
.context(ExternalSnafu)
|
||||
})
|
||||
.transpose()?
|
||||
// or use default QueryContext with catalog_name from info
|
||||
// to keep compatibility with old version
|
||||
.or_else(|| {
|
||||
Some(
|
||||
QueryContextBuilder::default()
|
||||
.current_catalog(info.catalog_name().to_string())
|
||||
.build(),
|
||||
)
|
||||
}),
|
||||
};
|
||||
if let Err(err) = self
|
||||
.create_flow(args)
|
||||
.await
|
||||
.map_err(BoxedError::new)
|
||||
.with_context(|_| CreateFlowSnafu {
|
||||
sql: info.raw_sql().clone(),
|
||||
})
|
||||
{
|
||||
errors.push((flow_id, err));
|
||||
}
|
||||
}
|
||||
for (flow_id, err) in errors {
|
||||
warn!("Failed to recreate flow {}, err={:#?}", flow_id, err);
|
||||
}
|
||||
} else {
|
||||
warn!(
|
||||
"Flownode {:?} found flows not exist in flownode, flow_ids={:?}",
|
||||
nodeid, to_be_created
|
||||
);
|
||||
}
|
||||
}
|
||||
if !to_be_dropped.is_empty() {
|
||||
if allow_drop {
|
||||
info!("Dropping flows: {:?}", to_be_dropped);
|
||||
let mut errors = vec![];
|
||||
for flow_id in to_be_dropped {
|
||||
let flow_id = *flow_id;
|
||||
if let Err(err) = self.remove_flow(flow_id).await {
|
||||
errors.push((flow_id, err));
|
||||
}
|
||||
}
|
||||
for (flow_id, err) in errors {
|
||||
warn!("Failed to drop flow {}, err={:#?}", flow_id, err);
|
||||
}
|
||||
} else {
|
||||
warn!(
|
||||
"Flownode {:?} found flows not exist in flownode, flow_ids={:?}",
|
||||
nodeid, to_be_dropped
|
||||
);
|
||||
}
|
||||
}
|
||||
Ok(())
|
||||
}
|
||||
|
||||
// TODO(discord9): consider sync this with heartbeat(might become necessary in the future)
|
||||
pub async fn start_flow_consistent_check_task(self: &Arc<Self>) -> Result<(), Error> {
|
||||
let mut check_task = self.check_task.lock().await;
|
||||
ensure!(
|
||||
check_task.is_none(),
|
||||
IllegalCheckTaskStateSnafu {
|
||||
reason: "Flow consistent check task already exists",
|
||||
}
|
||||
);
|
||||
let task = ConsistentCheckTask::start_check_task(self).await?;
|
||||
*check_task = Some(task);
|
||||
Ok(())
|
||||
}
|
||||
|
||||
pub async fn stop_flow_consistent_check_task(&self) -> Result<(), Error> {
|
||||
info!("Stopping flow consistent check task");
|
||||
let mut check_task = self.check_task.lock().await;
|
||||
|
||||
ensure!(
|
||||
check_task.is_some(),
|
||||
IllegalCheckTaskStateSnafu {
|
||||
reason: "Flow consistent check task does not exist",
|
||||
}
|
||||
);
|
||||
|
||||
check_task.take().unwrap().stop().await?;
|
||||
info!("Stopped flow consistent check task");
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// TODO(discord9): also add a `exists` api using flow metadata manager's `exists` method
|
||||
async fn flow_exist_in_metadata(&self, flow_id: FlowId) -> Result<bool, Error> {
|
||||
self.flow_metadata_manager
|
||||
.flow_info_manager()
|
||||
.get(flow_id as u32)
|
||||
.await
|
||||
.map_err(BoxedError::new)
|
||||
.context(ExternalSnafu)
|
||||
.map(|info| info.is_some())
|
||||
}
|
||||
}
|
||||
|
||||
struct ConsistentCheckTask {
|
||||
handle: JoinHandle<()>,
|
||||
shutdown_tx: tokio::sync::mpsc::Sender<()>,
|
||||
trigger_tx: tokio::sync::mpsc::Sender<(bool, bool, tokio::sync::oneshot::Sender<()>)>,
|
||||
}
|
||||
|
||||
impl ConsistentCheckTask {
|
||||
async fn start_check_task(engine: &Arc<FlowDualEngine>) -> Result<Self, Error> {
|
||||
// first do recover flows
|
||||
engine.check_flow_consistent(true, false).await?;
|
||||
|
||||
let inner = engine.clone();
|
||||
let (tx, mut rx) = tokio::sync::mpsc::channel(1);
|
||||
let (trigger_tx, mut trigger_rx) =
|
||||
tokio::sync::mpsc::channel::<(bool, bool, tokio::sync::oneshot::Sender<()>)>(10);
|
||||
let handle = common_runtime::spawn_global(async move {
|
||||
let (mut allow_create, mut allow_drop) = (false, false);
|
||||
let mut ret_signal: Option<tokio::sync::oneshot::Sender<()>> = None;
|
||||
loop {
|
||||
if let Err(err) = inner.check_flow_consistent(allow_create, allow_drop).await {
|
||||
error!(err; "Failed to check flow consistent");
|
||||
}
|
||||
if let Some(done) = ret_signal.take() {
|
||||
let _ = done.send(());
|
||||
}
|
||||
tokio::select! {
|
||||
_ = rx.recv() => break,
|
||||
incoming = trigger_rx.recv() => if let Some(incoming) = incoming {
|
||||
(allow_create, allow_drop) = (incoming.0, incoming.1);
|
||||
ret_signal = Some(incoming.2);
|
||||
},
|
||||
_ = tokio::time::sleep(std::time::Duration::from_secs(10)) => {
|
||||
(allow_create, allow_drop) = (false, false);
|
||||
},
|
||||
}
|
||||
}
|
||||
});
|
||||
Ok(ConsistentCheckTask {
|
||||
handle,
|
||||
shutdown_tx: tx,
|
||||
trigger_tx,
|
||||
})
|
||||
}
|
||||
|
||||
async fn trigger(&self, allow_create: bool, allow_drop: bool) -> Result<(), Error> {
|
||||
let (tx, rx) = tokio::sync::oneshot::channel();
|
||||
self.trigger_tx
|
||||
.send((allow_create, allow_drop, tx))
|
||||
.await
|
||||
.map_err(|_| {
|
||||
IllegalCheckTaskStateSnafu {
|
||||
reason: "Failed to send trigger signal",
|
||||
}
|
||||
.build()
|
||||
})?;
|
||||
rx.await.map_err(|_| {
|
||||
IllegalCheckTaskStateSnafu {
|
||||
reason: "Failed to receive trigger signal",
|
||||
}
|
||||
.build()
|
||||
})?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
async fn stop(self) -> Result<(), Error> {
|
||||
self.shutdown_tx.send(()).await.map_err(|_| {
|
||||
IllegalCheckTaskStateSnafu {
|
||||
reason: "Failed to send shutdown signal",
|
||||
}
|
||||
.build()
|
||||
})?;
|
||||
// abort so no need to wait
|
||||
self.handle.abort();
|
||||
Ok(())
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Default)]
|
||||
struct SrcTableToFlow {
|
||||
/// mapping of table ids to flow ids for streaming mode
|
||||
stream: HashMap<TableId, HashSet<FlowId>>,
|
||||
@@ -138,35 +497,49 @@ impl FlowEngine for FlowDualEngine {
|
||||
|
||||
self.src_table2flow
|
||||
.write()
|
||||
.unwrap()
|
||||
.await
|
||||
.add_flow(flow_id, flow_type, src_table_ids);
|
||||
|
||||
Ok(res)
|
||||
}
|
||||
|
||||
async fn remove_flow(&self, flow_id: FlowId) -> Result<(), Error> {
|
||||
let flow_type = self.src_table2flow.read().unwrap().get_flow_type(flow_id);
|
||||
let flow_type = self.src_table2flow.read().await.get_flow_type(flow_id);
|
||||
|
||||
match flow_type {
|
||||
Some(FlowType::Batching) => self.batching_engine.remove_flow(flow_id).await,
|
||||
Some(FlowType::Streaming) => self.streaming_engine.remove_flow(flow_id).await,
|
||||
None => FlowNotFoundSnafu { id: flow_id }.fail(),
|
||||
None => {
|
||||
// this can happen if flownode just restart, and is stilling creating the flow
|
||||
// since now that this flow should dropped, we need to trigger the consistent check and allow drop
|
||||
// this rely on drop flow ddl delete metadata first, see src/common/meta/src/ddl/drop_flow.rs
|
||||
warn!(
|
||||
"Flow {} is not exist in the underlying engine, but exist in metadata",
|
||||
flow_id
|
||||
);
|
||||
self.try_sync_with_check_task(flow_id, true).await?;
|
||||
|
||||
Ok(())
|
||||
}
|
||||
}?;
|
||||
// remove mapping
|
||||
self.src_table2flow.write().unwrap().remove_flow(flow_id);
|
||||
self.src_table2flow.write().await.remove_flow(flow_id);
|
||||
Ok(())
|
||||
}
|
||||
|
||||
async fn flush_flow(&self, flow_id: FlowId) -> Result<usize, Error> {
|
||||
let flow_type = self.src_table2flow.read().unwrap().get_flow_type(flow_id);
|
||||
// sync with check task
|
||||
self.try_sync_with_check_task(flow_id, false).await?;
|
||||
let flow_type = self.src_table2flow.read().await.get_flow_type(flow_id);
|
||||
match flow_type {
|
||||
Some(FlowType::Batching) => self.batching_engine.flush_flow(flow_id).await,
|
||||
Some(FlowType::Streaming) => self.streaming_engine.flush_flow(flow_id).await,
|
||||
None => FlowNotFoundSnafu { id: flow_id }.fail(),
|
||||
None => Ok(0),
|
||||
}
|
||||
}
|
||||
|
||||
async fn flow_exist(&self, flow_id: FlowId) -> Result<bool, Error> {
|
||||
let flow_type = self.src_table2flow.read().unwrap().get_flow_type(flow_id);
|
||||
let flow_type = self.src_table2flow.read().await.get_flow_type(flow_id);
|
||||
// not using `flow_type.is_some()` to make sure the flow is actually exist in the underlying engine
|
||||
match flow_type {
|
||||
Some(FlowType::Batching) => self.batching_engine.flow_exist(flow_id).await,
|
||||
@@ -175,6 +548,13 @@ impl FlowEngine for FlowDualEngine {
|
||||
}
|
||||
}
|
||||
|
||||
async fn list_flows(&self) -> Result<impl IntoIterator<Item = FlowId>, Error> {
|
||||
let stream_flows = self.streaming_engine.list_flows().await?;
|
||||
let batch_flows = self.batching_engine.list_flows().await?;
|
||||
|
||||
Ok(stream_flows.into_iter().chain(batch_flows))
|
||||
}
|
||||
|
||||
async fn handle_flow_inserts(
|
||||
&self,
|
||||
request: api::v1::region::InsertRequests,
|
||||
@@ -184,7 +564,7 @@ impl FlowEngine for FlowDualEngine {
|
||||
let mut to_batch_engine = request.requests;
|
||||
|
||||
{
|
||||
let src_table2flow = self.src_table2flow.read().unwrap();
|
||||
let src_table2flow = self.src_table2flow.read().await;
|
||||
to_batch_engine.retain(|req| {
|
||||
let region_id = RegionId::from(req.region_id);
|
||||
let table_id = region_id.table_id();
|
||||
@@ -221,12 +601,7 @@ impl FlowEngine for FlowDualEngine {
|
||||
requests: to_batch_engine,
|
||||
})
|
||||
.await?;
|
||||
stream_handler.await.map_err(|e| {
|
||||
crate::error::UnexpectedSnafu {
|
||||
reason: format!("JoinError when handle inserts for flow stream engine: {e:?}"),
|
||||
}
|
||||
.build()
|
||||
})??;
|
||||
stream_handler.await.context(JoinTaskSnafu)??;
|
||||
|
||||
Ok(())
|
||||
}
|
||||
@@ -307,14 +682,7 @@ impl common_meta::node_manager::Flownode for FlowDualEngine {
|
||||
..Default::default()
|
||||
})
|
||||
}
|
||||
None => UnexpectedSnafu {
|
||||
err_msg: "Missing request body",
|
||||
}
|
||||
.fail(),
|
||||
_ => UnexpectedSnafu {
|
||||
err_msg: "Invalid request body.",
|
||||
}
|
||||
.fail(),
|
||||
other => common_meta::error::InvalidFlowRequestBodySnafu { body: other }.fail(),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -339,7 +707,7 @@ fn to_meta_err(
|
||||
}
|
||||
|
||||
#[async_trait::async_trait]
|
||||
impl common_meta::node_manager::Flownode for FlowWorkerManager {
|
||||
impl common_meta::node_manager::Flownode for StreamingEngine {
|
||||
async fn handle(&self, request: FlowRequest) -> MetaResult<FlowResponse> {
|
||||
let query_ctx = request
|
||||
.header
|
||||
@@ -413,14 +781,7 @@ impl common_meta::node_manager::Flownode for FlowWorkerManager {
|
||||
..Default::default()
|
||||
})
|
||||
}
|
||||
None => UnexpectedSnafu {
|
||||
err_msg: "Missing request body",
|
||||
}
|
||||
.fail(),
|
||||
_ => UnexpectedSnafu {
|
||||
err_msg: "Invalid request body.",
|
||||
}
|
||||
.fail(),
|
||||
other => common_meta::error::InvalidFlowRequestBodySnafu { body: other }.fail(),
|
||||
}
|
||||
}
|
||||
|
||||
@@ -432,7 +793,7 @@ impl common_meta::node_manager::Flownode for FlowWorkerManager {
|
||||
}
|
||||
}
|
||||
|
||||
impl FlowEngine for FlowWorkerManager {
|
||||
impl FlowEngine for StreamingEngine {
|
||||
async fn create_flow(&self, args: CreateFlowArgs) -> Result<Option<FlowId>, Error> {
|
||||
self.create_flow_inner(args).await
|
||||
}
|
||||
@@ -449,6 +810,16 @@ impl FlowEngine for FlowWorkerManager {
|
||||
self.flow_exist_inner(flow_id).await
|
||||
}
|
||||
|
||||
async fn list_flows(&self) -> Result<impl IntoIterator<Item = FlowId>, Error> {
|
||||
Ok(self
|
||||
.flow_err_collectors
|
||||
.read()
|
||||
.await
|
||||
.keys()
|
||||
.cloned()
|
||||
.collect::<Vec<_>>())
|
||||
}
|
||||
|
||||
async fn handle_flow_inserts(
|
||||
&self,
|
||||
request: api::v1::region::InsertRequests,
|
||||
@@ -474,7 +845,7 @@ impl FetchFromRow {
|
||||
}
|
||||
}
|
||||
|
||||
impl FlowWorkerManager {
|
||||
impl StreamingEngine {
|
||||
async fn handle_inserts_inner(
|
||||
&self,
|
||||
request: InsertRequests,
|
||||
@@ -552,7 +923,7 @@ impl FlowWorkerManager {
|
||||
.copied()
|
||||
.map(FetchFromRow::Idx)
|
||||
.or_else(|| col_default_val.clone().map(FetchFromRow::Default))
|
||||
.with_context(|| crate::error::UnexpectedSnafu {
|
||||
.with_context(|| UnexpectedSnafu {
|
||||
reason: format!(
|
||||
"Column not found: {}, default_value: {:?}",
|
||||
col_name, col_default_val
|
||||
|
||||
@@ -31,7 +31,7 @@ use snafu::{ensure, OptionExt, ResultExt};
|
||||
use table::metadata::TableId;
|
||||
|
||||
use crate::adapter::table_source::ManagedTableSource;
|
||||
use crate::adapter::{FlowId, FlowWorkerManager, FlowWorkerManagerRef};
|
||||
use crate::adapter::{FlowId, FlowStreamingEngineRef, StreamingEngine};
|
||||
use crate::error::{FlowNotFoundSnafu, JoinTaskSnafu, UnexpectedSnafu};
|
||||
use crate::expr::error::ExternalSnafu;
|
||||
use crate::expr::utils::find_plan_time_window_expr_lower_bound;
|
||||
@@ -39,10 +39,10 @@ use crate::repr::RelationDesc;
|
||||
use crate::server::get_all_flow_ids;
|
||||
use crate::{Error, FrontendInvoker};
|
||||
|
||||
impl FlowWorkerManager {
|
||||
impl StreamingEngine {
|
||||
/// Create and start refill flow tasks in background
|
||||
pub async fn create_and_start_refill_flow_tasks(
|
||||
self: &FlowWorkerManagerRef,
|
||||
self: &FlowStreamingEngineRef,
|
||||
flow_metadata_manager: &FlowMetadataManagerRef,
|
||||
catalog_manager: &CatalogManagerRef,
|
||||
) -> Result<(), Error> {
|
||||
@@ -130,7 +130,7 @@ impl FlowWorkerManager {
|
||||
|
||||
/// Starting to refill flows, if any error occurs, will rebuild the flow and retry
|
||||
pub(crate) async fn starting_refill_flows(
|
||||
self: &FlowWorkerManagerRef,
|
||||
self: &FlowStreamingEngineRef,
|
||||
tasks: Vec<RefillTask>,
|
||||
) -> Result<(), Error> {
|
||||
// TODO(discord9): add a back pressure mechanism
|
||||
@@ -266,7 +266,7 @@ impl TaskState<()> {
|
||||
fn start_running(
|
||||
&mut self,
|
||||
task_data: &TaskData,
|
||||
manager: FlowWorkerManagerRef,
|
||||
manager: FlowStreamingEngineRef,
|
||||
mut output_stream: SendableRecordBatchStream,
|
||||
) -> Result<(), Error> {
|
||||
let data = (*task_data).clone();
|
||||
@@ -383,7 +383,7 @@ impl RefillTask {
|
||||
/// Start running the task in background, non-blocking
|
||||
pub async fn start_running(
|
||||
&mut self,
|
||||
manager: FlowWorkerManagerRef,
|
||||
manager: FlowStreamingEngineRef,
|
||||
invoker: &FrontendInvoker,
|
||||
) -> Result<(), Error> {
|
||||
let TaskState::Prepared { sql } = &mut self.state else {
|
||||
|
||||
@@ -16,9 +16,9 @@ use std::collections::BTreeMap;
|
||||
|
||||
use common_meta::key::flow::flow_state::FlowStat;
|
||||
|
||||
use crate::FlowWorkerManager;
|
||||
use crate::StreamingEngine;
|
||||
|
||||
impl FlowWorkerManager {
|
||||
impl StreamingEngine {
|
||||
pub async fn gen_state_report(&self) -> FlowStat {
|
||||
let mut full_report = BTreeMap::new();
|
||||
let mut last_exec_time_map = BTreeMap::new();
|
||||
|
||||
@@ -33,8 +33,8 @@ use crate::adapter::table_source::TableDesc;
|
||||
use crate::adapter::{TableName, WorkerHandle, AUTO_CREATED_PLACEHOLDER_TS_COL};
|
||||
use crate::error::{Error, ExternalSnafu, UnexpectedSnafu};
|
||||
use crate::repr::{ColumnType, RelationDesc, RelationType};
|
||||
use crate::FlowWorkerManager;
|
||||
impl FlowWorkerManager {
|
||||
use crate::StreamingEngine;
|
||||
impl StreamingEngine {
|
||||
/// Get a worker handle for creating flow, using round robin to select a worker
|
||||
pub(crate) async fn get_worker_handle_for_create_flow(&self) -> &WorkerHandle {
|
||||
let use_idx = {
|
||||
|
||||
@@ -32,3 +32,9 @@ pub const SLOW_QUERY_THRESHOLD: Duration = Duration::from_secs(60);
|
||||
|
||||
/// The minimum duration between two queries execution by batching mode task
|
||||
const MIN_REFRESH_DURATION: Duration = Duration::new(5, 0);
|
||||
|
||||
/// Grpc connection timeout
|
||||
const GRPC_CONN_TIMEOUT: Duration = Duration::from_secs(5);
|
||||
|
||||
/// Grpc max retry number
|
||||
const GRPC_MAX_RETRIES: u32 = 3;
|
||||
|
||||
@@ -17,14 +17,16 @@
|
||||
use std::collections::{BTreeMap, HashMap};
|
||||
use std::sync::Arc;
|
||||
|
||||
use catalog::CatalogManagerRef;
|
||||
use common_error::ext::BoxedError;
|
||||
use common_meta::ddl::create_flow::FlowType;
|
||||
use common_meta::key::flow::FlowMetadataManagerRef;
|
||||
use common_meta::key::table_info::TableInfoManager;
|
||||
use common_meta::key::table_info::{TableInfoManager, TableInfoValue};
|
||||
use common_meta::key::TableMetadataManagerRef;
|
||||
use common_runtime::JoinHandle;
|
||||
use common_telemetry::info;
|
||||
use common_telemetry::tracing::warn;
|
||||
use common_telemetry::{debug, info};
|
||||
use common_time::TimeToLive;
|
||||
use query::QueryEngineRef;
|
||||
use snafu::{ensure, OptionExt, ResultExt};
|
||||
use store_api::storage::RegionId;
|
||||
@@ -36,7 +38,9 @@ use crate::batching_mode::task::BatchingTask;
|
||||
use crate::batching_mode::time_window::{find_time_window_expr, TimeWindowExpr};
|
||||
use crate::batching_mode::utils::sql_to_df_plan;
|
||||
use crate::engine::FlowEngine;
|
||||
use crate::error::{ExternalSnafu, FlowAlreadyExistSnafu, TableNotFoundMetaSnafu, UnexpectedSnafu};
|
||||
use crate::error::{
|
||||
ExternalSnafu, FlowAlreadyExistSnafu, TableNotFoundMetaSnafu, UnexpectedSnafu, UnsupportedSnafu,
|
||||
};
|
||||
use crate::{CreateFlowArgs, Error, FlowId, TableName};
|
||||
|
||||
/// Batching mode Engine, responsible for driving all the batching mode tasks
|
||||
@@ -48,6 +52,7 @@ pub struct BatchingEngine {
|
||||
frontend_client: Arc<FrontendClient>,
|
||||
flow_metadata_manager: FlowMetadataManagerRef,
|
||||
table_meta: TableMetadataManagerRef,
|
||||
catalog_manager: CatalogManagerRef,
|
||||
query_engine: QueryEngineRef,
|
||||
}
|
||||
|
||||
@@ -57,6 +62,7 @@ impl BatchingEngine {
|
||||
query_engine: QueryEngineRef,
|
||||
flow_metadata_manager: FlowMetadataManagerRef,
|
||||
table_meta: TableMetadataManagerRef,
|
||||
catalog_manager: CatalogManagerRef,
|
||||
) -> Self {
|
||||
Self {
|
||||
tasks: Default::default(),
|
||||
@@ -64,6 +70,7 @@ impl BatchingEngine {
|
||||
frontend_client,
|
||||
flow_metadata_manager,
|
||||
table_meta,
|
||||
catalog_manager,
|
||||
query_engine,
|
||||
}
|
||||
}
|
||||
@@ -179,6 +186,16 @@ async fn get_table_name(
|
||||
table_info: &TableInfoManager,
|
||||
table_id: &TableId,
|
||||
) -> Result<TableName, Error> {
|
||||
get_table_info(table_info, table_id).await.map(|info| {
|
||||
let name = info.table_name();
|
||||
[name.catalog_name, name.schema_name, name.table_name]
|
||||
})
|
||||
}
|
||||
|
||||
async fn get_table_info(
|
||||
table_info: &TableInfoManager,
|
||||
table_id: &TableId,
|
||||
) -> Result<TableInfoValue, Error> {
|
||||
table_info
|
||||
.get(*table_id)
|
||||
.await
|
||||
@@ -187,8 +204,7 @@ async fn get_table_name(
|
||||
.with_context(|| UnexpectedSnafu {
|
||||
reason: format!("Table id = {:?}, couldn't found table name", table_id),
|
||||
})
|
||||
.map(|name| name.table_name())
|
||||
.map(|name| [name.catalog_name, name.schema_name, name.table_name])
|
||||
.map(|info| info.into_inner())
|
||||
}
|
||||
|
||||
impl BatchingEngine {
|
||||
@@ -248,7 +264,20 @@ impl BatchingEngine {
|
||||
let query_ctx = Arc::new(query_ctx);
|
||||
let mut source_table_names = Vec::with_capacity(2);
|
||||
for src_id in source_table_ids {
|
||||
// also check table option to see if ttl!=instant
|
||||
let table_name = get_table_name(self.table_meta.table_info_manager(), &src_id).await?;
|
||||
let table_info = get_table_info(self.table_meta.table_info_manager(), &src_id).await?;
|
||||
ensure!(
|
||||
table_info.table_info.meta.options.ttl != Some(TimeToLive::Instant),
|
||||
UnsupportedSnafu {
|
||||
reason: format!(
|
||||
"Source table `{}`(id={}) has instant TTL, Instant TTL is not supported under batching mode. Consider using a TTL longer than flush interval",
|
||||
table_name.join("."),
|
||||
src_id
|
||||
),
|
||||
}
|
||||
);
|
||||
|
||||
source_table_names.push(table_name);
|
||||
}
|
||||
|
||||
@@ -273,7 +302,14 @@ impl BatchingEngine {
|
||||
})
|
||||
.transpose()?;
|
||||
|
||||
info!("Flow id={}, found time window expr={:?}", flow_id, phy_expr);
|
||||
info!(
|
||||
"Flow id={}, found time window expr={}",
|
||||
flow_id,
|
||||
phy_expr
|
||||
.as_ref()
|
||||
.map(|phy_expr| phy_expr.to_string())
|
||||
.unwrap_or("None".to_string())
|
||||
);
|
||||
|
||||
let task = BatchingTask::new(
|
||||
flow_id,
|
||||
@@ -284,7 +320,7 @@ impl BatchingEngine {
|
||||
sink_table_name,
|
||||
source_table_names,
|
||||
query_ctx,
|
||||
self.table_meta.clone(),
|
||||
self.catalog_manager.clone(),
|
||||
rx,
|
||||
);
|
||||
|
||||
@@ -295,10 +331,11 @@ impl BatchingEngine {
|
||||
// check execute once first to detect any error early
|
||||
task.check_execute(&engine, &frontend).await?;
|
||||
|
||||
// TODO(discord9): also save handle & use time wheel or what for better
|
||||
let _handle = common_runtime::spawn_global(async move {
|
||||
// TODO(discord9): use time wheel or what for better
|
||||
let handle = common_runtime::spawn_global(async move {
|
||||
task_inner.start_executing_loop(engine, frontend).await;
|
||||
});
|
||||
task.state.write().unwrap().task_handle = Some(handle);
|
||||
|
||||
// only replace here not earlier because we want the old one intact if something went wrong before this line
|
||||
let replaced_old_task_opt = self.tasks.write().await.insert(flow_id, task);
|
||||
@@ -326,15 +363,23 @@ impl BatchingEngine {
|
||||
}
|
||||
|
||||
pub async fn flush_flow_inner(&self, flow_id: FlowId) -> Result<usize, Error> {
|
||||
debug!("Try flush flow {flow_id}");
|
||||
let task = self.tasks.read().await.get(&flow_id).cloned();
|
||||
let task = task.with_context(|| UnexpectedSnafu {
|
||||
reason: format!("Can't found task for flow {flow_id}"),
|
||||
})?;
|
||||
|
||||
task.mark_all_windows_as_dirty()?;
|
||||
|
||||
let res = task
|
||||
.gen_exec_once(&self.query_engine, &self.frontend_client)
|
||||
.await?;
|
||||
|
||||
let affected_rows = res.map(|(r, _)| r).unwrap_or_default() as usize;
|
||||
debug!(
|
||||
"Successfully flush flow {flow_id}, affected rows={}",
|
||||
affected_rows
|
||||
);
|
||||
Ok(affected_rows)
|
||||
}
|
||||
|
||||
@@ -357,6 +402,9 @@ impl FlowEngine for BatchingEngine {
|
||||
async fn flow_exist(&self, flow_id: FlowId) -> Result<bool, Error> {
|
||||
Ok(self.flow_exist_inner(flow_id).await)
|
||||
}
|
||||
async fn list_flows(&self) -> Result<impl IntoIterator<Item = FlowId>, Error> {
|
||||
Ok(self.tasks.read().await.keys().cloned().collect::<Vec<_>>())
|
||||
}
|
||||
async fn handle_flow_inserts(
|
||||
&self,
|
||||
request: api::v1::region::InsertRequests,
|
||||
|
||||
@@ -14,44 +14,109 @@
|
||||
|
||||
//! Frontend client to run flow as batching task which is time-window-aware normal query triggered every tick set by user
|
||||
|
||||
use std::sync::Arc;
|
||||
use std::sync::{Arc, Weak};
|
||||
|
||||
use client::{Client, Database, DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME};
|
||||
use common_error::ext::BoxedError;
|
||||
use api::v1::greptime_request::Request;
|
||||
use api::v1::CreateTableExpr;
|
||||
use client::{Client, Database};
|
||||
use common_error::ext::{BoxedError, ErrorExt};
|
||||
use common_grpc::channel_manager::{ChannelConfig, ChannelManager};
|
||||
use common_meta::cluster::{NodeInfo, NodeInfoKey, Role};
|
||||
use common_meta::peer::Peer;
|
||||
use common_meta::rpc::store::RangeRequest;
|
||||
use common_query::Output;
|
||||
use common_telemetry::warn;
|
||||
use meta_client::client::MetaClient;
|
||||
use snafu::ResultExt;
|
||||
use servers::query_handler::grpc::GrpcQueryHandler;
|
||||
use session::context::{QueryContextBuilder, QueryContextRef};
|
||||
use snafu::{OptionExt, ResultExt};
|
||||
|
||||
use crate::batching_mode::DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT;
|
||||
use crate::error::{ExternalSnafu, UnexpectedSnafu};
|
||||
use crate::batching_mode::{
|
||||
DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT, GRPC_CONN_TIMEOUT, GRPC_MAX_RETRIES,
|
||||
};
|
||||
use crate::error::{ExternalSnafu, InvalidRequestSnafu, UnexpectedSnafu};
|
||||
use crate::Error;
|
||||
|
||||
fn default_channel_mgr() -> ChannelManager {
|
||||
let cfg = ChannelConfig::new().timeout(DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT);
|
||||
ChannelManager::with_config(cfg)
|
||||
/// Just like [`GrpcQueryHandler`] but use BoxedError
|
||||
///
|
||||
/// basically just a specialized `GrpcQueryHandler<Error=BoxedError>`
|
||||
///
|
||||
/// this is only useful for flownode to
|
||||
/// invoke frontend Instance in standalone mode
|
||||
#[async_trait::async_trait]
|
||||
pub trait GrpcQueryHandlerWithBoxedError: Send + Sync + 'static {
|
||||
async fn do_query(
|
||||
&self,
|
||||
query: Request,
|
||||
ctx: QueryContextRef,
|
||||
) -> std::result::Result<Output, BoxedError>;
|
||||
}
|
||||
|
||||
fn client_from_urls(addrs: Vec<String>) -> Client {
|
||||
Client::with_manager_and_urls(default_channel_mgr(), addrs)
|
||||
/// auto impl
|
||||
#[async_trait::async_trait]
|
||||
impl<
|
||||
E: ErrorExt + Send + Sync + 'static,
|
||||
T: GrpcQueryHandler<Error = E> + Send + Sync + 'static,
|
||||
> GrpcQueryHandlerWithBoxedError for T
|
||||
{
|
||||
async fn do_query(
|
||||
&self,
|
||||
query: Request,
|
||||
ctx: QueryContextRef,
|
||||
) -> std::result::Result<Output, BoxedError> {
|
||||
self.do_query(query, ctx).await.map_err(BoxedError::new)
|
||||
}
|
||||
}
|
||||
|
||||
type HandlerMutable = Arc<std::sync::Mutex<Option<Weak<dyn GrpcQueryHandlerWithBoxedError>>>>;
|
||||
|
||||
/// A simple frontend client able to execute sql using grpc protocol
|
||||
#[derive(Debug)]
|
||||
///
|
||||
/// This is for computation-heavy query which need to offload computation to frontend, lifting the load from flownode
|
||||
#[derive(Debug, Clone)]
|
||||
pub enum FrontendClient {
|
||||
Distributed {
|
||||
meta_client: Arc<MetaClient>,
|
||||
chnl_mgr: ChannelManager,
|
||||
},
|
||||
Standalone {
|
||||
/// for the sake of simplicity still use grpc even in standalone mode
|
||||
/// notice the client here should all be lazy, so that can wait after frontend is booted then make conn
|
||||
/// TODO(discord9): not use grpc under standalone mode
|
||||
database_client: DatabaseWithPeer,
|
||||
database_client: HandlerMutable,
|
||||
},
|
||||
}
|
||||
|
||||
impl FrontendClient {
|
||||
/// Create a new empty frontend client, with a `HandlerMutable` to set the grpc handler later
|
||||
pub fn from_empty_grpc_handler() -> (Self, HandlerMutable) {
|
||||
let handler = Arc::new(std::sync::Mutex::new(None));
|
||||
(
|
||||
Self::Standalone {
|
||||
database_client: handler.clone(),
|
||||
},
|
||||
handler,
|
||||
)
|
||||
}
|
||||
|
||||
pub fn from_meta_client(meta_client: Arc<MetaClient>) -> Self {
|
||||
Self::Distributed {
|
||||
meta_client,
|
||||
chnl_mgr: {
|
||||
let cfg = ChannelConfig::new()
|
||||
.connect_timeout(GRPC_CONN_TIMEOUT)
|
||||
.timeout(DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT);
|
||||
ChannelManager::with_config(cfg)
|
||||
},
|
||||
}
|
||||
}
|
||||
|
||||
pub fn from_grpc_handler(grpc_handler: Weak<dyn GrpcQueryHandlerWithBoxedError>) -> Self {
|
||||
Self::Standalone {
|
||||
database_client: Arc::new(std::sync::Mutex::new(Some(grpc_handler))),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[derive(Debug, Clone)]
|
||||
pub struct DatabaseWithPeer {
|
||||
pub database: Database,
|
||||
@@ -64,25 +129,6 @@ impl DatabaseWithPeer {
|
||||
}
|
||||
}
|
||||
|
||||
impl FrontendClient {
|
||||
pub fn from_meta_client(meta_client: Arc<MetaClient>) -> Self {
|
||||
Self::Distributed { meta_client }
|
||||
}
|
||||
|
||||
pub fn from_static_grpc_addr(addr: String) -> Self {
|
||||
let peer = Peer {
|
||||
id: 0,
|
||||
addr: addr.clone(),
|
||||
};
|
||||
|
||||
let client = client_from_urls(vec![addr]);
|
||||
let database = Database::new(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, client);
|
||||
Self::Standalone {
|
||||
database_client: DatabaseWithPeer::new(database, peer),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
impl FrontendClient {
|
||||
async fn scan_for_frontend(&self) -> Result<Vec<(NodeInfoKey, NodeInfo)>, Error> {
|
||||
let Self::Distributed { meta_client, .. } = self else {
|
||||
@@ -115,10 +161,21 @@ impl FrontendClient {
|
||||
}
|
||||
|
||||
/// Get the database with max `last_activity_ts`
|
||||
async fn get_last_active_frontend(&self) -> Result<DatabaseWithPeer, Error> {
|
||||
if let Self::Standalone { database_client } = self {
|
||||
return Ok(database_client.clone());
|
||||
}
|
||||
async fn get_last_active_frontend(
|
||||
&self,
|
||||
catalog: &str,
|
||||
schema: &str,
|
||||
) -> Result<DatabaseWithPeer, Error> {
|
||||
let Self::Distributed {
|
||||
meta_client: _,
|
||||
chnl_mgr,
|
||||
} = self
|
||||
else {
|
||||
return UnexpectedSnafu {
|
||||
reason: "Expect distributed mode",
|
||||
}
|
||||
.fail();
|
||||
};
|
||||
|
||||
let frontends = self.scan_for_frontend().await?;
|
||||
let mut peer = None;
|
||||
@@ -133,16 +190,139 @@ impl FrontendClient {
|
||||
}
|
||||
.fail()?
|
||||
};
|
||||
let client = client_from_urls(vec![peer.addr.clone()]);
|
||||
let database = Database::new(DEFAULT_CATALOG_NAME, DEFAULT_SCHEMA_NAME, client);
|
||||
let client = Client::with_manager_and_urls(chnl_mgr.clone(), vec![peer.addr.clone()]);
|
||||
let database = Database::new(catalog, schema, client);
|
||||
Ok(DatabaseWithPeer::new(database, peer))
|
||||
}
|
||||
|
||||
/// Get a database client, and possibly update it before returning.
|
||||
pub async fn get_database_client(&self) -> Result<DatabaseWithPeer, Error> {
|
||||
pub async fn create(
|
||||
&self,
|
||||
create: CreateTableExpr,
|
||||
catalog: &str,
|
||||
schema: &str,
|
||||
) -> Result<u32, Error> {
|
||||
self.handle(
|
||||
Request::Ddl(api::v1::DdlRequest {
|
||||
expr: Some(api::v1::ddl_request::Expr::CreateTable(create)),
|
||||
}),
|
||||
catalog,
|
||||
schema,
|
||||
&mut None,
|
||||
)
|
||||
.await
|
||||
}
|
||||
|
||||
/// Handle a request to frontend
|
||||
pub(crate) async fn handle(
|
||||
&self,
|
||||
req: api::v1::greptime_request::Request,
|
||||
catalog: &str,
|
||||
schema: &str,
|
||||
peer_desc: &mut Option<PeerDesc>,
|
||||
) -> Result<u32, Error> {
|
||||
match self {
|
||||
Self::Standalone { database_client } => Ok(database_client.clone()),
|
||||
Self::Distributed { meta_client: _ } => self.get_last_active_frontend().await,
|
||||
FrontendClient::Distributed { .. } => {
|
||||
let db = self.get_last_active_frontend(catalog, schema).await?;
|
||||
|
||||
*peer_desc = Some(PeerDesc::Dist {
|
||||
peer: db.peer.clone(),
|
||||
});
|
||||
|
||||
let mut retry = 0;
|
||||
|
||||
loop {
|
||||
let ret = db.database.handle(req.clone()).await.with_context(|_| {
|
||||
InvalidRequestSnafu {
|
||||
context: format!("Failed to handle request: {:?}", req),
|
||||
}
|
||||
});
|
||||
if let Err(err) = ret {
|
||||
if retry < GRPC_MAX_RETRIES {
|
||||
retry += 1;
|
||||
warn!(
|
||||
"Failed to send request to grpc handle at Peer={:?}, retry = {}, error = {:?}",
|
||||
db.peer, retry, err
|
||||
);
|
||||
continue;
|
||||
} else {
|
||||
common_telemetry::error!(
|
||||
"Failed to send request to grpc handle at Peer={:?} after {} retries, error = {:?}",
|
||||
db.peer, retry, err
|
||||
);
|
||||
return Err(err);
|
||||
}
|
||||
}
|
||||
return ret;
|
||||
}
|
||||
}
|
||||
FrontendClient::Standalone { database_client } => {
|
||||
let ctx = QueryContextBuilder::default()
|
||||
.current_catalog(catalog.to_string())
|
||||
.current_schema(schema.to_string())
|
||||
.build();
|
||||
let ctx = Arc::new(ctx);
|
||||
{
|
||||
let database_client = {
|
||||
database_client
|
||||
.lock()
|
||||
.map_err(|e| {
|
||||
UnexpectedSnafu {
|
||||
reason: format!("Failed to lock database client: {e}"),
|
||||
}
|
||||
.build()
|
||||
})?
|
||||
.as_ref()
|
||||
.context(UnexpectedSnafu {
|
||||
reason: "Standalone's frontend instance is not set",
|
||||
})?
|
||||
.upgrade()
|
||||
.context(UnexpectedSnafu {
|
||||
reason: "Failed to upgrade database client",
|
||||
})?
|
||||
};
|
||||
let resp: common_query::Output = database_client
|
||||
.do_query(req.clone(), ctx)
|
||||
.await
|
||||
.map_err(BoxedError::new)
|
||||
.context(ExternalSnafu)?;
|
||||
match resp.data {
|
||||
common_query::OutputData::AffectedRows(rows) => {
|
||||
Ok(rows.try_into().map_err(|_| {
|
||||
UnexpectedSnafu {
|
||||
reason: format!("Failed to convert rows to u32: {}", rows),
|
||||
}
|
||||
.build()
|
||||
})?)
|
||||
}
|
||||
_ => UnexpectedSnafu {
|
||||
reason: "Unexpected output data",
|
||||
}
|
||||
.fail(),
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// Describe a peer of frontend
|
||||
#[derive(Debug, Default)]
|
||||
pub(crate) enum PeerDesc {
|
||||
/// Distributed mode's frontend peer address
|
||||
Dist {
|
||||
/// frontend peer address
|
||||
peer: Peer,
|
||||
},
|
||||
/// Standalone mode
|
||||
#[default]
|
||||
Standalone,
|
||||
}
|
||||
|
||||
impl std::fmt::Display for PeerDesc {
|
||||
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
|
||||
match self {
|
||||
PeerDesc::Dist { peer } => write!(f, "{}", peer.addr),
|
||||
PeerDesc::Standalone => write!(f, "standalone"),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -22,13 +22,14 @@ use common_telemetry::tracing::warn;
|
||||
use common_time::Timestamp;
|
||||
use datatypes::value::Value;
|
||||
use session::context::QueryContextRef;
|
||||
use snafu::ResultExt;
|
||||
use snafu::{OptionExt, ResultExt};
|
||||
use tokio::sync::oneshot;
|
||||
use tokio::time::Instant;
|
||||
|
||||
use crate::batching_mode::task::BatchingTask;
|
||||
use crate::batching_mode::time_window::TimeWindowExpr;
|
||||
use crate::batching_mode::MIN_REFRESH_DURATION;
|
||||
use crate::error::{DatatypesSnafu, InternalSnafu, TimeSnafu};
|
||||
use crate::error::{DatatypesSnafu, InternalSnafu, TimeSnafu, UnexpectedSnafu};
|
||||
use crate::{Error, FlowId};
|
||||
|
||||
/// The state of the [`BatchingTask`].
|
||||
@@ -46,6 +47,8 @@ pub struct TaskState {
|
||||
exec_state: ExecState,
|
||||
/// Shutdown receiver
|
||||
pub(crate) shutdown_rx: oneshot::Receiver<()>,
|
||||
/// Task handle
|
||||
pub(crate) task_handle: Option<tokio::task::JoinHandle<()>>,
|
||||
}
|
||||
impl TaskState {
|
||||
pub fn new(query_ctx: QueryContextRef, shutdown_rx: oneshot::Receiver<()>) -> Self {
|
||||
@@ -56,6 +59,7 @@ impl TaskState {
|
||||
dirty_time_windows: Default::default(),
|
||||
exec_state: ExecState::Idle,
|
||||
shutdown_rx,
|
||||
task_handle: None,
|
||||
}
|
||||
}
|
||||
|
||||
@@ -70,7 +74,11 @@ impl TaskState {
|
||||
/// wait for at least `last_query_duration`, at most `max_timeout` to start next query
|
||||
///
|
||||
/// if have more dirty time window, exec next query immediately
|
||||
pub fn get_next_start_query_time(&self, max_timeout: Option<Duration>) -> Instant {
|
||||
pub fn get_next_start_query_time(
|
||||
&self,
|
||||
flow_id: FlowId,
|
||||
max_timeout: Option<Duration>,
|
||||
) -> Instant {
|
||||
let next_duration = max_timeout
|
||||
.unwrap_or(self.last_query_duration)
|
||||
.min(self.last_query_duration);
|
||||
@@ -80,6 +88,12 @@ impl TaskState {
|
||||
if self.dirty_time_windows.windows.is_empty() {
|
||||
self.last_update_time + next_duration
|
||||
} else {
|
||||
debug!(
|
||||
"Flow id = {}, still have {} dirty time window({:?}), execute immediately",
|
||||
flow_id,
|
||||
self.dirty_time_windows.windows.len(),
|
||||
self.dirty_time_windows.windows
|
||||
);
|
||||
Instant::now()
|
||||
}
|
||||
}
|
||||
@@ -115,6 +129,15 @@ impl DirtyTimeWindows {
|
||||
}
|
||||
}
|
||||
|
||||
pub fn add_window(&mut self, start: Timestamp, end: Option<Timestamp>) {
|
||||
self.windows.insert(start, end);
|
||||
}
|
||||
|
||||
/// Clean all dirty time windows, useful when can't found time window expr
|
||||
pub fn clean(&mut self) {
|
||||
self.windows.clear();
|
||||
}
|
||||
|
||||
/// Generate all filter expressions consuming all time windows
|
||||
pub fn gen_filter_exprs(
|
||||
&mut self,
|
||||
@@ -177,6 +200,18 @@ impl DirtyTimeWindows {
|
||||
|
||||
let mut expr_lst = vec![];
|
||||
for (start, end) in first_nth.into_iter() {
|
||||
// align using time window exprs
|
||||
let (start, end) = if let Some(ctx) = task_ctx {
|
||||
let Some(time_window_expr) = &ctx.config.time_window_expr else {
|
||||
UnexpectedSnafu {
|
||||
reason: "time_window_expr is not set",
|
||||
}
|
||||
.fail()?
|
||||
};
|
||||
self.align_time_window(start, end, time_window_expr)?
|
||||
} else {
|
||||
(start, end)
|
||||
};
|
||||
debug!(
|
||||
"Time window start: {:?}, end: {:?}",
|
||||
start.to_iso8601_string(),
|
||||
@@ -199,6 +234,30 @@ impl DirtyTimeWindows {
|
||||
Ok(expr)
|
||||
}
|
||||
|
||||
fn align_time_window(
|
||||
&self,
|
||||
start: Timestamp,
|
||||
end: Option<Timestamp>,
|
||||
time_window_expr: &TimeWindowExpr,
|
||||
) -> Result<(Timestamp, Option<Timestamp>), Error> {
|
||||
let align_start = time_window_expr.eval(start)?.0.context(UnexpectedSnafu {
|
||||
reason: format!(
|
||||
"Failed to align start time {:?} with time window expr {:?}",
|
||||
start, time_window_expr
|
||||
),
|
||||
})?;
|
||||
let align_end = end
|
||||
.and_then(|end| {
|
||||
time_window_expr
|
||||
.eval(end)
|
||||
// if after aligned, end is the same, then use end(because it's already aligned) else use aligned end
|
||||
.map(|r| if r.0 == Some(end) { r.0 } else { r.1 })
|
||||
.transpose()
|
||||
})
|
||||
.transpose()?;
|
||||
Ok((align_start, align_end))
|
||||
}
|
||||
|
||||
/// Merge time windows that overlaps or get too close
|
||||
pub fn merge_dirty_time_windows(
|
||||
&mut self,
|
||||
@@ -287,8 +346,12 @@ enum ExecState {
|
||||
#[cfg(test)]
|
||||
mod test {
|
||||
use pretty_assertions::assert_eq;
|
||||
use session::context::QueryContext;
|
||||
|
||||
use super::*;
|
||||
use crate::batching_mode::time_window::find_time_window_expr;
|
||||
use crate::batching_mode::utils::sql_to_df_plan;
|
||||
use crate::test_utils::create_test_query_engine;
|
||||
|
||||
#[test]
|
||||
fn test_merge_dirty_time_windows() {
|
||||
@@ -404,4 +467,59 @@ mod test {
|
||||
assert_eq!(expected_filter_expr, to_sql.as_deref());
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_align_time_window() {
|
||||
type TimeWindow = (Timestamp, Option<Timestamp>);
|
||||
struct TestCase {
|
||||
sql: String,
|
||||
aligns: Vec<(TimeWindow, TimeWindow)>,
|
||||
}
|
||||
let testcases: Vec<TestCase> = vec![TestCase{
|
||||
sql: "SELECT date_bin(INTERVAL '5 second', ts) AS time_window FROM numbers_with_ts GROUP BY time_window;".to_string(),
|
||||
aligns: vec![
|
||||
((Timestamp::new_second(3), None), (Timestamp::new_second(0), None)),
|
||||
((Timestamp::new_second(8), None), (Timestamp::new_second(5), None)),
|
||||
((Timestamp::new_second(8), Some(Timestamp::new_second(10))), (Timestamp::new_second(5), Some(Timestamp::new_second(10)))),
|
||||
((Timestamp::new_second(8), Some(Timestamp::new_second(9))), (Timestamp::new_second(5), Some(Timestamp::new_second(10)))),
|
||||
],
|
||||
}];
|
||||
|
||||
let query_engine = create_test_query_engine();
|
||||
let ctx = QueryContext::arc();
|
||||
for TestCase { sql, aligns } in testcases {
|
||||
let plan = sql_to_df_plan(ctx.clone(), query_engine.clone(), &sql, true)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let (column_name, time_window_expr, _, df_schema) = find_time_window_expr(
|
||||
&plan,
|
||||
query_engine.engine_state().catalog_manager().clone(),
|
||||
ctx.clone(),
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let time_window_expr = time_window_expr
|
||||
.map(|expr| {
|
||||
TimeWindowExpr::from_expr(
|
||||
&expr,
|
||||
&column_name,
|
||||
&df_schema,
|
||||
&query_engine.engine_state().session_state(),
|
||||
)
|
||||
})
|
||||
.transpose()
|
||||
.unwrap()
|
||||
.unwrap();
|
||||
|
||||
let dirty = DirtyTimeWindows::default();
|
||||
for (before_align, expected_after_align) in aligns {
|
||||
let after_align = dirty
|
||||
.align_time_window(before_align.0, before_align.1, &time_window_expr)
|
||||
.unwrap();
|
||||
assert_eq!(expected_after_align, after_align);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -12,33 +12,32 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::collections::HashSet;
|
||||
use std::collections::{BTreeSet, HashSet};
|
||||
use std::ops::Deref;
|
||||
use std::sync::{Arc, RwLock};
|
||||
use std::time::{Duration, SystemTime, UNIX_EPOCH};
|
||||
|
||||
use api::v1::CreateTableExpr;
|
||||
use arrow_schema::Fields;
|
||||
use catalog::CatalogManagerRef;
|
||||
use common_error::ext::BoxedError;
|
||||
use common_meta::key::table_name::TableNameKey;
|
||||
use common_meta::key::TableMetadataManagerRef;
|
||||
use common_query::logical_plan::breakup_insert_plan;
|
||||
use common_telemetry::tracing::warn;
|
||||
use common_telemetry::{debug, info};
|
||||
use common_time::Timestamp;
|
||||
use datafusion::optimizer::analyzer::count_wildcard_rule::CountWildcardRule;
|
||||
use datafusion::optimizer::AnalyzerRule;
|
||||
use datafusion::sql::unparser::expr_to_sql;
|
||||
use datafusion_common::tree_node::TreeNode;
|
||||
use datafusion_common::tree_node::{Transformed, TreeNode};
|
||||
use datafusion_expr::{DmlStatement, LogicalPlan, WriteOp};
|
||||
use datatypes::prelude::ConcreteDataType;
|
||||
use datatypes::schema::constraint::NOW_FN;
|
||||
use datatypes::schema::{ColumnDefaultConstraint, ColumnSchema};
|
||||
use datatypes::value::Value;
|
||||
use datatypes::schema::{ColumnSchema, Schema};
|
||||
use operator::expr_helper::column_schemas_to_defs;
|
||||
use query::query_engine::DefaultSerializer;
|
||||
use query::QueryEngineRef;
|
||||
use session::context::QueryContextRef;
|
||||
use snafu::{OptionExt, ResultExt};
|
||||
use snafu::{ensure, OptionExt, ResultExt};
|
||||
use substrait::{DFLogicalSubstraitConvertor, SubstraitPlan};
|
||||
use table::metadata::RawTableMeta;
|
||||
use tokio::sync::oneshot;
|
||||
use tokio::sync::oneshot::error::TryRecvError;
|
||||
use tokio::time::Instant;
|
||||
@@ -48,14 +47,16 @@ use crate::batching_mode::frontend_client::FrontendClient;
|
||||
use crate::batching_mode::state::TaskState;
|
||||
use crate::batching_mode::time_window::TimeWindowExpr;
|
||||
use crate::batching_mode::utils::{
|
||||
sql_to_df_plan, AddAutoColumnRewriter, AddFilterRewriter, FindGroupByFinalName,
|
||||
get_table_info_df_schema, sql_to_df_plan, AddAutoColumnRewriter, AddFilterRewriter,
|
||||
FindGroupByFinalName,
|
||||
};
|
||||
use crate::batching_mode::{
|
||||
DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT, MIN_REFRESH_DURATION, SLOW_QUERY_THRESHOLD,
|
||||
};
|
||||
use crate::df_optimizer::apply_df_optimizer;
|
||||
use crate::error::{
|
||||
ConvertColumnSchemaSnafu, DatafusionSnafu, DatatypesSnafu, ExternalSnafu, InvalidRequestSnafu,
|
||||
SubstraitEncodeLogicalPlanSnafu, TableNotFoundMetaSnafu, TableNotFoundSnafu, UnexpectedSnafu,
|
||||
ConvertColumnSchemaSnafu, DatafusionSnafu, ExternalSnafu, InvalidQuerySnafu,
|
||||
SubstraitEncodeLogicalPlanSnafu, UnexpectedSnafu,
|
||||
};
|
||||
use crate::metrics::{
|
||||
METRIC_FLOW_BATCHING_ENGINE_QUERY_TIME, METRIC_FLOW_BATCHING_ENGINE_SLOW_QUERY,
|
||||
@@ -73,7 +74,7 @@ pub struct TaskConfig {
|
||||
pub expire_after: Option<i64>,
|
||||
sink_table_name: [String; 3],
|
||||
pub source_table_names: HashSet<[String; 3]>,
|
||||
table_meta: TableMetadataManagerRef,
|
||||
catalog_manager: CatalogManagerRef,
|
||||
}
|
||||
|
||||
#[derive(Clone)]
|
||||
@@ -93,7 +94,7 @@ impl BatchingTask {
|
||||
sink_table_name: [String; 3],
|
||||
source_table_names: Vec<[String; 3]>,
|
||||
query_ctx: QueryContextRef,
|
||||
table_meta: TableMetadataManagerRef,
|
||||
catalog_manager: CatalogManagerRef,
|
||||
shutdown_rx: oneshot::Receiver<()>,
|
||||
) -> Self {
|
||||
Self {
|
||||
@@ -105,12 +106,42 @@ impl BatchingTask {
|
||||
expire_after,
|
||||
sink_table_name,
|
||||
source_table_names: source_table_names.into_iter().collect(),
|
||||
table_meta,
|
||||
catalog_manager,
|
||||
}),
|
||||
state: Arc::new(RwLock::new(TaskState::new(query_ctx, shutdown_rx))),
|
||||
}
|
||||
}
|
||||
|
||||
/// mark time window range (now - expire_after, now) as dirty (or (0, now) if expire_after not set)
|
||||
///
|
||||
/// useful for flush_flow to flush dirty time windows range
|
||||
pub fn mark_all_windows_as_dirty(&self) -> Result<(), Error> {
|
||||
let now = SystemTime::now();
|
||||
let now = Timestamp::new_second(
|
||||
now.duration_since(UNIX_EPOCH)
|
||||
.expect("Time went backwards")
|
||||
.as_secs() as _,
|
||||
);
|
||||
let lower_bound = self
|
||||
.config
|
||||
.expire_after
|
||||
.map(|e| now.sub_duration(Duration::from_secs(e as _)))
|
||||
.transpose()
|
||||
.map_err(BoxedError::new)
|
||||
.context(ExternalSnafu)?
|
||||
.unwrap_or(Timestamp::new_second(0));
|
||||
debug!(
|
||||
"Flow {} mark range ({:?}, {:?}) as dirty",
|
||||
self.config.flow_id, lower_bound, now
|
||||
);
|
||||
self.state
|
||||
.write()
|
||||
.unwrap()
|
||||
.dirty_time_windows
|
||||
.add_window(lower_bound, Some(now));
|
||||
Ok(())
|
||||
}
|
||||
|
||||
/// Test execute, for check syntax or such
|
||||
pub async fn check_execute(
|
||||
&self,
|
||||
@@ -148,13 +179,8 @@ impl BatchingTask {
|
||||
|
||||
async fn is_table_exist(&self, table_name: &[String; 3]) -> Result<bool, Error> {
|
||||
self.config
|
||||
.table_meta
|
||||
.table_name_manager()
|
||||
.exists(TableNameKey {
|
||||
catalog: &table_name[0],
|
||||
schema: &table_name[1],
|
||||
table: &table_name[2],
|
||||
})
|
||||
.catalog_manager
|
||||
.table_exists(&table_name[0], &table_name[1], &table_name[2], None)
|
||||
.await
|
||||
.map_err(BoxedError::new)
|
||||
.context(ExternalSnafu)
|
||||
@@ -166,8 +192,10 @@ impl BatchingTask {
|
||||
frontend_client: &Arc<FrontendClient>,
|
||||
) -> Result<Option<(u32, Duration)>, Error> {
|
||||
if let Some(new_query) = self.gen_insert_plan(engine).await? {
|
||||
debug!("Generate new query: {:#?}", new_query);
|
||||
self.execute_logical_plan(frontend_client, &new_query).await
|
||||
} else {
|
||||
debug!("Generate no query");
|
||||
Ok(None)
|
||||
}
|
||||
}
|
||||
@@ -176,67 +204,35 @@ impl BatchingTask {
|
||||
&self,
|
||||
engine: &QueryEngineRef,
|
||||
) -> Result<Option<LogicalPlan>, Error> {
|
||||
let full_table_name = self.config.sink_table_name.clone().join(".");
|
||||
|
||||
let table_id = self
|
||||
.config
|
||||
.table_meta
|
||||
.table_name_manager()
|
||||
.get(common_meta::key::table_name::TableNameKey::new(
|
||||
&self.config.sink_table_name[0],
|
||||
&self.config.sink_table_name[1],
|
||||
&self.config.sink_table_name[2],
|
||||
))
|
||||
.await
|
||||
.with_context(|_| TableNotFoundMetaSnafu {
|
||||
msg: full_table_name.clone(),
|
||||
})?
|
||||
.map(|t| t.table_id())
|
||||
.with_context(|| TableNotFoundSnafu {
|
||||
name: full_table_name.clone(),
|
||||
})?;
|
||||
|
||||
let table = self
|
||||
.config
|
||||
.table_meta
|
||||
.table_info_manager()
|
||||
.get(table_id)
|
||||
.await
|
||||
.with_context(|_| TableNotFoundMetaSnafu {
|
||||
msg: full_table_name.clone(),
|
||||
})?
|
||||
.with_context(|| TableNotFoundSnafu {
|
||||
name: full_table_name.clone(),
|
||||
})?
|
||||
.into_inner();
|
||||
|
||||
let schema: datatypes::schema::Schema = table
|
||||
.table_info
|
||||
.meta
|
||||
.schema
|
||||
.clone()
|
||||
.try_into()
|
||||
.with_context(|_| DatatypesSnafu {
|
||||
extra: format!(
|
||||
"Failed to convert schema from raw schema, raw_schema={:?}",
|
||||
table.table_info.meta.schema
|
||||
),
|
||||
})?;
|
||||
|
||||
let df_schema = Arc::new(schema.arrow_schema().clone().try_into().with_context(|_| {
|
||||
DatafusionSnafu {
|
||||
context: format!(
|
||||
"Failed to convert arrow schema to datafusion schema, arrow_schema={:?}",
|
||||
schema.arrow_schema()
|
||||
),
|
||||
}
|
||||
})?);
|
||||
let (table, df_schema) = get_table_info_df_schema(
|
||||
self.config.catalog_manager.clone(),
|
||||
self.config.sink_table_name.clone(),
|
||||
)
|
||||
.await?;
|
||||
|
||||
let new_query = self
|
||||
.gen_query_with_time_window(engine.clone(), &table.table_info.meta)
|
||||
.gen_query_with_time_window(engine.clone(), &table.meta.schema)
|
||||
.await?;
|
||||
|
||||
let insert_into = if let Some((new_query, _column_cnt)) = new_query {
|
||||
// first check if all columns in input query exists in sink table
|
||||
// since insert into ref to names in record batch generate by given query
|
||||
let table_columns = df_schema
|
||||
.columns()
|
||||
.into_iter()
|
||||
.map(|c| c.name)
|
||||
.collect::<BTreeSet<_>>();
|
||||
for column in new_query.schema().columns() {
|
||||
ensure!(
|
||||
table_columns.contains(column.name()),
|
||||
InvalidQuerySnafu {
|
||||
reason: format!(
|
||||
"Column {} not found in sink table with columns {:?}",
|
||||
column, table_columns
|
||||
),
|
||||
}
|
||||
);
|
||||
}
|
||||
// update_at& time index placeholder (if exists) should have default value
|
||||
LogicalPlan::Dml(DmlStatement::new(
|
||||
datafusion_common::TableReference::Full {
|
||||
@@ -251,6 +247,9 @@ impl BatchingTask {
|
||||
} else {
|
||||
return Ok(None);
|
||||
};
|
||||
let insert_into = insert_into.recompute_schema().context(DatafusionSnafu {
|
||||
context: "Failed to recompute schema",
|
||||
})?;
|
||||
Ok(Some(insert_into))
|
||||
}
|
||||
|
||||
@@ -259,14 +258,11 @@ impl BatchingTask {
|
||||
frontend_client: &Arc<FrontendClient>,
|
||||
expr: CreateTableExpr,
|
||||
) -> Result<(), Error> {
|
||||
let db_client = frontend_client.get_database_client().await?;
|
||||
db_client
|
||||
.database
|
||||
.create(expr.clone())
|
||||
.await
|
||||
.with_context(|_| InvalidRequestSnafu {
|
||||
context: format!("Failed to create table with expr: {:?}", expr),
|
||||
})?;
|
||||
let catalog = &self.config.sink_table_name[0];
|
||||
let schema = &self.config.sink_table_name[1];
|
||||
frontend_client
|
||||
.create(expr.clone(), catalog, schema)
|
||||
.await?;
|
||||
Ok(())
|
||||
}
|
||||
|
||||
@@ -277,27 +273,78 @@ impl BatchingTask {
|
||||
) -> Result<Option<(u32, Duration)>, Error> {
|
||||
let instant = Instant::now();
|
||||
let flow_id = self.config.flow_id;
|
||||
let db_client = frontend_client.get_database_client().await?;
|
||||
let peer_addr = db_client.peer.addr;
|
||||
|
||||
debug!(
|
||||
"Executing flow {flow_id}(expire_after={:?} secs) on {:?} with query {}",
|
||||
self.config.expire_after, peer_addr, &plan
|
||||
"Executing flow {flow_id}(expire_after={:?} secs) with query {}",
|
||||
self.config.expire_after, &plan
|
||||
);
|
||||
|
||||
let timer = METRIC_FLOW_BATCHING_ENGINE_QUERY_TIME
|
||||
.with_label_values(&[flow_id.to_string().as_str()])
|
||||
.start_timer();
|
||||
let catalog = &self.config.sink_table_name[0];
|
||||
let schema = &self.config.sink_table_name[1];
|
||||
|
||||
let message = DFLogicalSubstraitConvertor {}
|
||||
.encode(plan, DefaultSerializer)
|
||||
.context(SubstraitEncodeLogicalPlanSnafu)?;
|
||||
// fix all table ref by make it fully qualified, i.e. "table_name" => "catalog_name.schema_name.table_name"
|
||||
let fixed_plan = plan
|
||||
.clone()
|
||||
.transform_down_with_subqueries(|p| {
|
||||
if let LogicalPlan::TableScan(mut table_scan) = p {
|
||||
let resolved = table_scan.table_name.resolve(catalog, schema);
|
||||
table_scan.table_name = resolved.into();
|
||||
Ok(Transformed::yes(LogicalPlan::TableScan(table_scan)))
|
||||
} else {
|
||||
Ok(Transformed::no(p))
|
||||
}
|
||||
})
|
||||
.with_context(|_| DatafusionSnafu {
|
||||
context: format!("Failed to fix table ref in logical plan, plan={:?}", plan),
|
||||
})?
|
||||
.data;
|
||||
|
||||
let req = api::v1::greptime_request::Request::Query(api::v1::QueryRequest {
|
||||
query: Some(api::v1::query_request::Query::LogicalPlan(message.to_vec())),
|
||||
});
|
||||
let expanded_plan = CountWildcardRule::new()
|
||||
.analyze(fixed_plan.clone(), &Default::default())
|
||||
.with_context(|_| DatafusionSnafu {
|
||||
context: format!(
|
||||
"Failed to expand wildcard in logical plan, plan={:?}",
|
||||
fixed_plan
|
||||
),
|
||||
})?;
|
||||
|
||||
let res = db_client.database.handle(req).await;
|
||||
drop(timer);
|
||||
let plan = expanded_plan;
|
||||
let mut peer_desc = None;
|
||||
|
||||
let res = {
|
||||
let _timer = METRIC_FLOW_BATCHING_ENGINE_QUERY_TIME
|
||||
.with_label_values(&[flow_id.to_string().as_str()])
|
||||
.start_timer();
|
||||
|
||||
// hack and special handling the insert logical plan
|
||||
let req = if let Some((insert_to, insert_plan)) =
|
||||
breakup_insert_plan(&plan, catalog, schema)
|
||||
{
|
||||
let message = DFLogicalSubstraitConvertor {}
|
||||
.encode(&insert_plan, DefaultSerializer)
|
||||
.context(SubstraitEncodeLogicalPlanSnafu)?;
|
||||
api::v1::greptime_request::Request::Query(api::v1::QueryRequest {
|
||||
query: Some(api::v1::query_request::Query::InsertIntoPlan(
|
||||
api::v1::InsertIntoPlan {
|
||||
table_name: Some(insert_to),
|
||||
logical_plan: message.to_vec(),
|
||||
},
|
||||
)),
|
||||
})
|
||||
} else {
|
||||
let message = DFLogicalSubstraitConvertor {}
|
||||
.encode(&plan, DefaultSerializer)
|
||||
.context(SubstraitEncodeLogicalPlanSnafu)?;
|
||||
|
||||
api::v1::greptime_request::Request::Query(api::v1::QueryRequest {
|
||||
query: Some(api::v1::query_request::Query::LogicalPlan(message.to_vec())),
|
||||
})
|
||||
};
|
||||
|
||||
frontend_client
|
||||
.handle(req, catalog, schema, &mut peer_desc)
|
||||
.await
|
||||
};
|
||||
|
||||
let elapsed = instant.elapsed();
|
||||
if let Ok(affected_rows) = &res {
|
||||
@@ -307,19 +354,23 @@ impl BatchingTask {
|
||||
);
|
||||
} else if let Err(err) = &res {
|
||||
warn!(
|
||||
"Failed to execute Flow {flow_id} on frontend {}, result: {err:?}, elapsed: {:?} with query: {}",
|
||||
peer_addr, elapsed, &plan
|
||||
"Failed to execute Flow {flow_id} on frontend {:?}, result: {err:?}, elapsed: {:?} with query: {}",
|
||||
peer_desc, elapsed, &plan
|
||||
);
|
||||
}
|
||||
|
||||
// record slow query
|
||||
if elapsed >= SLOW_QUERY_THRESHOLD {
|
||||
warn!(
|
||||
"Flow {flow_id} on frontend {} executed for {:?} before complete, query: {}",
|
||||
peer_addr, elapsed, &plan
|
||||
"Flow {flow_id} on frontend {:?} executed for {:?} before complete, query: {}",
|
||||
peer_desc, elapsed, &plan
|
||||
);
|
||||
METRIC_FLOW_BATCHING_ENGINE_SLOW_QUERY
|
||||
.with_label_values(&[flow_id.to_string().as_str(), &plan.to_string(), &peer_addr])
|
||||
.with_label_values(&[
|
||||
flow_id.to_string().as_str(),
|
||||
&plan.to_string(),
|
||||
&peer_desc.unwrap_or_default().to_string(),
|
||||
])
|
||||
.observe(elapsed.as_secs_f64());
|
||||
}
|
||||
|
||||
@@ -328,12 +379,7 @@ impl BatchingTask {
|
||||
.unwrap()
|
||||
.after_query_exec(elapsed, res.is_ok());
|
||||
|
||||
let res = res.context(InvalidRequestSnafu {
|
||||
context: format!(
|
||||
"Failed to execute query for flow={}: \'{}\'",
|
||||
self.config.flow_id, &plan
|
||||
),
|
||||
})?;
|
||||
let res = res?;
|
||||
|
||||
Ok(Some((res, elapsed)))
|
||||
}
|
||||
@@ -372,7 +418,10 @@ impl BatchingTask {
|
||||
}
|
||||
Err(TryRecvError::Empty) => (),
|
||||
}
|
||||
state.get_next_start_query_time(Some(DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT))
|
||||
state.get_next_start_query_time(
|
||||
self.config.flow_id,
|
||||
Some(DEFAULT_BATCHING_ENGINE_QUERY_TIMEOUT),
|
||||
)
|
||||
};
|
||||
tokio::time::sleep_until(sleep_until).await;
|
||||
}
|
||||
@@ -386,14 +435,18 @@ impl BatchingTask {
|
||||
continue;
|
||||
}
|
||||
// TODO(discord9): this error should have better place to go, but for now just print error, also more context is needed
|
||||
Err(err) => match new_query {
|
||||
Some(query) => {
|
||||
common_telemetry::error!(err; "Failed to execute query for flow={} with query: {query}", self.config.flow_id)
|
||||
Err(err) => {
|
||||
match new_query {
|
||||
Some(query) => {
|
||||
common_telemetry::error!(err; "Failed to execute query for flow={} with query: {query}", self.config.flow_id)
|
||||
}
|
||||
None => {
|
||||
common_telemetry::error!(err; "Failed to generate query for flow={}", self.config.flow_id)
|
||||
}
|
||||
}
|
||||
None => {
|
||||
common_telemetry::error!(err; "Failed to generate query for flow={}", self.config.flow_id)
|
||||
}
|
||||
},
|
||||
// also sleep for a little while before try again to prevent flooding logs
|
||||
tokio::time::sleep(MIN_REFRESH_DURATION).await;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -418,7 +471,7 @@ impl BatchingTask {
|
||||
async fn gen_query_with_time_window(
|
||||
&self,
|
||||
engine: QueryEngineRef,
|
||||
sink_table_meta: &RawTableMeta,
|
||||
sink_table_schema: &Arc<Schema>,
|
||||
) -> Result<Option<(LogicalPlan, usize)>, Error> {
|
||||
let query_ctx = self.state.read().unwrap().query_ctx.clone();
|
||||
let start = SystemTime::now();
|
||||
@@ -477,9 +530,11 @@ impl BatchingTask {
|
||||
debug!(
|
||||
"Flow id = {:?}, can't get window size: precise_lower_bound={expire_time_window_bound:?}, using the same query", self.config.flow_id
|
||||
);
|
||||
// clean dirty time window too, this could be from create flow's check_execute
|
||||
self.state.write().unwrap().dirty_time_windows.clean();
|
||||
|
||||
let mut add_auto_column =
|
||||
AddAutoColumnRewriter::new(sink_table_meta.schema.clone());
|
||||
AddAutoColumnRewriter::new(sink_table_schema.clone());
|
||||
let plan = self
|
||||
.config
|
||||
.plan
|
||||
@@ -487,7 +542,10 @@ impl BatchingTask {
|
||||
.clone()
|
||||
.rewrite(&mut add_auto_column)
|
||||
.with_context(|_| DatafusionSnafu {
|
||||
context: format!("Failed to rewrite plan {:?}", self.config.plan),
|
||||
context: format!(
|
||||
"Failed to rewrite plan:\n {}\n",
|
||||
self.config.plan
|
||||
),
|
||||
})?
|
||||
.data;
|
||||
let schema_len = plan.schema().fields().len();
|
||||
@@ -515,18 +573,23 @@ impl BatchingTask {
|
||||
return Ok(None);
|
||||
};
|
||||
|
||||
// TODO(discord9): add auto column or not? This might break compatibility for auto created sink table before this, but that's ok right?
|
||||
|
||||
let mut add_filter = AddFilterRewriter::new(expr);
|
||||
let mut add_auto_column = AddAutoColumnRewriter::new(sink_table_meta.schema.clone());
|
||||
// make a not optimized plan for clearer unparse
|
||||
let mut add_auto_column = AddAutoColumnRewriter::new(sink_table_schema.clone());
|
||||
|
||||
let plan = sql_to_df_plan(query_ctx.clone(), engine.clone(), &self.config.query, false)
|
||||
.await?;
|
||||
plan.clone()
|
||||
let rewrite = plan
|
||||
.clone()
|
||||
.rewrite(&mut add_filter)
|
||||
.and_then(|p| p.data.rewrite(&mut add_auto_column))
|
||||
.with_context(|_| DatafusionSnafu {
|
||||
context: format!("Failed to rewrite plan {plan:?}"),
|
||||
context: format!("Failed to rewrite plan:\n {}\n", plan),
|
||||
})?
|
||||
.data
|
||||
.data;
|
||||
// only apply optimize after complex rewrite is done
|
||||
apply_df_optimizer(rewrite).await?
|
||||
};
|
||||
|
||||
Ok(Some((new_plan, schema_len)))
|
||||
@@ -534,7 +597,7 @@ impl BatchingTask {
|
||||
}
|
||||
|
||||
// auto created table have a auto added column `update_at`, and optional have a `AUTO_CREATED_PLACEHOLDER_TS_COL` column for time index placeholder if no timestamp column is specified
|
||||
// TODO(discord9): unit test
|
||||
// TODO(discord9): for now no default value is set for auto added column for compatibility reason with streaming mode, but this might change in favor of simpler code?
|
||||
fn create_table_with_expr(
|
||||
plan: &LogicalPlan,
|
||||
sink_table_name: &[String; 3],
|
||||
@@ -558,11 +621,7 @@ fn create_table_with_expr(
|
||||
AUTO_CREATED_UPDATE_AT_TS_COL,
|
||||
ConcreteDataType::timestamp_millisecond_datatype(),
|
||||
true,
|
||||
)
|
||||
.with_default_constraint(Some(ColumnDefaultConstraint::Function(NOW_FN.to_string())))
|
||||
.context(DatatypesSnafu {
|
||||
extra: "Failed to build column `update_at TimestampMillisecond default now()`",
|
||||
})?;
|
||||
);
|
||||
column_schemas.push(update_at_schema);
|
||||
|
||||
let time_index = if let Some(time_index) = first_time_stamp {
|
||||
@@ -574,16 +633,7 @@ fn create_table_with_expr(
|
||||
ConcreteDataType::timestamp_millisecond_datatype(),
|
||||
false,
|
||||
)
|
||||
.with_time_index(true)
|
||||
.with_default_constraint(Some(ColumnDefaultConstraint::Value(Value::Timestamp(
|
||||
Timestamp::new_millisecond(0),
|
||||
))))
|
||||
.context(DatatypesSnafu {
|
||||
extra: format!(
|
||||
"Failed to build column `{} TimestampMillisecond TIME INDEX default 0`",
|
||||
AUTO_CREATED_PLACEHOLDER_TS_COL
|
||||
),
|
||||
})?,
|
||||
.with_time_index(true),
|
||||
);
|
||||
AUTO_CREATED_PLACEHOLDER_TS_COL.to_string()
|
||||
};
|
||||
@@ -675,20 +725,14 @@ mod test {
|
||||
AUTO_CREATED_UPDATE_AT_TS_COL,
|
||||
ConcreteDataType::timestamp_millisecond_datatype(),
|
||||
true,
|
||||
)
|
||||
.with_default_constraint(Some(ColumnDefaultConstraint::Function(NOW_FN.to_string())))
|
||||
.unwrap();
|
||||
);
|
||||
|
||||
let ts_placeholder_schema = ColumnSchema::new(
|
||||
AUTO_CREATED_PLACEHOLDER_TS_COL,
|
||||
ConcreteDataType::timestamp_millisecond_datatype(),
|
||||
false,
|
||||
)
|
||||
.with_time_index(true)
|
||||
.with_default_constraint(Some(ColumnDefaultConstraint::Value(Value::Timestamp(
|
||||
Timestamp::new_millisecond(0),
|
||||
))))
|
||||
.unwrap();
|
||||
.with_time_index(true);
|
||||
|
||||
let testcases = vec![
|
||||
TestCase {
|
||||
|
||||
@@ -72,6 +72,17 @@ pub struct TimeWindowExpr {
|
||||
df_schema: DFSchema,
|
||||
}
|
||||
|
||||
impl std::fmt::Display for TimeWindowExpr {
|
||||
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
|
||||
f.debug_struct("TimeWindowExpr")
|
||||
.field("phy_expr", &self.phy_expr.to_string())
|
||||
.field("column_name", &self.column_name)
|
||||
.field("logical_expr", &self.logical_expr.to_string())
|
||||
.field("df_schema", &self.df_schema)
|
||||
.finish()
|
||||
}
|
||||
}
|
||||
|
||||
impl TimeWindowExpr {
|
||||
pub fn from_expr(
|
||||
expr: &Expr,
|
||||
@@ -256,7 +267,7 @@ fn columnar_to_ts_vector(columnar: &ColumnarValue) -> Result<Vec<Option<Timestam
|
||||
Ok(val)
|
||||
}
|
||||
|
||||
/// Return (the column name of time index column, the time window expr, the expected time unit of time index column, the expr's schema for evaluating the time window)
|
||||
/// Return (`the column name of time index column`, `the time window expr`, `the expected time unit of time index column`, `the expr's schema for evaluating the time window`)
|
||||
///
|
||||
/// The time window expr is expected to have one input column with Timestamp type, and also return Timestamp type, the time window expr is expected
|
||||
/// to be monotonic increasing and appears in the innermost GROUP BY clause
|
||||
@@ -693,6 +704,28 @@ mod test {
|
||||
),
|
||||
"SELECT arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)') AS time_window FROM numbers_with_ts WHERE ((ts >= CAST('2025-02-24 10:48:00' AS TIMESTAMP)) AND (ts <= CAST('2025-02-24 10:49:00' AS TIMESTAMP))) GROUP BY arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)')"
|
||||
),
|
||||
// complex time window index with where
|
||||
(
|
||||
"SELECT arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)') AS time_window FROM numbers_with_ts WHERE number in (2, 3, 4) GROUP BY time_window;",
|
||||
Timestamp::new(1740394109, TimeUnit::Second),
|
||||
(
|
||||
"ts".to_string(),
|
||||
Some(Timestamp::new(1740394080, TimeUnit::Second)),
|
||||
Some(Timestamp::new(1740394140, TimeUnit::Second)),
|
||||
),
|
||||
"SELECT arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)') AS time_window FROM numbers_with_ts WHERE numbers_with_ts.number IN (2, 3, 4) AND ((ts >= CAST('2025-02-24 10:48:00' AS TIMESTAMP)) AND (ts <= CAST('2025-02-24 10:49:00' AS TIMESTAMP))) GROUP BY arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)')"
|
||||
),
|
||||
// complex time window index with between and
|
||||
(
|
||||
"SELECT arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)') AS time_window FROM numbers_with_ts WHERE number BETWEEN 2 AND 4 GROUP BY time_window;",
|
||||
Timestamp::new(1740394109, TimeUnit::Second),
|
||||
(
|
||||
"ts".to_string(),
|
||||
Some(Timestamp::new(1740394080, TimeUnit::Second)),
|
||||
Some(Timestamp::new(1740394140, TimeUnit::Second)),
|
||||
),
|
||||
"SELECT arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)') AS time_window FROM numbers_with_ts WHERE (numbers_with_ts.number BETWEEN 2 AND 4) AND ((ts >= CAST('2025-02-24 10:48:00' AS TIMESTAMP)) AND (ts <= CAST('2025-02-24 10:49:00' AS TIMESTAMP))) GROUP BY arrow_cast(date_bin(INTERVAL '1 MINS', numbers_with_ts.ts), 'Timestamp(Second, None)')"
|
||||
),
|
||||
// no time index
|
||||
(
|
||||
"SELECT date_bin('5 minutes', ts) FROM numbers_with_ts;",
|
||||
|
||||
@@ -14,29 +14,63 @@
|
||||
|
||||
//! some utils for helping with batching mode
|
||||
|
||||
use std::collections::HashSet;
|
||||
use std::collections::{BTreeSet, HashSet};
|
||||
use std::sync::Arc;
|
||||
|
||||
use catalog::CatalogManagerRef;
|
||||
use common_error::ext::BoxedError;
|
||||
use common_telemetry::{debug, info};
|
||||
use common_telemetry::debug;
|
||||
use datafusion::error::Result as DfResult;
|
||||
use datafusion::logical_expr::Expr;
|
||||
use datafusion::sql::unparser::Unparser;
|
||||
use datafusion_common::tree_node::{
|
||||
Transformed, TreeNodeRecursion, TreeNodeRewriter, TreeNodeVisitor,
|
||||
};
|
||||
use datafusion_common::DataFusionError;
|
||||
use datafusion_expr::{Distinct, LogicalPlan};
|
||||
use datatypes::schema::RawSchema;
|
||||
use datafusion_common::{DFSchema, DataFusionError, ScalarValue};
|
||||
use datafusion_expr::{Distinct, LogicalPlan, Projection};
|
||||
use datatypes::schema::SchemaRef;
|
||||
use query::parser::QueryLanguageParser;
|
||||
use query::QueryEngineRef;
|
||||
use session::context::QueryContextRef;
|
||||
use snafu::ResultExt;
|
||||
use snafu::{OptionExt, ResultExt};
|
||||
use table::metadata::TableInfo;
|
||||
|
||||
use crate::adapter::AUTO_CREATED_PLACEHOLDER_TS_COL;
|
||||
use crate::df_optimizer::apply_df_optimizer;
|
||||
use crate::error::{DatafusionSnafu, ExternalSnafu};
|
||||
use crate::Error;
|
||||
use crate::error::{DatafusionSnafu, ExternalSnafu, TableNotFoundSnafu};
|
||||
use crate::{Error, TableName};
|
||||
|
||||
pub async fn get_table_info_df_schema(
|
||||
catalog_mr: CatalogManagerRef,
|
||||
table_name: TableName,
|
||||
) -> Result<(Arc<TableInfo>, Arc<DFSchema>), Error> {
|
||||
let full_table_name = table_name.clone().join(".");
|
||||
let table = catalog_mr
|
||||
.table(&table_name[0], &table_name[1], &table_name[2], None)
|
||||
.await
|
||||
.map_err(BoxedError::new)
|
||||
.context(ExternalSnafu)?
|
||||
.context(TableNotFoundSnafu {
|
||||
name: &full_table_name,
|
||||
})?;
|
||||
let table_info = table.table_info().clone();
|
||||
|
||||
let schema = table_info.meta.schema.clone();
|
||||
|
||||
let df_schema: Arc<DFSchema> = Arc::new(
|
||||
schema
|
||||
.arrow_schema()
|
||||
.clone()
|
||||
.try_into()
|
||||
.with_context(|_| DatafusionSnafu {
|
||||
context: format!(
|
||||
"Failed to convert arrow schema to datafusion schema, arrow_schema={:?}",
|
||||
schema.arrow_schema()
|
||||
),
|
||||
})?,
|
||||
);
|
||||
Ok((table_info, df_schema))
|
||||
}
|
||||
|
||||
/// Convert sql to datafusion logical plan
|
||||
pub async fn sql_to_df_plan(
|
||||
@@ -164,14 +198,16 @@ impl TreeNodeVisitor<'_> for FindGroupByFinalName {
|
||||
/// (which doesn't necessary need to have exact name just need to be a extra timestamp column)
|
||||
/// and `__ts_placeholder`(this column need to have exact this name and be a timestamp)
|
||||
/// with values like `now()` and `0`
|
||||
///
|
||||
/// it also give existing columns alias to column in sink table if needed
|
||||
#[derive(Debug)]
|
||||
pub struct AddAutoColumnRewriter {
|
||||
pub schema: RawSchema,
|
||||
pub schema: SchemaRef,
|
||||
pub is_rewritten: bool,
|
||||
}
|
||||
|
||||
impl AddAutoColumnRewriter {
|
||||
pub fn new(schema: RawSchema) -> Self {
|
||||
pub fn new(schema: SchemaRef) -> Self {
|
||||
Self {
|
||||
schema,
|
||||
is_rewritten: false,
|
||||
@@ -181,37 +217,97 @@ impl AddAutoColumnRewriter {
|
||||
|
||||
impl TreeNodeRewriter for AddAutoColumnRewriter {
|
||||
type Node = LogicalPlan;
|
||||
fn f_down(&mut self, node: Self::Node) -> DfResult<Transformed<Self::Node>> {
|
||||
fn f_down(&mut self, mut node: Self::Node) -> DfResult<Transformed<Self::Node>> {
|
||||
if self.is_rewritten {
|
||||
return Ok(Transformed::no(node));
|
||||
}
|
||||
|
||||
// if is distinct all, go one level down
|
||||
if let LogicalPlan::Distinct(Distinct::All(_)) = node {
|
||||
return Ok(Transformed::no(node));
|
||||
// if is distinct all, wrap it in a projection
|
||||
if let LogicalPlan::Distinct(Distinct::All(_)) = &node {
|
||||
let mut exprs = vec![];
|
||||
|
||||
for field in node.schema().fields().iter() {
|
||||
exprs.push(Expr::Column(datafusion::common::Column::new_unqualified(
|
||||
field.name(),
|
||||
)));
|
||||
}
|
||||
|
||||
let projection =
|
||||
LogicalPlan::Projection(Projection::try_new(exprs, Arc::new(node.clone()))?);
|
||||
|
||||
node = projection;
|
||||
}
|
||||
// handle table_scan by wrap it in a projection
|
||||
else if let LogicalPlan::TableScan(table_scan) = node {
|
||||
let mut exprs = vec![];
|
||||
|
||||
for field in table_scan.projected_schema.fields().iter() {
|
||||
exprs.push(Expr::Column(datafusion::common::Column::new(
|
||||
Some(table_scan.table_name.clone()),
|
||||
field.name(),
|
||||
)));
|
||||
}
|
||||
|
||||
let projection = LogicalPlan::Projection(Projection::try_new(
|
||||
exprs,
|
||||
Arc::new(LogicalPlan::TableScan(table_scan)),
|
||||
)?);
|
||||
|
||||
node = projection;
|
||||
}
|
||||
|
||||
// FIXME(discord9): just read plan.expr and do stuffs
|
||||
let mut exprs = node.expressions();
|
||||
// only do rewrite if found the outermost projection
|
||||
let mut exprs = if let LogicalPlan::Projection(project) = &node {
|
||||
project.expr.clone()
|
||||
} else {
|
||||
return Ok(Transformed::no(node));
|
||||
};
|
||||
|
||||
let all_names = self
|
||||
.schema
|
||||
.column_schemas()
|
||||
.iter()
|
||||
.map(|c| c.name.clone())
|
||||
.collect::<BTreeSet<_>>();
|
||||
// first match by position
|
||||
for (idx, expr) in exprs.iter_mut().enumerate() {
|
||||
if !all_names.contains(&expr.qualified_name().1) {
|
||||
if let Some(col_name) = self
|
||||
.schema
|
||||
.column_schemas()
|
||||
.get(idx)
|
||||
.map(|c| c.name.clone())
|
||||
{
|
||||
// if the data type mismatched, later check_execute will error out
|
||||
// hence no need to check it here, beside, optimize pass might be able to cast it
|
||||
// so checking here is not necessary
|
||||
*expr = expr.clone().alias(col_name);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// add columns if have different column count
|
||||
let query_col_cnt = exprs.len();
|
||||
let table_col_cnt = self.schema.column_schemas.len();
|
||||
info!("query_col_cnt={query_col_cnt}, table_col_cnt={table_col_cnt}");
|
||||
let table_col_cnt = self.schema.column_schemas().len();
|
||||
debug!("query_col_cnt={query_col_cnt}, table_col_cnt={table_col_cnt}");
|
||||
|
||||
let placeholder_ts_expr =
|
||||
datafusion::logical_expr::lit(ScalarValue::TimestampMillisecond(Some(0), None))
|
||||
.alias(AUTO_CREATED_PLACEHOLDER_TS_COL);
|
||||
|
||||
if query_col_cnt == table_col_cnt {
|
||||
self.is_rewritten = true;
|
||||
return Ok(Transformed::no(node));
|
||||
// still need to add alias, see below
|
||||
} else if query_col_cnt + 1 == table_col_cnt {
|
||||
let last_col_schema = self.schema.column_schemas.last().unwrap();
|
||||
let last_col_schema = self.schema.column_schemas().last().unwrap();
|
||||
|
||||
// if time index column is auto created add it
|
||||
if last_col_schema.name == AUTO_CREATED_PLACEHOLDER_TS_COL
|
||||
&& self.schema.timestamp_index == Some(table_col_cnt - 1)
|
||||
&& self.schema.timestamp_index() == Some(table_col_cnt - 1)
|
||||
{
|
||||
exprs.push(datafusion::logical_expr::lit(0));
|
||||
exprs.push(placeholder_ts_expr);
|
||||
} else if last_col_schema.data_type.is_timestamp() {
|
||||
// is the update at column
|
||||
exprs.push(datafusion::prelude::now());
|
||||
exprs.push(datafusion::prelude::now().alias(&last_col_schema.name));
|
||||
} else {
|
||||
// helpful error message
|
||||
return Err(DataFusionError::Plan(format!(
|
||||
@@ -221,11 +317,11 @@ impl TreeNodeRewriter for AddAutoColumnRewriter {
|
||||
)));
|
||||
}
|
||||
} else if query_col_cnt + 2 == table_col_cnt {
|
||||
let mut col_iter = self.schema.column_schemas.iter().rev();
|
||||
let mut col_iter = self.schema.column_schemas().iter().rev();
|
||||
let last_col_schema = col_iter.next().unwrap();
|
||||
let second_last_col_schema = col_iter.next().unwrap();
|
||||
if second_last_col_schema.data_type.is_timestamp() {
|
||||
exprs.push(datafusion::prelude::now());
|
||||
exprs.push(datafusion::prelude::now().alias(&second_last_col_schema.name));
|
||||
} else {
|
||||
return Err(DataFusionError::Plan(format!(
|
||||
"Expect the second last column in the table to be timestamp column, found column {} with type {:?}",
|
||||
@@ -235,9 +331,9 @@ impl TreeNodeRewriter for AddAutoColumnRewriter {
|
||||
}
|
||||
|
||||
if last_col_schema.name == AUTO_CREATED_PLACEHOLDER_TS_COL
|
||||
&& self.schema.timestamp_index == Some(table_col_cnt - 1)
|
||||
&& self.schema.timestamp_index() == Some(table_col_cnt - 1)
|
||||
{
|
||||
exprs.push(datafusion::logical_expr::lit(0));
|
||||
exprs.push(placeholder_ts_expr);
|
||||
} else {
|
||||
return Err(DataFusionError::Plan(format!(
|
||||
"Expect timestamp column {}, found {:?}",
|
||||
@@ -247,7 +343,7 @@ impl TreeNodeRewriter for AddAutoColumnRewriter {
|
||||
} else {
|
||||
return Err(DataFusionError::Plan(format!(
|
||||
"Expect table have 0,1 or 2 columns more than query columns, found {} query columns {:?}, {} table columns {:?}",
|
||||
query_col_cnt, node.expressions(), table_col_cnt, self.schema.column_schemas
|
||||
query_col_cnt, exprs, table_col_cnt, self.schema.column_schemas()
|
||||
)));
|
||||
}
|
||||
|
||||
@@ -255,9 +351,12 @@ impl TreeNodeRewriter for AddAutoColumnRewriter {
|
||||
let new_plan = node.with_new_exprs(exprs, node.inputs().into_iter().cloned().collect())?;
|
||||
Ok(Transformed::yes(new_plan))
|
||||
}
|
||||
}
|
||||
|
||||
// TODO(discord9): a method to found out the precise time window
|
||||
/// We might add new columns, so we need to recompute the schema
|
||||
fn f_up(&mut self, node: Self::Node) -> DfResult<Transformed<Self::Node>> {
|
||||
node.recompute_schema().map(Transformed::yes)
|
||||
}
|
||||
}
|
||||
|
||||
/// Find out the `Filter` Node corresponding to innermost(deepest) `WHERE` and add a new filter expr to it
|
||||
#[derive(Debug)]
|
||||
@@ -301,11 +400,15 @@ impl TreeNodeRewriter for AddFilterRewriter {
|
||||
|
||||
#[cfg(test)]
|
||||
mod test {
|
||||
use std::sync::Arc;
|
||||
|
||||
use datafusion_common::tree_node::TreeNode as _;
|
||||
use datatypes::prelude::ConcreteDataType;
|
||||
use datatypes::schema::ColumnSchema;
|
||||
use datatypes::schema::{ColumnSchema, Schema};
|
||||
use pretty_assertions::assert_eq;
|
||||
use query::query_engine::DefaultSerializer;
|
||||
use session::context::QueryContext;
|
||||
use substrait::{DFLogicalSubstraitConvertor, SubstraitPlan};
|
||||
|
||||
use super::*;
|
||||
use crate::test_utils::create_test_query_engine;
|
||||
@@ -386,7 +489,7 @@ mod test {
|
||||
// add update_at
|
||||
(
|
||||
"SELECT number FROM numbers_with_ts",
|
||||
Ok("SELECT numbers_with_ts.number, now() FROM numbers_with_ts"),
|
||||
Ok("SELECT numbers_with_ts.number, now() AS ts FROM numbers_with_ts"),
|
||||
vec![
|
||||
ColumnSchema::new("number", ConcreteDataType::int32_datatype(), true),
|
||||
ColumnSchema::new(
|
||||
@@ -400,7 +503,7 @@ mod test {
|
||||
// add ts placeholder
|
||||
(
|
||||
"SELECT number FROM numbers_with_ts",
|
||||
Ok("SELECT numbers_with_ts.number, 0 FROM numbers_with_ts"),
|
||||
Ok("SELECT numbers_with_ts.number, CAST('1970-01-01 00:00:00' AS TIMESTAMP) AS __ts_placeholder FROM numbers_with_ts"),
|
||||
vec![
|
||||
ColumnSchema::new("number", ConcreteDataType::int32_datatype(), true),
|
||||
ColumnSchema::new(
|
||||
@@ -428,7 +531,7 @@ mod test {
|
||||
// add update_at and ts placeholder
|
||||
(
|
||||
"SELECT number FROM numbers_with_ts",
|
||||
Ok("SELECT numbers_with_ts.number, now(), 0 FROM numbers_with_ts"),
|
||||
Ok("SELECT numbers_with_ts.number, now() AS update_at, CAST('1970-01-01 00:00:00' AS TIMESTAMP) AS __ts_placeholder FROM numbers_with_ts"),
|
||||
vec![
|
||||
ColumnSchema::new("number", ConcreteDataType::int32_datatype(), true),
|
||||
ColumnSchema::new(
|
||||
@@ -447,7 +550,7 @@ mod test {
|
||||
// add ts placeholder
|
||||
(
|
||||
"SELECT number, ts FROM numbers_with_ts",
|
||||
Ok("SELECT numbers_with_ts.number, numbers_with_ts.ts, 0 FROM numbers_with_ts"),
|
||||
Ok("SELECT numbers_with_ts.number, numbers_with_ts.ts AS update_at, CAST('1970-01-01 00:00:00' AS TIMESTAMP) AS __ts_placeholder FROM numbers_with_ts"),
|
||||
vec![
|
||||
ColumnSchema::new("number", ConcreteDataType::int32_datatype(), true),
|
||||
ColumnSchema::new(
|
||||
@@ -466,7 +569,7 @@ mod test {
|
||||
// add update_at after time index column
|
||||
(
|
||||
"SELECT number, ts FROM numbers_with_ts",
|
||||
Ok("SELECT numbers_with_ts.number, numbers_with_ts.ts, now() FROM numbers_with_ts"),
|
||||
Ok("SELECT numbers_with_ts.number, numbers_with_ts.ts, now() AS update_atat FROM numbers_with_ts"),
|
||||
vec![
|
||||
ColumnSchema::new("number", ConcreteDataType::int32_datatype(), true),
|
||||
ColumnSchema::new(
|
||||
@@ -528,8 +631,8 @@ mod test {
|
||||
let query_engine = create_test_query_engine();
|
||||
let ctx = QueryContext::arc();
|
||||
for (before, after, column_schemas) in testcases {
|
||||
let raw_schema = RawSchema::new(column_schemas);
|
||||
let mut add_auto_column_rewriter = AddAutoColumnRewriter::new(raw_schema);
|
||||
let schema = Arc::new(Schema::new(column_schemas));
|
||||
let mut add_auto_column_rewriter = AddAutoColumnRewriter::new(schema);
|
||||
|
||||
let plan = sql_to_df_plan(ctx.clone(), query_engine.clone(), before, false)
|
||||
.await
|
||||
@@ -600,4 +703,18 @@ mod test {
|
||||
);
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_null_cast() {
|
||||
let query_engine = create_test_query_engine();
|
||||
let ctx = QueryContext::arc();
|
||||
let sql = "SELECT NULL::DOUBLE FROM numbers_with_ts";
|
||||
let plan = sql_to_df_plan(ctx, query_engine.clone(), sql, false)
|
||||
.await
|
||||
.unwrap();
|
||||
|
||||
let _sub_plan = DFLogicalSubstraitConvertor {}
|
||||
.encode(&plan, DefaultSerializer)
|
||||
.unwrap();
|
||||
}
|
||||
}
|
||||
|
||||
@@ -25,7 +25,6 @@ use datafusion::config::ConfigOptions;
|
||||
use datafusion::error::DataFusionError;
|
||||
use datafusion::functions_aggregate::count::count_udaf;
|
||||
use datafusion::functions_aggregate::sum::sum_udaf;
|
||||
use datafusion::optimizer::analyzer::count_wildcard_rule::CountWildcardRule;
|
||||
use datafusion::optimizer::analyzer::type_coercion::TypeCoercion;
|
||||
use datafusion::optimizer::common_subexpr_eliminate::CommonSubexprEliminate;
|
||||
use datafusion::optimizer::optimize_projections::OptimizeProjections;
|
||||
@@ -42,6 +41,7 @@ use datafusion_expr::{
|
||||
BinaryExpr, ColumnarValue, Expr, Operator, Projection, ScalarFunctionArgs, ScalarUDFImpl,
|
||||
Signature, TypeSignature, Volatility,
|
||||
};
|
||||
use query::optimizer::count_wildcard::CountWildcardToTimeIndexRule;
|
||||
use query::parser::QueryLanguageParser;
|
||||
use query::query_engine::DefaultSerializer;
|
||||
use query::QueryEngine;
|
||||
@@ -61,9 +61,9 @@ pub async fn apply_df_optimizer(
|
||||
) -> Result<datafusion_expr::LogicalPlan, Error> {
|
||||
let cfg = ConfigOptions::new();
|
||||
let analyzer = Analyzer::with_rules(vec![
|
||||
Arc::new(CountWildcardRule::new()),
|
||||
Arc::new(AvgExpandRule::new()),
|
||||
Arc::new(TumbleExpandRule::new()),
|
||||
Arc::new(CountWildcardToTimeIndexRule),
|
||||
Arc::new(AvgExpandRule),
|
||||
Arc::new(TumbleExpandRule),
|
||||
Arc::new(CheckGroupByRule::new()),
|
||||
Arc::new(TypeCoercion::new()),
|
||||
]);
|
||||
@@ -128,13 +128,7 @@ pub async fn sql_to_flow_plan(
|
||||
}
|
||||
|
||||
#[derive(Debug)]
|
||||
struct AvgExpandRule {}
|
||||
|
||||
impl AvgExpandRule {
|
||||
pub fn new() -> Self {
|
||||
Self {}
|
||||
}
|
||||
}
|
||||
struct AvgExpandRule;
|
||||
|
||||
impl AnalyzerRule for AvgExpandRule {
|
||||
fn analyze(
|
||||
@@ -331,13 +325,7 @@ impl TreeNodeRewriter for ExpandAvgRewriter<'_> {
|
||||
|
||||
/// expand tumble in aggr expr to tumble_start and tumble_end with column name like `window_start`
|
||||
#[derive(Debug)]
|
||||
struct TumbleExpandRule {}
|
||||
|
||||
impl TumbleExpandRule {
|
||||
pub fn new() -> Self {
|
||||
Self {}
|
||||
}
|
||||
}
|
||||
struct TumbleExpandRule;
|
||||
|
||||
impl AnalyzerRule for TumbleExpandRule {
|
||||
fn analyze(
|
||||
|
||||
@@ -49,6 +49,8 @@ pub trait FlowEngine {
|
||||
async fn flush_flow(&self, flow_id: FlowId) -> Result<usize, Error>;
|
||||
/// Check if the flow exists
|
||||
async fn flow_exist(&self, flow_id: FlowId) -> Result<bool, Error>;
|
||||
/// List all flows
|
||||
async fn list_flows(&self) -> Result<impl IntoIterator<Item = FlowId>, Error>;
|
||||
/// Handle the insert requests for the flow
|
||||
async fn handle_flow_inserts(
|
||||
&self,
|
||||
|
||||
@@ -149,6 +149,13 @@ pub enum Error {
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Unsupported: {reason}"))]
|
||||
Unsupported {
|
||||
reason: String,
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Unsupported temporal filter: {reason}"))]
|
||||
UnsupportedTemporalFilter {
|
||||
reason: String,
|
||||
@@ -189,6 +196,25 @@ pub enum Error {
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Illegal check task state: {reason}"))]
|
||||
IllegalCheckTaskState {
|
||||
reason: String,
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display(
|
||||
"Failed to sync with check task for flow {} with allow_drop={}",
|
||||
flow_id,
|
||||
allow_drop
|
||||
))]
|
||||
SyncCheckTask {
|
||||
flow_id: FlowId,
|
||||
allow_drop: bool,
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to start server"))]
|
||||
StartServer {
|
||||
#[snafu(implicit)]
|
||||
@@ -280,10 +306,12 @@ impl ErrorExt for Error {
|
||||
Self::CreateFlow { .. } | Self::Arrow { .. } | Self::Time { .. } => {
|
||||
StatusCode::EngineExecuteQuery
|
||||
}
|
||||
Self::Unexpected { .. } => StatusCode::Unexpected,
|
||||
Self::NotImplemented { .. } | Self::UnsupportedTemporalFilter { .. } => {
|
||||
StatusCode::Unsupported
|
||||
}
|
||||
Self::Unexpected { .. }
|
||||
| Self::SyncCheckTask { .. }
|
||||
| Self::IllegalCheckTaskState { .. } => StatusCode::Unexpected,
|
||||
Self::NotImplemented { .. }
|
||||
| Self::UnsupportedTemporalFilter { .. }
|
||||
| Self::Unsupported { .. } => StatusCode::Unsupported,
|
||||
Self::External { source, .. } => source.status_code(),
|
||||
Self::Internal { .. } | Self::CacheRequired { .. } => StatusCode::Internal,
|
||||
Self::StartServer { source, .. } | Self::ShutdownServer { source, .. } => {
|
||||
|
||||
@@ -43,8 +43,8 @@ mod utils;
|
||||
#[cfg(test)]
|
||||
mod test_utils;
|
||||
|
||||
pub use adapter::{FlowConfig, FlowWorkerManager, FlowWorkerManagerRef, FlownodeOptions};
|
||||
pub use batching_mode::frontend_client::FrontendClient;
|
||||
pub use adapter::{FlowConfig, FlowStreamingEngineRef, FlownodeOptions, StreamingEngine};
|
||||
pub use batching_mode::frontend_client::{FrontendClient, GrpcQueryHandlerWithBoxedError};
|
||||
pub(crate) use engine::{CreateFlowArgs, FlowId, TableName};
|
||||
pub use error::{Error, Result};
|
||||
pub use server::{
|
||||
|
||||
@@ -29,6 +29,7 @@ use common_meta::key::TableMetadataManagerRef;
|
||||
use common_meta::kv_backend::KvBackendRef;
|
||||
use common_meta::node_manager::{Flownode, NodeManagerRef};
|
||||
use common_query::Output;
|
||||
use common_runtime::JoinHandle;
|
||||
use common_telemetry::tracing::info;
|
||||
use futures::{FutureExt, TryStreamExt};
|
||||
use greptime_proto::v1::flow::{flow_server, FlowRequest, FlowResponse, InsertRequests};
|
||||
@@ -50,7 +51,10 @@ use tonic::codec::CompressionEncoding;
|
||||
use tonic::transport::server::TcpIncoming;
|
||||
use tonic::{Request, Response, Status};
|
||||
|
||||
use crate::adapter::{create_worker, FlowWorkerManagerRef};
|
||||
use crate::adapter::flownode_impl::{FlowDualEngine, FlowDualEngineRef};
|
||||
use crate::adapter::{create_worker, FlowStreamingEngineRef};
|
||||
use crate::batching_mode::engine::BatchingEngine;
|
||||
use crate::engine::FlowEngine;
|
||||
use crate::error::{
|
||||
to_status_with_last_err, CacheRequiredSnafu, CreateFlowSnafu, ExternalSnafu, FlowNotFoundSnafu,
|
||||
ListFlowsSnafu, ParseAddrSnafu, ShutdownServerSnafu, StartServerSnafu, UnexpectedSnafu,
|
||||
@@ -59,19 +63,20 @@ use crate::heartbeat::HeartbeatTask;
|
||||
use crate::metrics::{METRIC_FLOW_PROCESSING_TIME, METRIC_FLOW_ROWS};
|
||||
use crate::transform::register_function_to_query_engine;
|
||||
use crate::utils::{SizeReportSender, StateReportHandler};
|
||||
use crate::{CreateFlowArgs, Error, FlowWorkerManager, FlownodeOptions, FrontendClient};
|
||||
use crate::{CreateFlowArgs, Error, FlownodeOptions, FrontendClient, StreamingEngine};
|
||||
|
||||
pub const FLOW_NODE_SERVER_NAME: &str = "FLOW_NODE_SERVER";
|
||||
/// wrapping flow node manager to avoid orphan rule with Arc<...>
|
||||
#[derive(Clone)]
|
||||
pub struct FlowService {
|
||||
/// TODO(discord9): replace with dual engine
|
||||
pub manager: FlowWorkerManagerRef,
|
||||
pub dual_engine: FlowDualEngineRef,
|
||||
}
|
||||
|
||||
impl FlowService {
|
||||
pub fn new(manager: FlowWorkerManagerRef) -> Self {
|
||||
Self { manager }
|
||||
pub fn new(manager: FlowDualEngineRef) -> Self {
|
||||
Self {
|
||||
dual_engine: manager,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -86,7 +91,7 @@ impl flow_server::Flow for FlowService {
|
||||
.start_timer();
|
||||
|
||||
let request = request.into_inner();
|
||||
self.manager
|
||||
self.dual_engine
|
||||
.handle(request)
|
||||
.await
|
||||
.map_err(|err| {
|
||||
@@ -126,7 +131,7 @@ impl flow_server::Flow for FlowService {
|
||||
.with_label_values(&["in"])
|
||||
.inc_by(row_count as u64);
|
||||
|
||||
self.manager
|
||||
self.dual_engine
|
||||
.handle_inserts(request)
|
||||
.await
|
||||
.map(Response::new)
|
||||
@@ -139,11 +144,16 @@ pub struct FlownodeServer {
|
||||
inner: Arc<FlownodeServerInner>,
|
||||
}
|
||||
|
||||
/// FlownodeServerInner is the inner state of FlownodeServer,
|
||||
/// this struct mostly useful for construct/start and stop the
|
||||
/// flow node server
|
||||
struct FlownodeServerInner {
|
||||
/// worker shutdown signal, not to be confused with server_shutdown_tx
|
||||
worker_shutdown_tx: Mutex<broadcast::Sender<()>>,
|
||||
/// server shutdown signal for shutdown grpc server
|
||||
server_shutdown_tx: Mutex<broadcast::Sender<()>>,
|
||||
/// streaming task handler
|
||||
streaming_task_handler: Mutex<Option<JoinHandle<()>>>,
|
||||
flow_service: FlowService,
|
||||
}
|
||||
|
||||
@@ -156,16 +166,28 @@ impl FlownodeServer {
|
||||
flow_service,
|
||||
worker_shutdown_tx: Mutex::new(tx),
|
||||
server_shutdown_tx: Mutex::new(server_tx),
|
||||
streaming_task_handler: Mutex::new(None),
|
||||
}),
|
||||
}
|
||||
}
|
||||
|
||||
/// Start the background task for streaming computation.
|
||||
async fn start_workers(&self) -> Result<(), Error> {
|
||||
let manager_ref = self.inner.flow_service.manager.clone();
|
||||
let _handle = manager_ref
|
||||
.clone()
|
||||
let manager_ref = self.inner.flow_service.dual_engine.clone();
|
||||
let handle = manager_ref
|
||||
.streaming_engine()
|
||||
.run_background(Some(self.inner.worker_shutdown_tx.lock().await.subscribe()));
|
||||
self.inner
|
||||
.streaming_task_handler
|
||||
.lock()
|
||||
.await
|
||||
.replace(handle);
|
||||
|
||||
self.inner
|
||||
.flow_service
|
||||
.dual_engine
|
||||
.start_flow_consistent_check_task()
|
||||
.await?;
|
||||
|
||||
Ok(())
|
||||
}
|
||||
@@ -176,6 +198,11 @@ impl FlownodeServer {
|
||||
if tx.send(()).is_err() {
|
||||
info!("Receiver dropped, the flow node server has already shutdown");
|
||||
}
|
||||
self.inner
|
||||
.flow_service
|
||||
.dual_engine
|
||||
.stop_flow_consistent_check_task()
|
||||
.await?;
|
||||
Ok(())
|
||||
}
|
||||
}
|
||||
@@ -272,8 +299,8 @@ impl FlownodeInstance {
|
||||
&self.flownode_server
|
||||
}
|
||||
|
||||
pub fn flow_worker_manager(&self) -> FlowWorkerManagerRef {
|
||||
self.flownode_server.inner.flow_service.manager.clone()
|
||||
pub fn flow_engine(&self) -> FlowDualEngineRef {
|
||||
self.flownode_server.inner.flow_service.dual_engine.clone()
|
||||
}
|
||||
|
||||
pub fn setup_services(&mut self, services: ServerHandlers) {
|
||||
@@ -342,12 +369,21 @@ impl FlownodeBuilder {
|
||||
self.build_manager(query_engine_factory.query_engine())
|
||||
.await?,
|
||||
);
|
||||
let batching = Arc::new(BatchingEngine::new(
|
||||
self.frontend_client.clone(),
|
||||
query_engine_factory.query_engine(),
|
||||
self.flow_metadata_manager.clone(),
|
||||
self.table_meta.clone(),
|
||||
self.catalog_manager.clone(),
|
||||
));
|
||||
let dual = FlowDualEngine::new(
|
||||
manager.clone(),
|
||||
batching,
|
||||
self.flow_metadata_manager.clone(),
|
||||
self.catalog_manager.clone(),
|
||||
);
|
||||
|
||||
if let Err(err) = self.recover_flows(&manager).await {
|
||||
common_telemetry::error!(err; "Failed to recover flows");
|
||||
}
|
||||
|
||||
let server = FlownodeServer::new(FlowService::new(manager.clone()));
|
||||
let server = FlownodeServer::new(FlowService::new(Arc::new(dual)));
|
||||
|
||||
let heartbeat_task = self.heartbeat_task;
|
||||
|
||||
@@ -364,7 +400,7 @@ impl FlownodeBuilder {
|
||||
/// or recover all existing flow tasks if in standalone mode(nodeid is None)
|
||||
///
|
||||
/// TODO(discord9): persistent flow tasks with internal state
|
||||
async fn recover_flows(&self, manager: &FlowWorkerManagerRef) -> Result<usize, Error> {
|
||||
async fn recover_flows(&self, manager: &FlowDualEngine) -> Result<usize, Error> {
|
||||
let nodeid = self.opts.node_id;
|
||||
let to_be_recovered: Vec<_> = if let Some(nodeid) = nodeid {
|
||||
let to_be_recover = self
|
||||
@@ -401,6 +437,7 @@ impl FlownodeBuilder {
|
||||
let cnt = to_be_recovered.len();
|
||||
|
||||
// TODO(discord9): recover in parallel
|
||||
info!("Recovering {} flows: {:?}", cnt, to_be_recovered);
|
||||
for flow_id in to_be_recovered {
|
||||
let info = self
|
||||
.flow_metadata_manager
|
||||
@@ -416,6 +453,7 @@ impl FlownodeBuilder {
|
||||
info.sink_table_name().schema_name.clone(),
|
||||
info.sink_table_name().table_name.clone(),
|
||||
];
|
||||
|
||||
let args = CreateFlowArgs {
|
||||
flow_id: flow_id as _,
|
||||
sink_table_name,
|
||||
@@ -429,14 +467,27 @@ impl FlownodeBuilder {
|
||||
comment: Some(info.comment().clone()),
|
||||
sql: info.raw_sql().clone(),
|
||||
flow_options: info.options().clone(),
|
||||
query_ctx: Some(
|
||||
QueryContextBuilder::default()
|
||||
.current_catalog(info.catalog_name().clone())
|
||||
.build(),
|
||||
),
|
||||
query_ctx: info
|
||||
.query_context()
|
||||
.clone()
|
||||
.map(|ctx| {
|
||||
ctx.try_into()
|
||||
.map_err(BoxedError::new)
|
||||
.context(ExternalSnafu)
|
||||
})
|
||||
.transpose()?
|
||||
// or use default QueryContext with catalog_name from info
|
||||
// to keep compatibility with old version
|
||||
.or_else(|| {
|
||||
Some(
|
||||
QueryContextBuilder::default()
|
||||
.current_catalog(info.catalog_name().to_string())
|
||||
.build(),
|
||||
)
|
||||
}),
|
||||
};
|
||||
manager
|
||||
.create_flow_inner(args)
|
||||
.create_flow(args)
|
||||
.await
|
||||
.map_err(BoxedError::new)
|
||||
.with_context(|_| CreateFlowSnafu {
|
||||
@@ -452,7 +503,7 @@ impl FlownodeBuilder {
|
||||
async fn build_manager(
|
||||
&mut self,
|
||||
query_engine: Arc<dyn QueryEngine>,
|
||||
) -> Result<FlowWorkerManager, Error> {
|
||||
) -> Result<StreamingEngine, Error> {
|
||||
let table_meta = self.table_meta.clone();
|
||||
|
||||
register_function_to_query_engine(&query_engine);
|
||||
@@ -461,7 +512,7 @@ impl FlownodeBuilder {
|
||||
|
||||
let node_id = self.opts.node_id.map(|id| id as u32);
|
||||
|
||||
let mut man = FlowWorkerManager::new(node_id, query_engine, table_meta);
|
||||
let mut man = StreamingEngine::new(node_id, query_engine, table_meta);
|
||||
for worker_id in 0..num_workers {
|
||||
let (tx, rx) = oneshot::channel();
|
||||
|
||||
@@ -543,6 +594,10 @@ impl<'a> FlownodeServiceBuilder<'a> {
|
||||
}
|
||||
}
|
||||
|
||||
/// Basically a tiny frontend that communicates with datanode, different from [`FrontendClient`] which
|
||||
/// connect to a real frontend instead, this is used for flow's streaming engine. And is for simple query.
|
||||
///
|
||||
/// For heavy query use [`FrontendClient`] which offload computation to frontend, lifting the load from flownode
|
||||
#[derive(Clone)]
|
||||
pub struct FrontendInvoker {
|
||||
inserter: Arc<Inserter>,
|
||||
@@ -564,7 +619,7 @@ impl FrontendInvoker {
|
||||
}
|
||||
|
||||
pub async fn build_from(
|
||||
flow_worker_manager: FlowWorkerManagerRef,
|
||||
flow_streaming_engine: FlowStreamingEngineRef,
|
||||
catalog_manager: CatalogManagerRef,
|
||||
kv_backend: KvBackendRef,
|
||||
layered_cache_registry: LayeredCacheRegistryRef,
|
||||
@@ -599,7 +654,7 @@ impl FrontendInvoker {
|
||||
node_manager.clone(),
|
||||
));
|
||||
|
||||
let query_engine = flow_worker_manager.query_engine.clone();
|
||||
let query_engine = flow_streaming_engine.query_engine.clone();
|
||||
|
||||
let statement_executor = Arc::new(StatementExecutor::new(
|
||||
catalog_manager.clone(),
|
||||
|
||||
@@ -15,6 +15,7 @@ api.workspace = true
|
||||
arc-swap = "1.0"
|
||||
async-trait.workspace = true
|
||||
auth.workspace = true
|
||||
bytes.workspace = true
|
||||
cache.workspace = true
|
||||
catalog.workspace = true
|
||||
client.workspace = true
|
||||
@@ -39,6 +40,7 @@ datafusion.workspace = true
|
||||
datafusion-expr.workspace = true
|
||||
datanode.workspace = true
|
||||
datatypes.workspace = true
|
||||
futures.workspace = true
|
||||
humantime-serde.workspace = true
|
||||
lazy_static.workspace = true
|
||||
log-query.workspace = true
|
||||
@@ -47,6 +49,7 @@ meta-client.workspace = true
|
||||
num_cpus.workspace = true
|
||||
opentelemetry-proto.workspace = true
|
||||
operator.workspace = true
|
||||
otel-arrow-rust.workspace = true
|
||||
partition.workspace = true
|
||||
pipeline.workspace = true
|
||||
prometheus.workspace = true
|
||||
|
||||
@@ -19,6 +19,8 @@ use common_error::define_into_tonic_status;
|
||||
use common_error::ext::{BoxedError, ErrorExt};
|
||||
use common_error::status_code::StatusCode;
|
||||
use common_macro::stack_trace_debug;
|
||||
use common_query::error::datafusion_status_code;
|
||||
use datafusion::error::DataFusionError;
|
||||
use session::ReadPreference;
|
||||
use snafu::{Location, Snafu};
|
||||
use store_api::storage::RegionId;
|
||||
@@ -345,7 +347,15 @@ pub enum Error {
|
||||
SubstraitDecodeLogicalPlan {
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
source: substrait::error::Error,
|
||||
source: common_query::error::Error,
|
||||
},
|
||||
|
||||
#[snafu(display("DataFusionError"))]
|
||||
DataFusion {
|
||||
#[snafu(source)]
|
||||
error: DataFusionError,
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
},
|
||||
}
|
||||
|
||||
@@ -423,6 +433,8 @@ impl ErrorExt for Error {
|
||||
Error::TableOperation { source, .. } => source.status_code(),
|
||||
|
||||
Error::InFlightWriteBytesExceeded { .. } => StatusCode::RateLimited,
|
||||
|
||||
Error::DataFusion { error, .. } => datafusion_status_code::<Self>(error, None),
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@@ -278,7 +278,7 @@ impl SqlQueryHandler for Instance {
|
||||
// plan should be prepared before exec
|
||||
// we'll do check there
|
||||
self.query_engine
|
||||
.execute(plan, query_ctx)
|
||||
.execute(plan.clone(), query_ctx)
|
||||
.await
|
||||
.context(ExecLogicalPlanSnafu)
|
||||
}
|
||||
|
||||
@@ -12,29 +12,33 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::sync::Arc;
|
||||
|
||||
use api::v1::ddl_request::{Expr as DdlExpr, Expr};
|
||||
use api::v1::greptime_request::Request;
|
||||
use api::v1::query_request::Query;
|
||||
use api::v1::{DeleteRequests, DropFlowExpr, InsertRequests, RowDeleteRequests, RowInsertRequests};
|
||||
use api::v1::{
|
||||
DeleteRequests, DropFlowExpr, InsertIntoPlan, InsertRequests, RowDeleteRequests,
|
||||
RowInsertRequests,
|
||||
};
|
||||
use async_trait::async_trait;
|
||||
use auth::{PermissionChecker, PermissionCheckerRef, PermissionReq};
|
||||
use common_base::AffectedRows;
|
||||
use common_query::logical_plan::add_insert_to_logical_plan;
|
||||
use common_query::Output;
|
||||
use common_telemetry::tracing::{self};
|
||||
use datafusion::execution::SessionStateBuilder;
|
||||
use query::parser::PromQuery;
|
||||
use servers::interceptor::{GrpcQueryInterceptor, GrpcQueryInterceptorRef};
|
||||
use servers::query_handler::grpc::{GrpcQueryHandler, RawRecordBatch};
|
||||
use servers::query_handler::sql::SqlQueryHandler;
|
||||
use session::context::QueryContextRef;
|
||||
use snafu::{ensure, OptionExt, ResultExt};
|
||||
use substrait::{DFLogicalSubstraitConvertor, SubstraitPlan};
|
||||
use table::table_name::TableName;
|
||||
|
||||
use crate::error::{
|
||||
CatalogSnafu, Error, InFlightWriteBytesExceededSnafu, IncompleteGrpcRequestSnafu,
|
||||
NotSupportedSnafu, PermissionSnafu, Result, SubstraitDecodeLogicalPlanSnafu,
|
||||
TableNotFoundSnafu, TableOperationSnafu,
|
||||
CatalogSnafu, DataFusionSnafu, Error, InFlightWriteBytesExceededSnafu,
|
||||
IncompleteGrpcRequestSnafu, NotSupportedSnafu, PermissionSnafu, PlanStatementSnafu, Result,
|
||||
SubstraitDecodeLogicalPlanSnafu, TableNotFoundSnafu, TableOperationSnafu,
|
||||
};
|
||||
use crate::instance::{attach_timer, Instance};
|
||||
use crate::metrics::{
|
||||
@@ -91,14 +95,31 @@ impl GrpcQueryHandler for Instance {
|
||||
Query::LogicalPlan(plan) => {
|
||||
// this path is useful internally when flownode needs to execute a logical plan through gRPC interface
|
||||
let timer = GRPC_HANDLE_PLAN_ELAPSED.start_timer();
|
||||
let plan = DFLogicalSubstraitConvertor {}
|
||||
.decode(&*plan, SessionStateBuilder::default().build())
|
||||
|
||||
// use dummy catalog to provide table
|
||||
let plan_decoder = self
|
||||
.query_engine()
|
||||
.engine_context(ctx.clone())
|
||||
.new_plan_decoder()
|
||||
.context(PlanStatementSnafu)?;
|
||||
|
||||
let dummy_catalog_list =
|
||||
Arc::new(catalog::table_source::dummy_catalog::DummyCatalogList::new(
|
||||
self.catalog_manager().clone(),
|
||||
));
|
||||
|
||||
let logical_plan = plan_decoder
|
||||
.decode(bytes::Bytes::from(plan), dummy_catalog_list, true)
|
||||
.await
|
||||
.context(SubstraitDecodeLogicalPlanSnafu)?;
|
||||
let output = SqlQueryHandler::do_exec_plan(self, plan, ctx.clone()).await?;
|
||||
let output =
|
||||
SqlQueryHandler::do_exec_plan(self, logical_plan, ctx.clone()).await?;
|
||||
|
||||
attach_timer(output, timer)
|
||||
}
|
||||
Query::InsertIntoPlan(insert) => {
|
||||
self.handle_insert_plan(insert, ctx.clone()).await?
|
||||
}
|
||||
Query::PromRangeQuery(promql) => {
|
||||
let timer = GRPC_HANDLE_PROMQL_ELAPSED.start_timer();
|
||||
let prom_query = PromQuery {
|
||||
@@ -284,6 +305,91 @@ fn fill_catalog_and_schema_from_context(ddl_expr: &mut DdlExpr, ctx: &QueryConte
|
||||
}
|
||||
|
||||
impl Instance {
|
||||
async fn handle_insert_plan(
|
||||
&self,
|
||||
insert: InsertIntoPlan,
|
||||
ctx: QueryContextRef,
|
||||
) -> Result<Output> {
|
||||
let timer = GRPC_HANDLE_PLAN_ELAPSED.start_timer();
|
||||
let table_name = insert.table_name.context(IncompleteGrpcRequestSnafu {
|
||||
err_msg: "'table_name' is absent in InsertIntoPlan",
|
||||
})?;
|
||||
|
||||
// use dummy catalog to provide table
|
||||
let plan_decoder = self
|
||||
.query_engine()
|
||||
.engine_context(ctx.clone())
|
||||
.new_plan_decoder()
|
||||
.context(PlanStatementSnafu)?;
|
||||
|
||||
let dummy_catalog_list =
|
||||
Arc::new(catalog::table_source::dummy_catalog::DummyCatalogList::new(
|
||||
self.catalog_manager().clone(),
|
||||
));
|
||||
|
||||
// no optimize yet since we still need to add stuff
|
||||
let logical_plan = plan_decoder
|
||||
.decode(
|
||||
bytes::Bytes::from(insert.logical_plan),
|
||||
dummy_catalog_list,
|
||||
false,
|
||||
)
|
||||
.await
|
||||
.context(SubstraitDecodeLogicalPlanSnafu)?;
|
||||
|
||||
let table = self
|
||||
.catalog_manager()
|
||||
.table(
|
||||
&table_name.catalog_name,
|
||||
&table_name.schema_name,
|
||||
&table_name.table_name,
|
||||
None,
|
||||
)
|
||||
.await
|
||||
.context(CatalogSnafu)?
|
||||
.with_context(|| TableNotFoundSnafu {
|
||||
table_name: [
|
||||
table_name.catalog_name.clone(),
|
||||
table_name.schema_name.clone(),
|
||||
table_name.table_name.clone(),
|
||||
]
|
||||
.join("."),
|
||||
})?;
|
||||
|
||||
let table_info = table.table_info();
|
||||
|
||||
let df_schema = Arc::new(
|
||||
table_info
|
||||
.meta
|
||||
.schema
|
||||
.arrow_schema()
|
||||
.clone()
|
||||
.try_into()
|
||||
.context(DataFusionSnafu)?,
|
||||
);
|
||||
|
||||
let insert_into = add_insert_to_logical_plan(table_name, df_schema, logical_plan)
|
||||
.context(SubstraitDecodeLogicalPlanSnafu)?;
|
||||
|
||||
let engine_ctx = self.query_engine().engine_context(ctx.clone());
|
||||
let state = engine_ctx.state();
|
||||
// Analyze the plan
|
||||
let analyzed_plan = state
|
||||
.analyzer()
|
||||
.execute_and_check(insert_into, state.config_options(), |_, _| {})
|
||||
.context(common_query::error::GeneralDataFusionSnafu)
|
||||
.context(SubstraitDecodeLogicalPlanSnafu)?;
|
||||
|
||||
// Optimize the plan
|
||||
let optimized_plan = state
|
||||
.optimize(&analyzed_plan)
|
||||
.context(common_query::error::GeneralDataFusionSnafu)
|
||||
.context(SubstraitDecodeLogicalPlanSnafu)?;
|
||||
|
||||
let output = SqlQueryHandler::do_exec_plan(self, optimized_plan, ctx.clone()).await?;
|
||||
|
||||
Ok(attach_timer(output, timer))
|
||||
}
|
||||
#[tracing::instrument(skip_all)]
|
||||
pub async fn handle_inserts(
|
||||
&self,
|
||||
|
||||
@@ -27,6 +27,7 @@ use servers::http::{HttpServer, HttpServerBuilder};
|
||||
use servers::interceptor::LogIngestInterceptorRef;
|
||||
use servers::metrics_handler::MetricsHandler;
|
||||
use servers::mysql::server::{MysqlServer, MysqlSpawnConfig, MysqlSpawnRef};
|
||||
use servers::otel_arrow::OtelArrowServiceHandler;
|
||||
use servers::postgres::PostgresServer;
|
||||
use servers::query_handler::grpc::ServerGrpcQueryHandlerAdapter;
|
||||
use servers::query_handler::sql::ServerSqlQueryHandlerAdapter;
|
||||
@@ -162,6 +163,7 @@ where
|
||||
let grpc_server = builder
|
||||
.database_handler(greptime_request_handler.clone())
|
||||
.prometheus_handler(self.instance.clone(), user_provider.clone())
|
||||
.otel_arrow_handler(OtelArrowServiceHandler(self.instance.clone()))
|
||||
.flight_handler(Arc::new(greptime_request_handler))
|
||||
.build();
|
||||
Ok(grpc_server)
|
||||
|
||||
@@ -46,7 +46,11 @@ pub struct ChineseTokenizer;
|
||||
|
||||
impl Tokenizer for ChineseTokenizer {
|
||||
fn tokenize<'a>(&self, text: &'a str) -> Vec<&'a str> {
|
||||
JIEBA.cut(text, false)
|
||||
if text.is_ascii() {
|
||||
EnglishTokenizer {}.tokenize(text)
|
||||
} else {
|
||||
JIEBA.cut(text, false)
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
@@ -66,10 +66,12 @@ use crate::election::postgres::PgElection;
|
||||
#[cfg(any(feature = "pg_kvbackend", feature = "mysql_kvbackend"))]
|
||||
use crate::election::CANDIDATE_LEASE_SECS;
|
||||
use crate::metasrv::builder::MetasrvBuilder;
|
||||
use crate::metasrv::{BackendImpl, Metasrv, MetasrvOptions, SelectorRef};
|
||||
use crate::metasrv::{BackendImpl, Metasrv, MetasrvOptions, SelectTarget, SelectorRef};
|
||||
use crate::node_excluder::NodeExcluderRef;
|
||||
use crate::selector::lease_based::LeaseBasedSelector;
|
||||
use crate::selector::load_based::LoadBasedSelector;
|
||||
use crate::selector::round_robin::RoundRobinSelector;
|
||||
use crate::selector::weight_compute::RegionNumsBasedWeightCompute;
|
||||
use crate::selector::SelectorType;
|
||||
use crate::service::admin;
|
||||
use crate::{error, Result};
|
||||
@@ -294,10 +296,31 @@ pub async fn metasrv_builder(
|
||||
|
||||
let in_memory = Arc::new(MemoryKvBackend::new()) as ResettableKvBackendRef;
|
||||
|
||||
let selector = match opts.selector {
|
||||
SelectorType::LoadBased => Arc::new(LoadBasedSelector::default()) as SelectorRef,
|
||||
SelectorType::LeaseBased => Arc::new(LeaseBasedSelector) as SelectorRef,
|
||||
SelectorType::RoundRobin => Arc::new(RoundRobinSelector::default()) as SelectorRef,
|
||||
let node_excluder = plugins
|
||||
.get::<NodeExcluderRef>()
|
||||
.unwrap_or_else(|| Arc::new(Vec::new()) as NodeExcluderRef);
|
||||
let selector = if let Some(selector) = plugins.get::<SelectorRef>() {
|
||||
info!("Using selector from plugins");
|
||||
selector
|
||||
} else {
|
||||
let selector = match opts.selector {
|
||||
SelectorType::LoadBased => Arc::new(LoadBasedSelector::new(
|
||||
RegionNumsBasedWeightCompute,
|
||||
node_excluder,
|
||||
)) as SelectorRef,
|
||||
SelectorType::LeaseBased => {
|
||||
Arc::new(LeaseBasedSelector::new(node_excluder)) as SelectorRef
|
||||
}
|
||||
SelectorType::RoundRobin => Arc::new(RoundRobinSelector::new(
|
||||
SelectTarget::Datanode,
|
||||
node_excluder,
|
||||
)) as SelectorRef,
|
||||
};
|
||||
info!(
|
||||
"Using selector from options, selector type: {}",
|
||||
opts.selector.as_ref()
|
||||
);
|
||||
selector
|
||||
};
|
||||
|
||||
Ok(MetasrvBuilder::new()
|
||||
|
||||
@@ -336,6 +336,13 @@ pub enum Error {
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Region's leader peer changed: {}", msg))]
|
||||
LeaderPeerChanged {
|
||||
msg: String,
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Invalid arguments: {}", err_msg))]
|
||||
InvalidArguments {
|
||||
err_msg: String,
|
||||
@@ -914,7 +921,8 @@ impl ErrorExt for Error {
|
||||
| Error::ProcedureNotFound { .. }
|
||||
| Error::TooManyPartitions { .. }
|
||||
| Error::TomlFormat { .. }
|
||||
| Error::HandlerNotFound { .. } => StatusCode::InvalidArguments,
|
||||
| Error::HandlerNotFound { .. }
|
||||
| Error::LeaderPeerChanged { .. } => StatusCode::InvalidArguments,
|
||||
Error::LeaseKeyFromUtf8 { .. }
|
||||
| Error::LeaseValueFromUtf8 { .. }
|
||||
| Error::InvalidRegionKeyFromUtf8 { .. }
|
||||
|
||||
@@ -12,6 +12,8 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::collections::HashSet;
|
||||
|
||||
use common_error::ext::BoxedError;
|
||||
use common_meta::ddl::flow_meta::PartitionPeerAllocator;
|
||||
use common_meta::peer::Peer;
|
||||
@@ -40,6 +42,7 @@ impl PartitionPeerAllocator for FlowPeerAllocator {
|
||||
SelectorOptions {
|
||||
min_required_items: partitions,
|
||||
allow_duplication: true,
|
||||
exclude_peer_ids: HashSet::new(),
|
||||
},
|
||||
)
|
||||
.await
|
||||
|
||||
@@ -31,6 +31,7 @@ pub mod metasrv;
|
||||
pub mod metrics;
|
||||
#[cfg(feature = "mock")]
|
||||
pub mod mocks;
|
||||
pub mod node_excluder;
|
||||
pub mod procedure;
|
||||
pub mod pubsub;
|
||||
pub mod region;
|
||||
|
||||
@@ -111,6 +111,11 @@ pub struct MetasrvOptions {
|
||||
pub use_memory_store: bool,
|
||||
/// Whether to enable region failover.
|
||||
pub enable_region_failover: bool,
|
||||
/// Whether to allow region failover on local WAL.
|
||||
///
|
||||
/// If it's true, the region failover will be allowed even if the local WAL is used.
|
||||
/// Note that this option is not recommended to be set to true, because it may lead to data loss during failover.
|
||||
pub allow_region_failover_on_local_wal: bool,
|
||||
/// The HTTP server options.
|
||||
pub http: HttpOptions,
|
||||
/// The logging options.
|
||||
@@ -173,6 +178,7 @@ impl Default for MetasrvOptions {
|
||||
selector: SelectorType::default(),
|
||||
use_memory_store: false,
|
||||
enable_region_failover: false,
|
||||
allow_region_failover_on_local_wal: false,
|
||||
http: HttpOptions::default(),
|
||||
logging: LoggingOptions {
|
||||
dir: format!("{METASRV_HOME}/logs"),
|
||||
|
||||
@@ -40,7 +40,8 @@ use common_meta::state_store::KvStateStore;
|
||||
use common_meta::wal_options_allocator::{build_kafka_client, build_wal_options_allocator};
|
||||
use common_procedure::local::{LocalManager, ManagerConfig};
|
||||
use common_procedure::ProcedureManagerRef;
|
||||
use snafu::ResultExt;
|
||||
use common_telemetry::warn;
|
||||
use snafu::{ensure, ResultExt};
|
||||
|
||||
use crate::cache_invalidator::MetasrvCacheInvalidator;
|
||||
use crate::cluster::{MetaPeerClientBuilder, MetaPeerClientRef};
|
||||
@@ -190,7 +191,7 @@ impl MetasrvBuilder {
|
||||
|
||||
let meta_peer_client = meta_peer_client
|
||||
.unwrap_or_else(|| build_default_meta_peer_client(&election, &in_memory));
|
||||
let selector = selector.unwrap_or_else(|| Arc::new(LeaseBasedSelector));
|
||||
let selector = selector.unwrap_or_else(|| Arc::new(LeaseBasedSelector::default()));
|
||||
let pushers = Pushers::default();
|
||||
let mailbox = build_mailbox(&kv_backend, &pushers);
|
||||
let procedure_manager = build_procedure_manager(&options, &kv_backend);
|
||||
@@ -234,13 +235,17 @@ impl MetasrvBuilder {
|
||||
))
|
||||
});
|
||||
|
||||
let flow_selector = Arc::new(RoundRobinSelector::new(
|
||||
SelectTarget::Flownode,
|
||||
Arc::new(Vec::new()),
|
||||
)) as SelectorRef;
|
||||
|
||||
let flow_metadata_allocator = {
|
||||
// for now flownode just use round-robin selector
|
||||
let flow_selector = RoundRobinSelector::new(SelectTarget::Flownode);
|
||||
let flow_selector_ctx = selector_ctx.clone();
|
||||
let peer_allocator = Arc::new(FlowPeerAllocator::new(
|
||||
flow_selector_ctx,
|
||||
Arc::new(flow_selector),
|
||||
flow_selector.clone(),
|
||||
));
|
||||
let seq = Arc::new(
|
||||
SequenceBuilder::new(FLOW_ID_SEQ, kv_backend.clone())
|
||||
@@ -272,18 +277,25 @@ impl MetasrvBuilder {
|
||||
},
|
||||
));
|
||||
let peer_lookup_service = Arc::new(MetaPeerLookupService::new(meta_peer_client.clone()));
|
||||
|
||||
if !is_remote_wal && options.enable_region_failover {
|
||||
return error::UnexpectedSnafu {
|
||||
violated: "Region failover is not supported in the local WAL implementation!",
|
||||
ensure!(
|
||||
options.allow_region_failover_on_local_wal,
|
||||
error::UnexpectedSnafu {
|
||||
violated: "Region failover is not supported in the local WAL implementation!
|
||||
If you want to enable region failover for local WAL, please set `allow_region_failover_on_local_wal` to true.",
|
||||
}
|
||||
);
|
||||
if options.allow_region_failover_on_local_wal {
|
||||
warn!("Region failover is force enabled in the local WAL implementation! This may lead to data loss during failover!");
|
||||
}
|
||||
.fail();
|
||||
}
|
||||
|
||||
let (tx, rx) = RegionSupervisor::channel();
|
||||
let (region_failure_detector_controller, region_supervisor_ticker): (
|
||||
RegionFailureDetectorControllerRef,
|
||||
Option<std::sync::Arc<RegionSupervisorTicker>>,
|
||||
) = if options.enable_region_failover && is_remote_wal {
|
||||
) = if options.enable_region_failover {
|
||||
(
|
||||
Arc::new(RegionFailureDetectorControl::new(tx.clone())) as _,
|
||||
Some(Arc::new(RegionSupervisorTicker::new(
|
||||
@@ -309,7 +321,7 @@ impl MetasrvBuilder {
|
||||
));
|
||||
region_migration_manager.try_start()?;
|
||||
|
||||
let region_failover_handler = if options.enable_region_failover && is_remote_wal {
|
||||
let region_failover_handler = if options.enable_region_failover {
|
||||
let region_supervisor = RegionSupervisor::new(
|
||||
rx,
|
||||
options.failure_detector,
|
||||
@@ -420,7 +432,7 @@ impl MetasrvBuilder {
|
||||
meta_peer_client: meta_peer_client.clone(),
|
||||
selector,
|
||||
// TODO(jeremy): We do not allow configuring the flow selector.
|
||||
flow_selector: Arc::new(RoundRobinSelector::new(SelectTarget::Flownode)),
|
||||
flow_selector,
|
||||
handler_group: RwLock::new(None),
|
||||
handler_group_builder: Mutex::new(Some(handler_group_builder)),
|
||||
election,
|
||||
|
||||
@@ -62,11 +62,22 @@ lazy_static! {
|
||||
register_int_counter!("greptime_meta_region_migration_fail", "meta region migration fail").unwrap();
|
||||
// The heartbeat stat memory size histogram.
|
||||
pub static ref METRIC_META_HEARTBEAT_STAT_MEMORY_SIZE: Histogram =
|
||||
register_histogram!("greptime_meta_heartbeat_stat_memory_size", "meta heartbeat stat memory size").unwrap();
|
||||
register_histogram!("greptime_meta_heartbeat_stat_memory_size", "meta heartbeat stat memory size", vec![
|
||||
100.0, 500.0, 1000.0, 1500.0, 2000.0, 3000.0, 5000.0, 10000.0, 20000.0
|
||||
]).unwrap();
|
||||
// The heartbeat rate counter.
|
||||
pub static ref METRIC_META_HEARTBEAT_RATE: IntCounter =
|
||||
register_int_counter!("greptime_meta_heartbeat_rate", "meta heartbeat arrival rate").unwrap();
|
||||
/// The remote WAL prune execute counter.
|
||||
pub static ref METRIC_META_REMOTE_WAL_PRUNE_EXECUTE: IntCounterVec =
|
||||
register_int_counter_vec!("greptime_meta_remote_wal_prune_execute", "meta remote wal prune execute", &["topic_name"]).unwrap();
|
||||
/// The migration stage elapsed histogram.
|
||||
pub static ref METRIC_META_REGION_MIGRATION_STAGE_ELAPSED: HistogramVec = register_histogram_vec!(
|
||||
"greptime_meta_region_migration_stage_elapsed",
|
||||
"meta region migration stage elapsed",
|
||||
&["stage"],
|
||||
// 0.01 ~ 1000
|
||||
exponential_buckets(0.01, 10.0, 7).unwrap(),
|
||||
)
|
||||
.unwrap();
|
||||
}
|
||||
|
||||
32
src/meta-srv/src/node_excluder.rs
Normal file
32
src/meta-srv/src/node_excluder.rs
Normal file
@@ -0,0 +1,32 @@
|
||||
// Copyright 2023 Greptime Team
|
||||
//
|
||||
// Licensed under the Apache License, Version 2.0 (the "License");
|
||||
// you may not use this file except in compliance with the License.
|
||||
// You may obtain a copy of the License at
|
||||
//
|
||||
// http://www.apache.org/licenses/LICENSE-2.0
|
||||
//
|
||||
// Unless required by applicable law or agreed to in writing, software
|
||||
// distributed under the License is distributed on an "AS IS" BASIS,
|
||||
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::sync::Arc;
|
||||
|
||||
use common_meta::DatanodeId;
|
||||
|
||||
pub type NodeExcluderRef = Arc<dyn NodeExcluder>;
|
||||
|
||||
/// [NodeExcluder] is used to help decide whether some nodes should be excluded (out of consideration)
|
||||
/// in certain situations. For example, in some node selectors.
|
||||
pub trait NodeExcluder: Send + Sync {
|
||||
/// Returns the excluded datanode ids.
|
||||
fn excluded_datanode_ids(&self) -> &Vec<DatanodeId>;
|
||||
}
|
||||
|
||||
impl NodeExcluder for Vec<DatanodeId> {
|
||||
fn excluded_datanode_ids(&self) -> &Vec<DatanodeId> {
|
||||
self
|
||||
}
|
||||
}
|
||||
@@ -25,7 +25,7 @@ pub(crate) mod update_metadata;
|
||||
pub(crate) mod upgrade_candidate_region;
|
||||
|
||||
use std::any::Any;
|
||||
use std::fmt::Debug;
|
||||
use std::fmt::{Debug, Display};
|
||||
use std::time::Duration;
|
||||
|
||||
use common_error::ext::BoxedError;
|
||||
@@ -43,7 +43,7 @@ use common_procedure::error::{
|
||||
Error as ProcedureError, FromJsonSnafu, Result as ProcedureResult, ToJsonSnafu,
|
||||
};
|
||||
use common_procedure::{Context as ProcedureContext, LockKey, Procedure, Status, StringKey};
|
||||
use common_telemetry::info;
|
||||
use common_telemetry::{error, info};
|
||||
use manager::RegionMigrationProcedureGuard;
|
||||
pub use manager::{
|
||||
RegionMigrationManagerRef, RegionMigrationProcedureTask, RegionMigrationProcedureTracker,
|
||||
@@ -55,9 +55,15 @@ use tokio::time::Instant;
|
||||
|
||||
use self::migration_start::RegionMigrationStart;
|
||||
use crate::error::{self, Result};
|
||||
use crate::metrics::{METRIC_META_REGION_MIGRATION_ERROR, METRIC_META_REGION_MIGRATION_EXECUTE};
|
||||
use crate::metrics::{
|
||||
METRIC_META_REGION_MIGRATION_ERROR, METRIC_META_REGION_MIGRATION_EXECUTE,
|
||||
METRIC_META_REGION_MIGRATION_STAGE_ELAPSED,
|
||||
};
|
||||
use crate::service::mailbox::MailboxRef;
|
||||
|
||||
/// The default timeout for region migration.
|
||||
pub const DEFAULT_REGION_MIGRATION_TIMEOUT: Duration = Duration::from_secs(120);
|
||||
|
||||
/// It's shared in each step and available even after recovering.
|
||||
///
|
||||
/// It will only be updated/stored after the Red node has succeeded.
|
||||
@@ -100,6 +106,82 @@ impl PersistentContext {
|
||||
}
|
||||
}
|
||||
|
||||
/// Metrics of region migration.
|
||||
#[derive(Debug, Clone, Default)]
|
||||
pub struct Metrics {
|
||||
/// Elapsed time of downgrading region and upgrading region.
|
||||
operations_elapsed: Duration,
|
||||
/// Elapsed time of downgrading leader region.
|
||||
downgrade_leader_region_elapsed: Duration,
|
||||
/// Elapsed time of open candidate region.
|
||||
open_candidate_region_elapsed: Duration,
|
||||
/// Elapsed time of upgrade candidate region.
|
||||
upgrade_candidate_region_elapsed: Duration,
|
||||
}
|
||||
|
||||
impl Display for Metrics {
|
||||
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
|
||||
write!(
|
||||
f,
|
||||
"operations_elapsed: {:?}, downgrade_leader_region_elapsed: {:?}, open_candidate_region_elapsed: {:?}, upgrade_candidate_region_elapsed: {:?}",
|
||||
self.operations_elapsed,
|
||||
self.downgrade_leader_region_elapsed,
|
||||
self.open_candidate_region_elapsed,
|
||||
self.upgrade_candidate_region_elapsed
|
||||
)
|
||||
}
|
||||
}
|
||||
|
||||
impl Metrics {
|
||||
/// Updates the elapsed time of downgrading region and upgrading region.
|
||||
pub fn update_operations_elapsed(&mut self, elapsed: Duration) {
|
||||
self.operations_elapsed += elapsed;
|
||||
}
|
||||
|
||||
/// Updates the elapsed time of downgrading leader region.
|
||||
pub fn update_downgrade_leader_region_elapsed(&mut self, elapsed: Duration) {
|
||||
self.downgrade_leader_region_elapsed += elapsed;
|
||||
}
|
||||
|
||||
/// Updates the elapsed time of open candidate region.
|
||||
pub fn update_open_candidate_region_elapsed(&mut self, elapsed: Duration) {
|
||||
self.open_candidate_region_elapsed += elapsed;
|
||||
}
|
||||
|
||||
/// Updates the elapsed time of upgrade candidate region.
|
||||
pub fn update_upgrade_candidate_region_elapsed(&mut self, elapsed: Duration) {
|
||||
self.upgrade_candidate_region_elapsed += elapsed;
|
||||
}
|
||||
}
|
||||
|
||||
impl Drop for Metrics {
|
||||
fn drop(&mut self) {
|
||||
if !self.operations_elapsed.is_zero() {
|
||||
METRIC_META_REGION_MIGRATION_STAGE_ELAPSED
|
||||
.with_label_values(&["operations"])
|
||||
.observe(self.operations_elapsed.as_secs_f64());
|
||||
}
|
||||
|
||||
if !self.downgrade_leader_region_elapsed.is_zero() {
|
||||
METRIC_META_REGION_MIGRATION_STAGE_ELAPSED
|
||||
.with_label_values(&["downgrade_leader_region"])
|
||||
.observe(self.downgrade_leader_region_elapsed.as_secs_f64());
|
||||
}
|
||||
|
||||
if !self.open_candidate_region_elapsed.is_zero() {
|
||||
METRIC_META_REGION_MIGRATION_STAGE_ELAPSED
|
||||
.with_label_values(&["open_candidate_region"])
|
||||
.observe(self.open_candidate_region_elapsed.as_secs_f64());
|
||||
}
|
||||
|
||||
if !self.upgrade_candidate_region_elapsed.is_zero() {
|
||||
METRIC_META_REGION_MIGRATION_STAGE_ELAPSED
|
||||
.with_label_values(&["upgrade_candidate_region"])
|
||||
.observe(self.upgrade_candidate_region_elapsed.as_secs_f64());
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// It's shared in each step and available in executing (including retrying).
|
||||
///
|
||||
/// It will be dropped if the procedure runner crashes.
|
||||
@@ -129,8 +211,8 @@ pub struct VolatileContext {
|
||||
leader_region_last_entry_id: Option<u64>,
|
||||
/// The last_entry_id of leader metadata region (Only used for metric engine).
|
||||
leader_region_metadata_last_entry_id: Option<u64>,
|
||||
/// Elapsed time of downgrading region and upgrading region.
|
||||
operations_elapsed: Duration,
|
||||
/// Metrics of region migration.
|
||||
metrics: Metrics,
|
||||
}
|
||||
|
||||
impl VolatileContext {
|
||||
@@ -228,12 +310,35 @@ impl Context {
|
||||
pub fn next_operation_timeout(&self) -> Option<Duration> {
|
||||
self.persistent_ctx
|
||||
.timeout
|
||||
.checked_sub(self.volatile_ctx.operations_elapsed)
|
||||
.checked_sub(self.volatile_ctx.metrics.operations_elapsed)
|
||||
}
|
||||
|
||||
/// Updates operations elapsed.
|
||||
pub fn update_operations_elapsed(&mut self, instant: Instant) {
|
||||
self.volatile_ctx.operations_elapsed += instant.elapsed();
|
||||
self.volatile_ctx
|
||||
.metrics
|
||||
.update_operations_elapsed(instant.elapsed());
|
||||
}
|
||||
|
||||
/// Updates the elapsed time of downgrading leader region.
|
||||
pub fn update_downgrade_leader_region_elapsed(&mut self, instant: Instant) {
|
||||
self.volatile_ctx
|
||||
.metrics
|
||||
.update_downgrade_leader_region_elapsed(instant.elapsed());
|
||||
}
|
||||
|
||||
/// Updates the elapsed time of open candidate region.
|
||||
pub fn update_open_candidate_region_elapsed(&mut self, instant: Instant) {
|
||||
self.volatile_ctx
|
||||
.metrics
|
||||
.update_open_candidate_region_elapsed(instant.elapsed());
|
||||
}
|
||||
|
||||
/// Updates the elapsed time of upgrade candidate region.
|
||||
pub fn update_upgrade_candidate_region_elapsed(&mut self, instant: Instant) {
|
||||
self.volatile_ctx
|
||||
.metrics
|
||||
.update_upgrade_candidate_region_elapsed(instant.elapsed());
|
||||
}
|
||||
|
||||
/// Returns address of meta server.
|
||||
@@ -547,6 +652,14 @@ impl Procedure for RegionMigrationProcedure {
|
||||
.inc();
|
||||
ProcedureError::retry_later(e)
|
||||
} else {
|
||||
error!(
|
||||
e;
|
||||
"Region migration procedure failed, region_id: {}, from_peer: {}, to_peer: {}, {}",
|
||||
self.context.region_id(),
|
||||
self.context.persistent_ctx.from_peer,
|
||||
self.context.persistent_ctx.to_peer,
|
||||
self.context.volatile_ctx.metrics,
|
||||
);
|
||||
METRIC_META_REGION_MIGRATION_ERROR
|
||||
.with_label_values(&[name, "external"])
|
||||
.inc();
|
||||
|
||||
@@ -46,7 +46,13 @@ impl State for CloseDowngradedRegion {
|
||||
let region_id = ctx.region_id();
|
||||
warn!(err; "Failed to close downgraded leader region: {region_id} on datanode {:?}", downgrade_leader_datanode);
|
||||
}
|
||||
|
||||
info!(
|
||||
"Region migration is finished: region_id: {}, from_peer: {}, to_peer: {}, {}",
|
||||
ctx.region_id(),
|
||||
ctx.persistent_ctx.from_peer,
|
||||
ctx.persistent_ctx.to_peer,
|
||||
ctx.volatile_ctx.metrics,
|
||||
);
|
||||
Ok((Box::new(RegionMigrationEnd), Status::done()))
|
||||
}
|
||||
|
||||
|
||||
@@ -54,6 +54,7 @@ impl Default for DowngradeLeaderRegion {
|
||||
#[typetag::serde]
|
||||
impl State for DowngradeLeaderRegion {
|
||||
async fn next(&mut self, ctx: &mut Context) -> Result<(Box<dyn State>, Status)> {
|
||||
let now = Instant::now();
|
||||
// Ensures the `leader_region_lease_deadline` must exist after recovering.
|
||||
ctx.volatile_ctx
|
||||
.set_leader_region_lease_deadline(Duration::from_secs(REGION_LEASE_SECS));
|
||||
@@ -77,6 +78,7 @@ impl State for DowngradeLeaderRegion {
|
||||
}
|
||||
}
|
||||
}
|
||||
ctx.update_downgrade_leader_region_elapsed(now);
|
||||
|
||||
Ok((
|
||||
Box::new(UpgradeCandidateRegion::default()),
|
||||
@@ -348,7 +350,8 @@ mod tests {
|
||||
let env = TestingEnv::new();
|
||||
let mut ctx = env.context_factory().new_context(persistent_context);
|
||||
prepare_table_metadata(&ctx, HashMap::default()).await;
|
||||
ctx.volatile_ctx.operations_elapsed = ctx.persistent_ctx.timeout + Duration::from_secs(1);
|
||||
ctx.volatile_ctx.metrics.operations_elapsed =
|
||||
ctx.persistent_ctx.timeout + Duration::from_secs(1);
|
||||
|
||||
let err = state.downgrade_region(&mut ctx).await.unwrap_err();
|
||||
|
||||
@@ -591,7 +594,8 @@ mod tests {
|
||||
let mut ctx = env.context_factory().new_context(persistent_context);
|
||||
let mailbox_ctx = env.mailbox_context();
|
||||
let mailbox = mailbox_ctx.mailbox().clone();
|
||||
ctx.volatile_ctx.operations_elapsed = ctx.persistent_ctx.timeout + Duration::from_secs(1);
|
||||
ctx.volatile_ctx.metrics.operations_elapsed =
|
||||
ctx.persistent_ctx.timeout + Duration::from_secs(1);
|
||||
|
||||
let (tx, rx) = tokio::sync::mpsc::channel(1);
|
||||
mailbox_ctx
|
||||
|
||||
@@ -267,8 +267,8 @@ impl RegionMigrationManager {
|
||||
|
||||
ensure!(
|
||||
leader_peer.id == task.from_peer.id,
|
||||
error::InvalidArgumentsSnafu {
|
||||
err_msg: format!(
|
||||
error::LeaderPeerChangedSnafu {
|
||||
msg: format!(
|
||||
"Region's leader peer({}) is not the `from_peer`({}), region: {}",
|
||||
leader_peer.id, task.from_peer.id, task.region_id
|
||||
),
|
||||
@@ -507,8 +507,8 @@ mod test {
|
||||
.await;
|
||||
|
||||
let err = manager.submit_procedure(task).await.unwrap_err();
|
||||
assert_matches!(err, error::Error::InvalidArguments { .. });
|
||||
assert_eq!(err.to_string(), "Invalid arguments: Region's leader peer(3) is not the `from_peer`(1), region: 4398046511105(1024, 1)");
|
||||
assert_matches!(err, error::Error::LeaderPeerChanged { .. });
|
||||
assert_eq!(err.to_string(), "Region's leader peer changed: Region's leader peer(3) is not the `from_peer`(1), region: 4398046511105(1024, 1)");
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
|
||||
@@ -15,6 +15,7 @@
|
||||
use std::any::Any;
|
||||
|
||||
use common_procedure::Status;
|
||||
use common_telemetry::warn;
|
||||
use serde::{Deserialize, Serialize};
|
||||
|
||||
use crate::error::{self, Result};
|
||||
@@ -37,7 +38,15 @@ impl RegionMigrationAbort {
|
||||
#[async_trait::async_trait]
|
||||
#[typetag::serde]
|
||||
impl State for RegionMigrationAbort {
|
||||
async fn next(&mut self, _: &mut Context) -> Result<(Box<dyn State>, Status)> {
|
||||
async fn next(&mut self, ctx: &mut Context) -> Result<(Box<dyn State>, Status)> {
|
||||
warn!(
|
||||
"Region migration is aborted: {}, region_id: {}, from_peer: {}, to_peer: {}, {}",
|
||||
self.reason,
|
||||
ctx.region_id(),
|
||||
ctx.persistent_ctx.from_peer,
|
||||
ctx.persistent_ctx.to_peer,
|
||||
ctx.volatile_ctx.metrics,
|
||||
);
|
||||
error::MigrationAbortSnafu {
|
||||
reason: &self.reason,
|
||||
}
|
||||
|
||||
@@ -13,7 +13,7 @@
|
||||
// limitations under the License.
|
||||
|
||||
use std::any::Any;
|
||||
use std::time::{Duration, Instant};
|
||||
use std::time::Duration;
|
||||
|
||||
use api::v1::meta::MailboxMessage;
|
||||
use common_meta::distributed_time_constants::REGION_LEASE_SECS;
|
||||
@@ -24,6 +24,7 @@ use common_procedure::Status;
|
||||
use common_telemetry::info;
|
||||
use serde::{Deserialize, Serialize};
|
||||
use snafu::{OptionExt, ResultExt};
|
||||
use tokio::time::Instant;
|
||||
|
||||
use crate::error::{self, Result};
|
||||
use crate::handler::HeartbeatMailbox;
|
||||
@@ -42,7 +43,9 @@ pub struct OpenCandidateRegion;
|
||||
impl State for OpenCandidateRegion {
|
||||
async fn next(&mut self, ctx: &mut Context) -> Result<(Box<dyn State>, Status)> {
|
||||
let instruction = self.build_open_region_instruction(ctx).await?;
|
||||
let now = Instant::now();
|
||||
self.open_candidate_region(ctx, instruction).await?;
|
||||
ctx.update_open_candidate_region_elapsed(now);
|
||||
|
||||
Ok((
|
||||
Box::new(UpdateMetadata::Downgrade),
|
||||
|
||||
@@ -54,9 +54,12 @@ impl Default for UpgradeCandidateRegion {
|
||||
#[typetag::serde]
|
||||
impl State for UpgradeCandidateRegion {
|
||||
async fn next(&mut self, ctx: &mut Context) -> Result<(Box<dyn State>, Status)> {
|
||||
let now = Instant::now();
|
||||
if self.upgrade_region_with_retry(ctx).await {
|
||||
ctx.update_upgrade_candidate_region_elapsed(now);
|
||||
Ok((Box::new(UpdateMetadata::Upgrade), Status::executing(false)))
|
||||
} else {
|
||||
ctx.update_upgrade_candidate_region_elapsed(now);
|
||||
Ok((Box::new(UpdateMetadata::Rollback), Status::executing(false)))
|
||||
}
|
||||
}
|
||||
@@ -288,7 +291,8 @@ mod tests {
|
||||
let persistent_context = new_persistent_context();
|
||||
let env = TestingEnv::new();
|
||||
let mut ctx = env.context_factory().new_context(persistent_context);
|
||||
ctx.volatile_ctx.operations_elapsed = ctx.persistent_ctx.timeout + Duration::from_secs(1);
|
||||
ctx.volatile_ctx.metrics.operations_elapsed =
|
||||
ctx.persistent_ctx.timeout + Duration::from_secs(1);
|
||||
|
||||
let err = state.upgrade_region(&ctx).await.unwrap_err();
|
||||
|
||||
@@ -558,7 +562,8 @@ mod tests {
|
||||
let mut ctx = env.context_factory().new_context(persistent_context);
|
||||
let mailbox_ctx = env.mailbox_context();
|
||||
let mailbox = mailbox_ctx.mailbox().clone();
|
||||
ctx.volatile_ctx.operations_elapsed = ctx.persistent_ctx.timeout + Duration::from_secs(1);
|
||||
ctx.volatile_ctx.metrics.operations_elapsed =
|
||||
ctx.persistent_ctx.timeout + Duration::from_secs(1);
|
||||
|
||||
let (tx, rx) = tokio::sync::mpsc::channel(1);
|
||||
mailbox_ctx
|
||||
|
||||
@@ -335,22 +335,21 @@ impl WalPruneProcedure {
|
||||
})?;
|
||||
partition_client
|
||||
.delete_records(
|
||||
(self.data.prunable_entry_id + 1) as i64,
|
||||
// notice here no "+1" is needed because the offset arg is exclusive, and it's defensive programming just in case somewhere else have a off by one error, see https://kafka.apache.org/36/javadoc/org/apache/kafka/clients/consumer/KafkaConsumer.html#endOffsets(java.util.Collection) which we use to get the end offset from high watermark
|
||||
self.data.prunable_entry_id as i64,
|
||||
DELETE_RECORDS_TIMEOUT.as_millis() as i32,
|
||||
)
|
||||
.await
|
||||
.context(DeleteRecordsSnafu {
|
||||
topic: &self.data.topic,
|
||||
partition: DEFAULT_PARTITION,
|
||||
offset: (self.data.prunable_entry_id + 1),
|
||||
offset: self.data.prunable_entry_id,
|
||||
})
|
||||
.map_err(BoxedError::new)
|
||||
.with_context(|_| error::RetryLaterWithSourceSnafu {
|
||||
reason: format!(
|
||||
"Failed to delete records for topic: {}, partition: {}, offset: {}",
|
||||
self.data.topic,
|
||||
DEFAULT_PARTITION,
|
||||
self.data.prunable_entry_id + 1
|
||||
self.data.topic, DEFAULT_PARTITION, self.data.prunable_entry_id
|
||||
),
|
||||
})?;
|
||||
info!(
|
||||
@@ -605,19 +604,19 @@ mod tests {
|
||||
// Step 3: Test `on_prune`.
|
||||
let status = procedure.on_prune().await.unwrap();
|
||||
assert_matches!(status, Status::Done { output: None });
|
||||
// Check if the entry ids after `prunable_entry_id` still exist.
|
||||
check_entry_id_existence(
|
||||
procedure.context.client.clone(),
|
||||
&topic_name,
|
||||
procedure.data.prunable_entry_id as i64 + 1,
|
||||
true,
|
||||
)
|
||||
.await;
|
||||
// Check if the entry s before `prunable_entry_id` are deleted.
|
||||
// Check if the entry ids after(include) `prunable_entry_id` still exist.
|
||||
check_entry_id_existence(
|
||||
procedure.context.client.clone(),
|
||||
&topic_name,
|
||||
procedure.data.prunable_entry_id as i64,
|
||||
true,
|
||||
)
|
||||
.await;
|
||||
// Check if the entry ids before `prunable_entry_id` are deleted.
|
||||
check_entry_id_existence(
|
||||
procedure.context.client.clone(),
|
||||
&topic_name,
|
||||
procedure.data.prunable_entry_id as i64 - 1,
|
||||
false,
|
||||
)
|
||||
.await;
|
||||
|
||||
@@ -12,6 +12,7 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::collections::{HashMap, HashSet};
|
||||
use std::fmt::Debug;
|
||||
use std::sync::{Arc, Mutex};
|
||||
use std::time::Duration;
|
||||
@@ -24,9 +25,9 @@ use common_meta::leadership_notifier::LeadershipChangeListener;
|
||||
use common_meta::peer::PeerLookupServiceRef;
|
||||
use common_meta::DatanodeId;
|
||||
use common_runtime::JoinHandle;
|
||||
use common_telemetry::{error, info, warn};
|
||||
use common_telemetry::{debug, error, info, warn};
|
||||
use common_time::util::current_time_millis;
|
||||
use error::Error::{MigrationRunning, TableRouteNotFound};
|
||||
use error::Error::{LeaderPeerChanged, MigrationRunning, TableRouteNotFound};
|
||||
use snafu::{OptionExt, ResultExt};
|
||||
use store_api::storage::RegionId;
|
||||
use tokio::sync::mpsc::{Receiver, Sender};
|
||||
@@ -36,7 +37,9 @@ use crate::error::{self, Result};
|
||||
use crate::failure_detector::PhiAccrualFailureDetectorOptions;
|
||||
use crate::metasrv::{SelectorContext, SelectorRef};
|
||||
use crate::procedure::region_migration::manager::RegionMigrationManagerRef;
|
||||
use crate::procedure::region_migration::RegionMigrationProcedureTask;
|
||||
use crate::procedure::region_migration::{
|
||||
RegionMigrationProcedureTask, DEFAULT_REGION_MIGRATION_TIMEOUT,
|
||||
};
|
||||
use crate::region::failure_detector::RegionFailureDetector;
|
||||
use crate::selector::SelectorOptions;
|
||||
|
||||
@@ -205,6 +208,8 @@ pub const DEFAULT_TICK_INTERVAL: Duration = Duration::from_secs(1);
|
||||
pub struct RegionSupervisor {
|
||||
/// Used to detect the failure of regions.
|
||||
failure_detector: RegionFailureDetector,
|
||||
/// Tracks the number of failovers for each region.
|
||||
failover_counts: HashMap<DetectingRegion, u32>,
|
||||
/// Receives [Event]s.
|
||||
receiver: Receiver<Event>,
|
||||
/// The context of [`SelectorRef`]
|
||||
@@ -290,6 +295,7 @@ impl RegionSupervisor {
|
||||
) -> Self {
|
||||
Self {
|
||||
failure_detector: RegionFailureDetector::new(options),
|
||||
failover_counts: HashMap::new(),
|
||||
receiver: event_receiver,
|
||||
selector_context,
|
||||
selector,
|
||||
@@ -333,13 +339,14 @@ impl RegionSupervisor {
|
||||
}
|
||||
}
|
||||
|
||||
async fn deregister_failure_detectors(&self, detecting_regions: Vec<DetectingRegion>) {
|
||||
async fn deregister_failure_detectors(&mut self, detecting_regions: Vec<DetectingRegion>) {
|
||||
for region in detecting_regions {
|
||||
self.failure_detector.remove(®ion)
|
||||
self.failure_detector.remove(®ion);
|
||||
self.failover_counts.remove(®ion);
|
||||
}
|
||||
}
|
||||
|
||||
async fn handle_region_failures(&self, mut regions: Vec<(DatanodeId, RegionId)>) {
|
||||
async fn handle_region_failures(&mut self, mut regions: Vec<(DatanodeId, RegionId)>) {
|
||||
if regions.is_empty() {
|
||||
return;
|
||||
}
|
||||
@@ -362,16 +369,15 @@ impl RegionSupervisor {
|
||||
.collect::<Vec<_>>();
|
||||
|
||||
for (datanode_id, region_id) in migrating_regions {
|
||||
self.failure_detector.remove(&(datanode_id, region_id));
|
||||
debug!(
|
||||
"Removed region failover for region: {region_id}, datanode: {datanode_id} because it's migrating"
|
||||
);
|
||||
}
|
||||
|
||||
warn!("Detects region failures: {:?}", regions);
|
||||
for (datanode_id, region_id) in regions {
|
||||
match self.do_failover(datanode_id, region_id).await {
|
||||
Ok(_) => self.failure_detector.remove(&(datanode_id, region_id)),
|
||||
Err(err) => {
|
||||
error!(err; "Failed to execute region failover for region: {region_id}, datanode: {datanode_id}");
|
||||
}
|
||||
if let Err(err) = self.do_failover(datanode_id, region_id).await {
|
||||
error!(err; "Failed to execute region failover for region: {region_id}, datanode: {datanode_id}");
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -383,7 +389,12 @@ impl RegionSupervisor {
|
||||
.context(error::MaintenanceModeManagerSnafu)
|
||||
}
|
||||
|
||||
async fn do_failover(&self, datanode_id: DatanodeId, region_id: RegionId) -> Result<()> {
|
||||
async fn do_failover(&mut self, datanode_id: DatanodeId, region_id: RegionId) -> Result<()> {
|
||||
let count = *self
|
||||
.failover_counts
|
||||
.entry((datanode_id, region_id))
|
||||
.and_modify(|count| *count += 1)
|
||||
.or_insert(1);
|
||||
let from_peer = self
|
||||
.peer_lookup
|
||||
.datanode(datanode_id)
|
||||
@@ -401,6 +412,7 @@ impl RegionSupervisor {
|
||||
SelectorOptions {
|
||||
min_required_items: 1,
|
||||
allow_duplication: false,
|
||||
exclude_peer_ids: HashSet::from([from_peer.id]),
|
||||
},
|
||||
)
|
||||
.await?;
|
||||
@@ -411,17 +423,44 @@ impl RegionSupervisor {
|
||||
);
|
||||
return Ok(());
|
||||
}
|
||||
info!(
|
||||
"Failover for region: {region_id}, from_peer: {from_peer}, to_peer: {to_peer}, tries: {count}"
|
||||
);
|
||||
let task = RegionMigrationProcedureTask {
|
||||
region_id,
|
||||
from_peer,
|
||||
to_peer,
|
||||
timeout: Duration::from_secs(60),
|
||||
timeout: DEFAULT_REGION_MIGRATION_TIMEOUT * count,
|
||||
};
|
||||
|
||||
if let Err(err) = self.region_migration_manager.submit_procedure(task).await {
|
||||
return match err {
|
||||
// Returns Ok if it's running or table is dropped.
|
||||
MigrationRunning { .. } | TableRouteNotFound { .. } => Ok(()),
|
||||
MigrationRunning { .. } => {
|
||||
info!(
|
||||
"Another region migration is running, skip failover for region: {}, datanode: {}",
|
||||
region_id, datanode_id
|
||||
);
|
||||
Ok(())
|
||||
}
|
||||
TableRouteNotFound { .. } => {
|
||||
self.deregister_failure_detectors(vec![(datanode_id, region_id)])
|
||||
.await;
|
||||
info!(
|
||||
"Table route is not found, the table is dropped, removed failover detector for region: {}, datanode: {}",
|
||||
region_id, datanode_id
|
||||
);
|
||||
Ok(())
|
||||
}
|
||||
LeaderPeerChanged { .. } => {
|
||||
self.deregister_failure_detectors(vec![(datanode_id, region_id)])
|
||||
.await;
|
||||
info!(
|
||||
"Region's leader peer changed, removed failover detector for region: {}, datanode: {}",
|
||||
region_id, datanode_id
|
||||
);
|
||||
Ok(())
|
||||
}
|
||||
err => Err(err),
|
||||
};
|
||||
};
|
||||
|
||||
@@ -12,15 +12,18 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
mod common;
|
||||
pub mod common;
|
||||
pub mod lease_based;
|
||||
pub mod load_based;
|
||||
pub mod round_robin;
|
||||
#[cfg(test)]
|
||||
pub(crate) mod test_utils;
|
||||
mod weight_compute;
|
||||
mod weighted_choose;
|
||||
pub mod weight_compute;
|
||||
pub mod weighted_choose;
|
||||
use std::collections::HashSet;
|
||||
|
||||
use serde::{Deserialize, Serialize};
|
||||
use strum::AsRefStr;
|
||||
|
||||
use crate::error;
|
||||
use crate::error::Result;
|
||||
@@ -39,6 +42,8 @@ pub struct SelectorOptions {
|
||||
pub min_required_items: usize,
|
||||
/// Whether duplicates are allowed in the selected result, default false.
|
||||
pub allow_duplication: bool,
|
||||
/// The peers to exclude from the selection.
|
||||
pub exclude_peer_ids: HashSet<u64>,
|
||||
}
|
||||
|
||||
impl Default for SelectorOptions {
|
||||
@@ -46,12 +51,13 @@ impl Default for SelectorOptions {
|
||||
Self {
|
||||
min_required_items: 1,
|
||||
allow_duplication: false,
|
||||
exclude_peer_ids: HashSet::new(),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/// [`SelectorType`] refers to the load balancer used when creating tables.
|
||||
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize, Default)]
|
||||
#[derive(Clone, Debug, PartialEq, Eq, Serialize, Deserialize, Default, AsRefStr)]
|
||||
#[serde(try_from = "String")]
|
||||
pub enum SelectorType {
|
||||
/// The current load balancing is based on the number of regions on each datanode node;
|
||||
|
||||
@@ -12,15 +12,25 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::collections::HashSet;
|
||||
|
||||
use common_meta::peer::Peer;
|
||||
use snafu::ensure;
|
||||
|
||||
use crate::error;
|
||||
use crate::error::Result;
|
||||
use crate::metasrv::SelectTarget;
|
||||
use crate::selector::weighted_choose::WeightedChoose;
|
||||
use crate::selector::weighted_choose::{WeightedChoose, WeightedItem};
|
||||
use crate::selector::SelectorOptions;
|
||||
|
||||
/// Filter out the excluded peers from the `weight_array`.
|
||||
pub fn filter_out_excluded_peers(
|
||||
weight_array: &mut Vec<WeightedItem<Peer>>,
|
||||
exclude_peer_ids: &HashSet<u64>,
|
||||
) {
|
||||
weight_array.retain(|peer| !exclude_peer_ids.contains(&peer.item.id));
|
||||
}
|
||||
|
||||
/// According to the `opts`, choose peers from the `weight_array` through `weighted_choose`.
|
||||
pub fn choose_items<W>(opts: &SelectorOptions, weighted_choose: &mut W) -> Result<Vec<Peer>>
|
||||
where
|
||||
@@ -80,7 +90,7 @@ mod tests {
|
||||
|
||||
use common_meta::peer::Peer;
|
||||
|
||||
use crate::selector::common::choose_items;
|
||||
use crate::selector::common::{choose_items, filter_out_excluded_peers};
|
||||
use crate::selector::weighted_choose::{RandomWeightedChoose, WeightedItem};
|
||||
use crate::selector::SelectorOptions;
|
||||
|
||||
@@ -92,35 +102,35 @@ mod tests {
|
||||
id: 1,
|
||||
addr: "127.0.0.1:3001".to_string(),
|
||||
},
|
||||
weight: 1,
|
||||
weight: 1.0,
|
||||
},
|
||||
WeightedItem {
|
||||
item: Peer {
|
||||
id: 2,
|
||||
addr: "127.0.0.1:3001".to_string(),
|
||||
},
|
||||
weight: 1,
|
||||
weight: 1.0,
|
||||
},
|
||||
WeightedItem {
|
||||
item: Peer {
|
||||
id: 3,
|
||||
addr: "127.0.0.1:3001".to_string(),
|
||||
},
|
||||
weight: 1,
|
||||
weight: 1.0,
|
||||
},
|
||||
WeightedItem {
|
||||
item: Peer {
|
||||
id: 4,
|
||||
addr: "127.0.0.1:3001".to_string(),
|
||||
},
|
||||
weight: 1,
|
||||
weight: 1.0,
|
||||
},
|
||||
WeightedItem {
|
||||
item: Peer {
|
||||
id: 5,
|
||||
addr: "127.0.0.1:3001".to_string(),
|
||||
},
|
||||
weight: 1,
|
||||
weight: 1.0,
|
||||
},
|
||||
];
|
||||
|
||||
@@ -128,6 +138,7 @@ mod tests {
|
||||
let opts = SelectorOptions {
|
||||
min_required_items: i,
|
||||
allow_duplication: false,
|
||||
exclude_peer_ids: HashSet::new(),
|
||||
};
|
||||
|
||||
let selected_peers: HashSet<_> =
|
||||
@@ -142,6 +153,7 @@ mod tests {
|
||||
let opts = SelectorOptions {
|
||||
min_required_items: 6,
|
||||
allow_duplication: false,
|
||||
exclude_peer_ids: HashSet::new(),
|
||||
};
|
||||
|
||||
let selected_result =
|
||||
@@ -152,6 +164,7 @@ mod tests {
|
||||
let opts = SelectorOptions {
|
||||
min_required_items: i,
|
||||
allow_duplication: true,
|
||||
exclude_peer_ids: HashSet::new(),
|
||||
};
|
||||
|
||||
let selected_peers =
|
||||
@@ -160,4 +173,30 @@ mod tests {
|
||||
assert_eq!(i, selected_peers.len());
|
||||
}
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn test_filter_out_excluded_peers() {
|
||||
let mut weight_array = vec![
|
||||
WeightedItem {
|
||||
item: Peer {
|
||||
id: 1,
|
||||
addr: "127.0.0.1:3001".to_string(),
|
||||
},
|
||||
weight: 1.0,
|
||||
},
|
||||
WeightedItem {
|
||||
item: Peer {
|
||||
id: 2,
|
||||
addr: "127.0.0.1:3002".to_string(),
|
||||
},
|
||||
weight: 1.0,
|
||||
},
|
||||
];
|
||||
|
||||
let exclude_peer_ids = HashSet::from([1]);
|
||||
filter_out_excluded_peers(&mut weight_array, &exclude_peer_ids);
|
||||
|
||||
assert_eq!(weight_array.len(), 1);
|
||||
assert_eq!(weight_array[0].item.id, 2);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -12,17 +12,37 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::collections::HashSet;
|
||||
use std::sync::Arc;
|
||||
|
||||
use common_meta::peer::Peer;
|
||||
|
||||
use crate::error::Result;
|
||||
use crate::lease;
|
||||
use crate::metasrv::SelectorContext;
|
||||
use crate::selector::common::choose_items;
|
||||
use crate::node_excluder::NodeExcluderRef;
|
||||
use crate::selector::common::{choose_items, filter_out_excluded_peers};
|
||||
use crate::selector::weighted_choose::{RandomWeightedChoose, WeightedItem};
|
||||
use crate::selector::{Selector, SelectorOptions};
|
||||
|
||||
/// Select all alive datanodes based using a random weighted choose.
|
||||
pub struct LeaseBasedSelector;
|
||||
pub struct LeaseBasedSelector {
|
||||
node_excluder: NodeExcluderRef,
|
||||
}
|
||||
|
||||
impl LeaseBasedSelector {
|
||||
pub fn new(node_excluder: NodeExcluderRef) -> Self {
|
||||
Self { node_excluder }
|
||||
}
|
||||
}
|
||||
|
||||
impl Default for LeaseBasedSelector {
|
||||
fn default() -> Self {
|
||||
Self {
|
||||
node_excluder: Arc::new(Vec::new()),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
#[async_trait::async_trait]
|
||||
impl Selector for LeaseBasedSelector {
|
||||
@@ -35,18 +55,26 @@ impl Selector for LeaseBasedSelector {
|
||||
lease::alive_datanodes(&ctx.meta_peer_client, ctx.datanode_lease_secs).await?;
|
||||
|
||||
// 2. compute weight array, but the weight of each item is the same.
|
||||
let weight_array = lease_kvs
|
||||
let mut weight_array = lease_kvs
|
||||
.into_iter()
|
||||
.map(|(k, v)| WeightedItem {
|
||||
item: Peer {
|
||||
id: k.node_id,
|
||||
addr: v.node_addr.clone(),
|
||||
},
|
||||
weight: 1,
|
||||
weight: 1.0,
|
||||
})
|
||||
.collect();
|
||||
|
||||
// 3. choose peers by weight_array.
|
||||
let mut exclude_peer_ids = self
|
||||
.node_excluder
|
||||
.excluded_datanode_ids()
|
||||
.iter()
|
||||
.cloned()
|
||||
.collect::<HashSet<_>>();
|
||||
exclude_peer_ids.extend(opts.exclude_peer_ids.iter());
|
||||
filter_out_excluded_peers(&mut weight_array, &exclude_peer_ids);
|
||||
let mut weighted_choose = RandomWeightedChoose::new(weight_array);
|
||||
let selected = choose_items(&opts, &mut weighted_choose)?;
|
||||
|
||||
|
||||
@@ -12,7 +12,8 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::collections::HashMap;
|
||||
use std::collections::{HashMap, HashSet};
|
||||
use std::sync::Arc;
|
||||
|
||||
use common_meta::datanode::{DatanodeStatKey, DatanodeStatValue};
|
||||
use common_meta::key::TableMetadataManager;
|
||||
@@ -26,18 +27,23 @@ use crate::error::{self, Result};
|
||||
use crate::key::{DatanodeLeaseKey, LeaseValue};
|
||||
use crate::lease;
|
||||
use crate::metasrv::SelectorContext;
|
||||
use crate::selector::common::choose_items;
|
||||
use crate::node_excluder::NodeExcluderRef;
|
||||
use crate::selector::common::{choose_items, filter_out_excluded_peers};
|
||||
use crate::selector::weight_compute::{RegionNumsBasedWeightCompute, WeightCompute};
|
||||
use crate::selector::weighted_choose::RandomWeightedChoose;
|
||||
use crate::selector::{Selector, SelectorOptions};
|
||||
|
||||
pub struct LoadBasedSelector<C> {
|
||||
weight_compute: C,
|
||||
node_excluder: NodeExcluderRef,
|
||||
}
|
||||
|
||||
impl<C> LoadBasedSelector<C> {
|
||||
pub fn new(weight_compute: C) -> Self {
|
||||
Self { weight_compute }
|
||||
pub fn new(weight_compute: C, node_excluder: NodeExcluderRef) -> Self {
|
||||
Self {
|
||||
weight_compute,
|
||||
node_excluder,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -45,6 +51,7 @@ impl Default for LoadBasedSelector<RegionNumsBasedWeightCompute> {
|
||||
fn default() -> Self {
|
||||
Self {
|
||||
weight_compute: RegionNumsBasedWeightCompute,
|
||||
node_excluder: Arc::new(Vec::new()),
|
||||
}
|
||||
}
|
||||
}
|
||||
@@ -85,9 +92,17 @@ where
|
||||
};
|
||||
|
||||
// 4. compute weight array.
|
||||
let weight_array = self.weight_compute.compute(&stat_kvs);
|
||||
let mut weight_array = self.weight_compute.compute(&stat_kvs);
|
||||
|
||||
// 5. choose peers by weight_array.
|
||||
let mut exclude_peer_ids = self
|
||||
.node_excluder
|
||||
.excluded_datanode_ids()
|
||||
.iter()
|
||||
.cloned()
|
||||
.collect::<HashSet<_>>();
|
||||
exclude_peer_ids.extend(opts.exclude_peer_ids.iter());
|
||||
filter_out_excluded_peers(&mut weight_array, &exclude_peer_ids);
|
||||
let mut weighted_choose = RandomWeightedChoose::new(weight_array);
|
||||
let selected = choose_items(&opts, &mut weighted_choose)?;
|
||||
|
||||
|
||||
@@ -12,7 +12,9 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::collections::HashSet;
|
||||
use std::sync::atomic::AtomicUsize;
|
||||
use std::sync::Arc;
|
||||
|
||||
use common_meta::peer::Peer;
|
||||
use snafu::ensure;
|
||||
@@ -20,6 +22,7 @@ use snafu::ensure;
|
||||
use crate::error::{NoEnoughAvailableNodeSnafu, Result};
|
||||
use crate::lease;
|
||||
use crate::metasrv::{SelectTarget, SelectorContext};
|
||||
use crate::node_excluder::NodeExcluderRef;
|
||||
use crate::selector::{Selector, SelectorOptions};
|
||||
|
||||
/// Round-robin selector that returns the next peer in the list in sequence.
|
||||
@@ -32,6 +35,7 @@ use crate::selector::{Selector, SelectorOptions};
|
||||
pub struct RoundRobinSelector {
|
||||
select_target: SelectTarget,
|
||||
counter: AtomicUsize,
|
||||
node_excluder: NodeExcluderRef,
|
||||
}
|
||||
|
||||
impl Default for RoundRobinSelector {
|
||||
@@ -39,32 +43,38 @@ impl Default for RoundRobinSelector {
|
||||
Self {
|
||||
select_target: SelectTarget::Datanode,
|
||||
counter: AtomicUsize::new(0),
|
||||
node_excluder: Arc::new(Vec::new()),
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
impl RoundRobinSelector {
|
||||
pub fn new(select_target: SelectTarget) -> Self {
|
||||
pub fn new(select_target: SelectTarget, node_excluder: NodeExcluderRef) -> Self {
|
||||
Self {
|
||||
select_target,
|
||||
node_excluder,
|
||||
..Default::default()
|
||||
}
|
||||
}
|
||||
|
||||
async fn get_peers(
|
||||
&self,
|
||||
min_required_items: usize,
|
||||
ctx: &SelectorContext,
|
||||
) -> Result<Vec<Peer>> {
|
||||
async fn get_peers(&self, opts: &SelectorOptions, ctx: &SelectorContext) -> Result<Vec<Peer>> {
|
||||
let mut peers = match self.select_target {
|
||||
SelectTarget::Datanode => {
|
||||
// 1. get alive datanodes.
|
||||
let lease_kvs =
|
||||
lease::alive_datanodes(&ctx.meta_peer_client, ctx.datanode_lease_secs).await?;
|
||||
|
||||
let mut exclude_peer_ids = self
|
||||
.node_excluder
|
||||
.excluded_datanode_ids()
|
||||
.iter()
|
||||
.cloned()
|
||||
.collect::<HashSet<_>>();
|
||||
exclude_peer_ids.extend(opts.exclude_peer_ids.iter());
|
||||
// 2. map into peers
|
||||
lease_kvs
|
||||
.into_iter()
|
||||
.filter(|(k, _)| !exclude_peer_ids.contains(&k.node_id))
|
||||
.map(|(k, v)| Peer::new(k.node_id, v.node_addr))
|
||||
.collect::<Vec<_>>()
|
||||
}
|
||||
@@ -84,8 +94,8 @@ impl RoundRobinSelector {
|
||||
ensure!(
|
||||
!peers.is_empty(),
|
||||
NoEnoughAvailableNodeSnafu {
|
||||
required: min_required_items,
|
||||
available: 0usize,
|
||||
required: opts.min_required_items,
|
||||
available: peers.len(),
|
||||
select_target: self.select_target
|
||||
}
|
||||
);
|
||||
@@ -103,7 +113,7 @@ impl Selector for RoundRobinSelector {
|
||||
type Output = Vec<Peer>;
|
||||
|
||||
async fn select(&self, ctx: &Self::Context, opts: SelectorOptions) -> Result<Vec<Peer>> {
|
||||
let peers = self.get_peers(opts.min_required_items, ctx).await?;
|
||||
let peers = self.get_peers(&opts, ctx).await?;
|
||||
// choose peers
|
||||
let mut selected = Vec::with_capacity(opts.min_required_items);
|
||||
for _ in 0..opts.min_required_items {
|
||||
@@ -120,6 +130,8 @@ impl Selector for RoundRobinSelector {
|
||||
|
||||
#[cfg(test)]
|
||||
mod test {
|
||||
use std::collections::HashSet;
|
||||
|
||||
use super::*;
|
||||
use crate::test_util::{create_selector_context, put_datanodes};
|
||||
|
||||
@@ -149,6 +161,7 @@ mod test {
|
||||
SelectorOptions {
|
||||
min_required_items: 4,
|
||||
allow_duplication: true,
|
||||
exclude_peer_ids: HashSet::new(),
|
||||
},
|
||||
)
|
||||
.await
|
||||
@@ -165,6 +178,7 @@ mod test {
|
||||
SelectorOptions {
|
||||
min_required_items: 2,
|
||||
allow_duplication: true,
|
||||
exclude_peer_ids: HashSet::new(),
|
||||
},
|
||||
)
|
||||
.await
|
||||
@@ -172,4 +186,42 @@ mod test {
|
||||
assert_eq!(peers.len(), 2);
|
||||
assert_eq!(peers, vec![peer2.clone(), peer3.clone()]);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_round_robin_selector_with_exclude_peer_ids() {
|
||||
let selector = RoundRobinSelector::new(SelectTarget::Datanode, Arc::new(vec![5]));
|
||||
let ctx = create_selector_context();
|
||||
// add three nodes
|
||||
let peer1 = Peer {
|
||||
id: 2,
|
||||
addr: "node1".to_string(),
|
||||
};
|
||||
let peer2 = Peer {
|
||||
id: 5,
|
||||
addr: "node2".to_string(),
|
||||
};
|
||||
let peer3 = Peer {
|
||||
id: 8,
|
||||
addr: "node3".to_string(),
|
||||
};
|
||||
put_datanodes(
|
||||
&ctx.meta_peer_client,
|
||||
vec![peer1.clone(), peer2.clone(), peer3.clone()],
|
||||
)
|
||||
.await;
|
||||
|
||||
let peers = selector
|
||||
.select(
|
||||
&ctx,
|
||||
SelectorOptions {
|
||||
min_required_items: 1,
|
||||
allow_duplication: true,
|
||||
exclude_peer_ids: HashSet::from([2]),
|
||||
},
|
||||
)
|
||||
.await
|
||||
.unwrap();
|
||||
assert_eq!(peers.len(), 1);
|
||||
assert_eq!(peers, vec![peer3.clone()]);
|
||||
}
|
||||
}
|
||||
|
||||
@@ -84,7 +84,7 @@ impl WeightCompute for RegionNumsBasedWeightCompute {
|
||||
.zip(region_nums)
|
||||
.map(|(peer, region_num)| WeightedItem {
|
||||
item: peer,
|
||||
weight: (max_weight - region_num + base_weight) as usize,
|
||||
weight: (max_weight - region_num + base_weight) as f64,
|
||||
})
|
||||
.collect()
|
||||
}
|
||||
@@ -148,7 +148,7 @@ mod tests {
|
||||
2,
|
||||
);
|
||||
for weight in weight_array.iter() {
|
||||
assert_eq!(*expected.get(&weight.item).unwrap(), weight.weight,);
|
||||
assert_eq!(*expected.get(&weight.item).unwrap(), weight.weight as usize);
|
||||
}
|
||||
|
||||
let mut expected = HashMap::new();
|
||||
|
||||
@@ -42,10 +42,10 @@ pub trait WeightedChoose<Item>: Send + Sync {
|
||||
}
|
||||
|
||||
/// The struct represents a weighted item.
|
||||
#[derive(Debug, Clone, PartialEq, Eq)]
|
||||
#[derive(Debug, Clone, PartialEq)]
|
||||
pub struct WeightedItem<Item> {
|
||||
pub item: Item,
|
||||
pub weight: usize,
|
||||
pub weight: f64,
|
||||
}
|
||||
|
||||
/// A implementation of weighted balance: random weighted choose.
|
||||
@@ -87,7 +87,7 @@ where
|
||||
// unwrap safety: whether weighted_index is none has been checked before.
|
||||
let item = self
|
||||
.items
|
||||
.choose_weighted(&mut rng(), |item| item.weight as f64)
|
||||
.choose_weighted(&mut rng(), |item| item.weight)
|
||||
.context(error::ChooseItemsSnafu)?
|
||||
.item
|
||||
.clone();
|
||||
@@ -95,11 +95,11 @@ where
|
||||
}
|
||||
|
||||
fn choose_multiple(&mut self, amount: usize) -> Result<Vec<Item>> {
|
||||
let amount = amount.min(self.items.iter().filter(|item| item.weight > 0).count());
|
||||
let amount = amount.min(self.items.iter().filter(|item| item.weight > 0.0).count());
|
||||
|
||||
Ok(self
|
||||
.items
|
||||
.choose_multiple_weighted(&mut rng(), amount, |item| item.weight as f64)
|
||||
.choose_multiple_weighted(&mut rng(), amount, |item| item.weight)
|
||||
.context(error::ChooseItemsSnafu)?
|
||||
.cloned()
|
||||
.map(|item| item.item)
|
||||
@@ -120,9 +120,12 @@ mod tests {
|
||||
let mut choose = RandomWeightedChoose::new(vec![
|
||||
WeightedItem {
|
||||
item: 1,
|
||||
weight: 100,
|
||||
weight: 100.0,
|
||||
},
|
||||
WeightedItem {
|
||||
item: 2,
|
||||
weight: 0.0,
|
||||
},
|
||||
WeightedItem { item: 2, weight: 0 },
|
||||
]);
|
||||
|
||||
for _ in 0..100 {
|
||||
|
||||
@@ -12,6 +12,8 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::collections::HashSet;
|
||||
|
||||
use async_trait::async_trait;
|
||||
use common_error::ext::BoxedError;
|
||||
use common_meta::ddl::table_meta::PeerAllocator;
|
||||
@@ -51,6 +53,7 @@ impl MetasrvPeerAllocator {
|
||||
SelectorOptions {
|
||||
min_required_items: regions,
|
||||
allow_duplication: true,
|
||||
exclude_peer_ids: HashSet::new(),
|
||||
},
|
||||
)
|
||||
.await?;
|
||||
|
||||
@@ -18,11 +18,13 @@ common-error.workspace = true
|
||||
common-macro.workspace = true
|
||||
common-query.workspace = true
|
||||
common-recordbatch.workspace = true
|
||||
common-runtime.workspace = true
|
||||
common-telemetry.workspace = true
|
||||
common-time.workspace = true
|
||||
datafusion.workspace = true
|
||||
datatypes.workspace = true
|
||||
futures-util.workspace = true
|
||||
humantime-serde.workspace = true
|
||||
itertools.workspace = true
|
||||
lazy_static = "1.4"
|
||||
mito2.workspace = true
|
||||
|
||||
@@ -12,9 +12,49 @@
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::time::Duration;
|
||||
|
||||
use common_telemetry::warn;
|
||||
use serde::{Deserialize, Serialize};
|
||||
|
||||
#[derive(Debug, Clone, Default, Serialize, Deserialize, PartialEq, Eq)]
|
||||
/// The default flush interval of the metadata region.
|
||||
pub(crate) const DEFAULT_FLUSH_METADATA_REGION_INTERVAL: Duration = Duration::from_secs(30);
|
||||
|
||||
/// Configuration for the metric engine.
|
||||
#[derive(Debug, Clone, Serialize, Deserialize, PartialEq, Eq)]
|
||||
pub struct EngineConfig {
|
||||
/// Experimental feature to use sparse primary key encoding.
|
||||
pub experimental_sparse_primary_key_encoding: bool,
|
||||
/// The flush interval of the metadata region.
|
||||
#[serde(
|
||||
with = "humantime_serde",
|
||||
default = "EngineConfig::default_flush_metadata_region_interval"
|
||||
)]
|
||||
pub flush_metadata_region_interval: Duration,
|
||||
}
|
||||
|
||||
impl Default for EngineConfig {
|
||||
fn default() -> Self {
|
||||
Self {
|
||||
flush_metadata_region_interval: DEFAULT_FLUSH_METADATA_REGION_INTERVAL,
|
||||
experimental_sparse_primary_key_encoding: false,
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
impl EngineConfig {
|
||||
fn default_flush_metadata_region_interval() -> Duration {
|
||||
DEFAULT_FLUSH_METADATA_REGION_INTERVAL
|
||||
}
|
||||
|
||||
/// Sanitizes the configuration.
|
||||
pub fn sanitize(&mut self) {
|
||||
if self.flush_metadata_region_interval.is_zero() {
|
||||
warn!(
|
||||
"Flush metadata region interval is zero, override with default value: {:?}. Disable metadata region flush is forbidden.",
|
||||
DEFAULT_FLUSH_METADATA_REGION_INTERVAL
|
||||
);
|
||||
self.flush_metadata_region_interval = DEFAULT_FLUSH_METADATA_REGION_INTERVAL;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
@@ -34,9 +34,11 @@ use api::region::RegionResponse;
|
||||
use async_trait::async_trait;
|
||||
use common_error::ext::{BoxedError, ErrorExt};
|
||||
use common_error::status_code::StatusCode;
|
||||
use common_runtime::RepeatedTask;
|
||||
use mito2::engine::MitoEngine;
|
||||
pub(crate) use options::IndexOptions;
|
||||
use snafu::ResultExt;
|
||||
pub(crate) use state::MetricEngineState;
|
||||
use store_api::metadata::RegionMetadataRef;
|
||||
use store_api::metric_engine_consts::METRIC_ENGINE_NAME;
|
||||
use store_api::region_engine::{
|
||||
@@ -47,11 +49,11 @@ use store_api::region_engine::{
|
||||
use store_api::region_request::{BatchRegionDdlRequest, RegionRequest};
|
||||
use store_api::storage::{RegionId, ScanRequest, SequenceNumber};
|
||||
|
||||
use self::state::MetricEngineState;
|
||||
use crate::config::EngineConfig;
|
||||
use crate::data_region::DataRegion;
|
||||
use crate::error::{self, Result, UnsupportedRegionRequestSnafu};
|
||||
use crate::error::{self, Error, Result, StartRepeatedTaskSnafu, UnsupportedRegionRequestSnafu};
|
||||
use crate::metadata_region::MetadataRegion;
|
||||
use crate::repeated_task::FlushMetadataRegionTask;
|
||||
use crate::row_modifier::RowModifier;
|
||||
use crate::utils::{self, get_region_statistic};
|
||||
|
||||
@@ -359,19 +361,32 @@ impl RegionEngine for MetricEngine {
|
||||
}
|
||||
|
||||
impl MetricEngine {
|
||||
pub fn new(mito: MitoEngine, config: EngineConfig) -> Self {
|
||||
pub fn try_new(mito: MitoEngine, mut config: EngineConfig) -> Result<Self> {
|
||||
let metadata_region = MetadataRegion::new(mito.clone());
|
||||
let data_region = DataRegion::new(mito.clone());
|
||||
Self {
|
||||
inner: Arc::new(MetricEngineInner {
|
||||
mito,
|
||||
metadata_region,
|
||||
data_region,
|
||||
state: RwLock::default(),
|
||||
config,
|
||||
row_modifier: RowModifier::new(),
|
||||
}),
|
||||
}
|
||||
let state = Arc::new(RwLock::default());
|
||||
config.sanitize();
|
||||
let flush_interval = config.flush_metadata_region_interval;
|
||||
let inner = Arc::new(MetricEngineInner {
|
||||
mito: mito.clone(),
|
||||
metadata_region,
|
||||
data_region,
|
||||
state: state.clone(),
|
||||
config,
|
||||
row_modifier: RowModifier::new(),
|
||||
flush_task: RepeatedTask::new(
|
||||
flush_interval,
|
||||
Box::new(FlushMetadataRegionTask {
|
||||
state: state.clone(),
|
||||
mito: mito.clone(),
|
||||
}),
|
||||
),
|
||||
});
|
||||
inner
|
||||
.flush_task
|
||||
.start(common_runtime::global_runtime())
|
||||
.context(StartRepeatedTaskSnafu { name: "flush_task" })?;
|
||||
Ok(Self { inner })
|
||||
}
|
||||
|
||||
pub fn mito(&self) -> MitoEngine {
|
||||
@@ -426,15 +441,21 @@ impl MetricEngine {
|
||||
) -> Result<common_recordbatch::SendableRecordBatchStream, BoxedError> {
|
||||
self.inner.scan_to_stream(region_id, request).await
|
||||
}
|
||||
|
||||
/// Returns the configuration of the engine.
|
||||
pub fn config(&self) -> &EngineConfig {
|
||||
&self.inner.config
|
||||
}
|
||||
}
|
||||
|
||||
struct MetricEngineInner {
|
||||
mito: MitoEngine,
|
||||
metadata_region: MetadataRegion,
|
||||
data_region: DataRegion,
|
||||
state: RwLock<MetricEngineState>,
|
||||
state: Arc<RwLock<MetricEngineState>>,
|
||||
config: EngineConfig,
|
||||
row_modifier: RowModifier,
|
||||
flush_task: RepeatedTask<Error>,
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
|
||||
@@ -737,7 +737,7 @@ mod test {
|
||||
|
||||
// set up
|
||||
let env = TestEnv::new().await;
|
||||
let engine = MetricEngine::new(env.mito(), EngineConfig::default());
|
||||
let engine = MetricEngine::try_new(env.mito(), EngineConfig::default()).unwrap();
|
||||
let engine_inner = engine.inner;
|
||||
|
||||
// check create data region request
|
||||
|
||||
@@ -282,6 +282,14 @@ pub enum Error {
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
},
|
||||
|
||||
#[snafu(display("Failed to start repeated task: {}", name))]
|
||||
StartRepeatedTask {
|
||||
name: String,
|
||||
source: common_runtime::error::Error,
|
||||
#[snafu(implicit)]
|
||||
location: Location,
|
||||
},
|
||||
}
|
||||
|
||||
pub type Result<T, E = Error> = std::result::Result<T, E>;
|
||||
@@ -335,6 +343,8 @@ impl ErrorExt for Error {
|
||||
|
||||
CollectRecordBatchStream { source, .. } => source.status_code(),
|
||||
|
||||
StartRepeatedTask { source, .. } => source.status_code(),
|
||||
|
||||
MetricManifestInfo { .. } => StatusCode::Internal,
|
||||
}
|
||||
}
|
||||
|
||||
@@ -59,6 +59,7 @@ pub mod engine;
|
||||
pub mod error;
|
||||
mod metadata_region;
|
||||
mod metrics;
|
||||
mod repeated_task;
|
||||
pub mod row_modifier;
|
||||
#[cfg(test)]
|
||||
mod test_util;
|
||||
|
||||
167
src/metric-engine/src/repeated_task.rs
Normal file
167
src/metric-engine/src/repeated_task.rs
Normal file
@@ -0,0 +1,167 @@
|
||||
// Copyright 2023 Greptime Team
|
||||
//
|
||||
// Licensed under the Apache License, Version 2.0 (the "License");
|
||||
// you may not use this file except in compliance with the License.
|
||||
// You may obtain a copy of the License at
|
||||
//
|
||||
// http://www.apache.org/licenses/LICENSE-2.0
|
||||
//
|
||||
// Unless required by applicable law or agreed to in writing, software
|
||||
// distributed under the License is distributed on an "AS IS" BASIS,
|
||||
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
||||
// See the License for the specific language governing permissions and
|
||||
// limitations under the License.
|
||||
|
||||
use std::sync::{Arc, RwLock};
|
||||
use std::time::Instant;
|
||||
|
||||
use common_runtime::TaskFunction;
|
||||
use common_telemetry::{debug, error};
|
||||
use mito2::engine::MitoEngine;
|
||||
use store_api::region_engine::{RegionEngine, RegionRole};
|
||||
use store_api::region_request::{RegionFlushRequest, RegionRequest};
|
||||
|
||||
use crate::engine::MetricEngineState;
|
||||
use crate::error::{Error, Result};
|
||||
use crate::utils;
|
||||
|
||||
/// Task to flush metadata regions.
|
||||
///
|
||||
/// This task is used to send flush requests to the metadata regions
|
||||
/// periodically.
|
||||
pub(crate) struct FlushMetadataRegionTask {
|
||||
pub(crate) state: Arc<RwLock<MetricEngineState>>,
|
||||
pub(crate) mito: MitoEngine,
|
||||
}
|
||||
|
||||
#[async_trait::async_trait]
|
||||
impl TaskFunction<Error> for FlushMetadataRegionTask {
|
||||
fn name(&self) -> &str {
|
||||
"FlushMetadataRegionTask"
|
||||
}
|
||||
|
||||
async fn call(&mut self) -> Result<()> {
|
||||
let region_ids = {
|
||||
let state = self.state.read().unwrap();
|
||||
state
|
||||
.physical_region_states()
|
||||
.keys()
|
||||
.cloned()
|
||||
.collect::<Vec<_>>()
|
||||
};
|
||||
|
||||
let num_region = region_ids.len();
|
||||
let now = Instant::now();
|
||||
for region_id in region_ids {
|
||||
let Some(role) = self.mito.role(region_id) else {
|
||||
continue;
|
||||
};
|
||||
if role == RegionRole::Follower {
|
||||
continue;
|
||||
}
|
||||
let metadata_region_id = utils::to_metadata_region_id(region_id);
|
||||
if let Err(e) = self
|
||||
.mito
|
||||
.handle_request(
|
||||
metadata_region_id,
|
||||
RegionRequest::Flush(RegionFlushRequest {
|
||||
row_group_size: None,
|
||||
}),
|
||||
)
|
||||
.await
|
||||
{
|
||||
error!(e; "Failed to flush metadata region {}", metadata_region_id);
|
||||
}
|
||||
}
|
||||
debug!(
|
||||
"Flushed {} metadata regions, elapsed: {:?}",
|
||||
num_region,
|
||||
now.elapsed()
|
||||
);
|
||||
|
||||
Ok(())
|
||||
}
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use std::assert_matches::assert_matches;
|
||||
use std::time::Duration;
|
||||
|
||||
use store_api::region_engine::{RegionEngine, RegionManifestInfo};
|
||||
|
||||
use crate::config::{EngineConfig, DEFAULT_FLUSH_METADATA_REGION_INTERVAL};
|
||||
use crate::test_util::TestEnv;
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_flush_metadata_region_task() {
|
||||
let env = TestEnv::with_prefix_and_config(
|
||||
"test_flush_metadata_region_task",
|
||||
EngineConfig {
|
||||
flush_metadata_region_interval: Duration::from_millis(100),
|
||||
..Default::default()
|
||||
},
|
||||
)
|
||||
.await;
|
||||
env.init_metric_region().await;
|
||||
let engine = env.metric();
|
||||
// Wait for flush task run
|
||||
tokio::time::sleep(Duration::from_millis(200)).await;
|
||||
let physical_region_id = env.default_physical_region_id();
|
||||
let stat = engine.region_statistic(physical_region_id).unwrap();
|
||||
|
||||
assert_matches!(
|
||||
stat.manifest,
|
||||
RegionManifestInfo::Metric {
|
||||
metadata_manifest_version: 1,
|
||||
metadata_flushed_entry_id: 1,
|
||||
..
|
||||
}
|
||||
)
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_flush_metadata_region_task_with_long_interval() {
|
||||
let env = TestEnv::with_prefix_and_config(
|
||||
"test_flush_metadata_region_task_with_long_interval",
|
||||
EngineConfig {
|
||||
flush_metadata_region_interval: Duration::from_secs(60),
|
||||
..Default::default()
|
||||
},
|
||||
)
|
||||
.await;
|
||||
env.init_metric_region().await;
|
||||
let engine = env.metric();
|
||||
// Wait for flush task run, should not flush metadata region
|
||||
tokio::time::sleep(Duration::from_millis(200)).await;
|
||||
let physical_region_id = env.default_physical_region_id();
|
||||
let stat = engine.region_statistic(physical_region_id).unwrap();
|
||||
|
||||
assert_matches!(
|
||||
stat.manifest,
|
||||
RegionManifestInfo::Metric {
|
||||
metadata_manifest_version: 0,
|
||||
metadata_flushed_entry_id: 0,
|
||||
..
|
||||
}
|
||||
)
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn test_flush_metadata_region_sanitize() {
|
||||
let env = TestEnv::with_prefix_and_config(
|
||||
"test_flush_metadata_region_sanitize",
|
||||
EngineConfig {
|
||||
flush_metadata_region_interval: Duration::from_secs(0),
|
||||
..Default::default()
|
||||
},
|
||||
)
|
||||
.await;
|
||||
let metric = env.metric();
|
||||
let config = metric.config();
|
||||
assert_eq!(
|
||||
config.flush_metadata_region_interval,
|
||||
DEFAULT_FLUSH_METADATA_REGION_INTERVAL
|
||||
);
|
||||
}
|
||||
}
|
||||
@@ -54,9 +54,14 @@ impl TestEnv {
|
||||
|
||||
/// Returns a new env with specific `prefix` for test.
|
||||
pub async fn with_prefix(prefix: &str) -> Self {
|
||||
Self::with_prefix_and_config(prefix, EngineConfig::default()).await
|
||||
}
|
||||
|
||||
/// Returns a new env with specific `prefix` and `config` for test.
|
||||
pub async fn with_prefix_and_config(prefix: &str, config: EngineConfig) -> Self {
|
||||
let mut mito_env = MitoTestEnv::with_prefix(prefix);
|
||||
let mito = mito_env.create_engine(MitoConfig::default()).await;
|
||||
let metric = MetricEngine::new(mito.clone(), EngineConfig::default());
|
||||
let metric = MetricEngine::try_new(mito.clone(), config).unwrap();
|
||||
Self {
|
||||
mito_env,
|
||||
mito,
|
||||
@@ -84,7 +89,7 @@ impl TestEnv {
|
||||
.mito_env
|
||||
.create_follower_engine(MitoConfig::default())
|
||||
.await;
|
||||
let metric = MetricEngine::new(mito.clone(), EngineConfig::default());
|
||||
let metric = MetricEngine::try_new(mito.clone(), EngineConfig::default()).unwrap();
|
||||
|
||||
let region_id = self.default_physical_region_id();
|
||||
debug!("opening default physical region: {region_id}");
|
||||
|
||||
Some files were not shown because too many files have changed in this diff Show More
Reference in New Issue
Block a user