refactor: simplify WAL pruning procedure and introduce region flush trigger (#6741)

* chore: add logs

Signed-off-by: WenyXu <wenymedia@gmail.com>

* refactor: update wal config for metasrv

Signed-off-by: WenyXu <wenymedia@gmail.com>

* feat: introduce region flush trigger

Signed-off-by: WenyXu <wenymedia@gmail.com>

* fix: debug assert

Signed-off-by: WenyXu <wenymedia@gmail.com>

* chore: log level

Signed-off-by: WenyXu <wenymedia@gmail.com>

* refactor: simplify wal prune procedure

Signed-off-by: WenyXu <wenymedia@gmail.com>

* chore: upgrade rskafka

Signed-off-by: WenyXu <wenymedia@gmail.com>

* feat: always flush inactive regions

Signed-off-by: WenyXu <wenymedia@gmail.com>

* refactor: refactor flush trigger

Signed-off-by: WenyXu <wenymedia@gmail.com>

* chore: remove unused code

Signed-off-by: WenyXu <wenymedia@gmail.com>

* fix: typo

Signed-off-by: WenyXu <wenymedia@gmail.com>

* fix: update unit tests

Signed-off-by: WenyXu <wenymedia@gmail.com>

* chore: add metrics

Signed-off-by: WenyXu <wenymedia@gmail.com>

* chore: rename

Signed-off-by: WenyXu <wenymedia@gmail.com>

* chore: apply suggestions from CR

Signed-off-by: WenyXu <wenymedia@gmail.com>

---------

Signed-off-by: WenyXu <wenymedia@gmail.com>
This commit is contained in:
Weny Xu
2025-08-14 22:15:30 +08:00
committed by GitHub
parent 2a3e4c7a82
commit 021ad09c21
28 changed files with 940 additions and 394 deletions

View File

@@ -375,16 +375,16 @@
| `datanode.client.tcp_nodelay` | Bool | `true` | `TCP_NODELAY` option for accepted connections. |
| `wal` | -- | -- | -- |
| `wal.provider` | String | `raft_engine` | -- |
| `wal.broker_endpoints` | Array | -- | The broker endpoints of the Kafka cluster. |
| `wal.auto_create_topics` | Bool | `true` | Automatically create topics for WAL.<br/>Set to `true` to automatically create topics for WAL.<br/>Otherwise, use topics named `topic_name_prefix_[0..num_topics)` |
| `wal.auto_prune_interval` | String | `0s` | Interval of automatically WAL pruning.<br/>Set to `0s` to disable automatically WAL pruning which delete unused remote WAL entries periodically. |
| `wal.trigger_flush_threshold` | Integer | `0` | The threshold to trigger a flush operation of a region in automatically WAL pruning.<br/>Metasrv will send a flush request to flush the region when:<br/>`trigger_flush_threshold` + `prunable_entry_id` < `max_prunable_entry_id`<br/>where:<br/>- `prunable_entry_id` is the maximum entry id that can be pruned of the region.<br/>- `max_prunable_entry_id` is the maximum prunable entry id among all regions in the same topic.<br/>Set to `0` to disable the flush operation. |
| `wal.auto_prune_parallelism` | Integer | `10` | Concurrent task limit for automatically WAL pruning. |
| `wal.num_topics` | Integer | `64` | Number of topics. |
| `wal.selector_type` | String | `round_robin` | Topic selector type.<br/>Available selector types:<br/>- `round_robin` (default) |
| `wal.topic_name_prefix` | String | `greptimedb_wal_topic` | A Kafka topic is constructed by concatenating `topic_name_prefix` and `topic_id`.<br/>Only accepts strings that match the following regular expression pattern:<br/>[a-zA-Z_:-][a-zA-Z0-9_:\-\.@#]*<br/>i.g., greptimedb_wal_topic_0, greptimedb_wal_topic_1. |
| `wal.replication_factor` | Integer | `1` | Expected number of replicas of each partition. |
| `wal.create_topic_timeout` | String | `30s` | Above which a topic creation operation will be cancelled. |
| `wal.broker_endpoints` | Array | -- | The broker endpoints of the Kafka cluster.<br/><br/>**It's only used when the provider is `kafka`**. |
| `wal.auto_create_topics` | Bool | `true` | Automatically create topics for WAL.<br/>Set to `true` to automatically create topics for WAL.<br/>Otherwise, use topics named `topic_name_prefix_[0..num_topics)`<br/>**It's only used when the provider is `kafka`**. |
| `wal.auto_prune_interval` | String | `10m` | Interval of automatically WAL pruning.<br/>Set to `0s` to disable automatically WAL pruning which delete unused remote WAL entries periodically.<br/>**It's only used when the provider is `kafka`**. |
| `wal.flush_trigger_size` | String | `512MB` | Estimated size threshold to trigger a flush when using Kafka remote WAL.<br/>Since multiple regions may share a Kafka topic, the estimated size is calculated as:<br/> (latest_entry_id - flushed_entry_id) * avg_record_size<br/>MetaSrv triggers a flush for a region when this estimated size exceeds `flush_trigger_size`.<br/>- `latest_entry_id`: The latest entry ID in the topic.<br/>- `flushed_entry_id`: The last flushed entry ID for the region.<br/>Set to "0" to let the system decide the flush trigger size.<br/>**It's only used when the provider is `kafka`**. |
| `wal.auto_prune_parallelism` | Integer | `10` | Concurrent task limit for automatically WAL pruning.<br/>**It's only used when the provider is `kafka`**. |
| `wal.num_topics` | Integer | `64` | Number of topics used for remote WAL.<br/>**It's only used when the provider is `kafka`**. |
| `wal.selector_type` | String | `round_robin` | Topic selector type.<br/>Available selector types:<br/>- `round_robin` (default)<br/>**It's only used when the provider is `kafka`**. |
| `wal.topic_name_prefix` | String | `greptimedb_wal_topic` | A Kafka topic is constructed by concatenating `topic_name_prefix` and `topic_id`.<br/>Only accepts strings that match the following regular expression pattern:<br/>[a-zA-Z_:-][a-zA-Z0-9_:\-\.@#]*<br/>i.g., greptimedb_wal_topic_0, greptimedb_wal_topic_1.<br/>**It's only used when the provider is `kafka`**. |
| `wal.replication_factor` | Integer | `1` | Expected number of replicas of each partition.<br/>**It's only used when the provider is `kafka`**. |
| `wal.create_topic_timeout` | String | `30s` | The timeout for creating a Kafka topic.<br/>**It's only used when the provider is `kafka`**. |
| `event_recorder` | -- | -- | Configuration options for the event recorder. |
| `event_recorder.ttl` | String | `90d` | TTL for the events table that will be used to store the events. Default is `90d`. |
| `logging` | -- | -- | The logging options. |

View File

@@ -176,50 +176,61 @@ tcp_nodelay = true
# - `kafka`: metasrv **have to be** configured with kafka wal config when using kafka wal provider in datanode.
provider = "raft_engine"
# Kafka wal config.
## The broker endpoints of the Kafka cluster.
##
## **It's only used when the provider is `kafka`**.
broker_endpoints = ["127.0.0.1:9092"]
## Automatically create topics for WAL.
## Set to `true` to automatically create topics for WAL.
## Otherwise, use topics named `topic_name_prefix_[0..num_topics)`
## **It's only used when the provider is `kafka`**.
auto_create_topics = true
## Interval of automatically WAL pruning.
## Set to `0s` to disable automatically WAL pruning which delete unused remote WAL entries periodically.
auto_prune_interval = "0s"
## **It's only used when the provider is `kafka`**.
auto_prune_interval = "10m"
## The threshold to trigger a flush operation of a region in automatically WAL pruning.
## Metasrv will send a flush request to flush the region when:
## `trigger_flush_threshold` + `prunable_entry_id` < `max_prunable_entry_id`
## where:
## - `prunable_entry_id` is the maximum entry id that can be pruned of the region.
## - `max_prunable_entry_id` is the maximum prunable entry id among all regions in the same topic.
## Set to `0` to disable the flush operation.
trigger_flush_threshold = 0
## Estimated size threshold to trigger a flush when using Kafka remote WAL.
## Since multiple regions may share a Kafka topic, the estimated size is calculated as:
## (latest_entry_id - flushed_entry_id) * avg_record_size
## MetaSrv triggers a flush for a region when this estimated size exceeds `flush_trigger_size`.
## - `latest_entry_id`: The latest entry ID in the topic.
## - `flushed_entry_id`: The last flushed entry ID for the region.
## Set to "0" to let the system decide the flush trigger size.
## **It's only used when the provider is `kafka`**.
flush_trigger_size = "512MB"
## Concurrent task limit for automatically WAL pruning.
## **It's only used when the provider is `kafka`**.
auto_prune_parallelism = 10
## Number of topics.
## Number of topics used for remote WAL.
## **It's only used when the provider is `kafka`**.
num_topics = 64
## Topic selector type.
## Available selector types:
## - `round_robin` (default)
## **It's only used when the provider is `kafka`**.
selector_type = "round_robin"
## A Kafka topic is constructed by concatenating `topic_name_prefix` and `topic_id`.
## Only accepts strings that match the following regular expression pattern:
## [a-zA-Z_:-][a-zA-Z0-9_:\-\.@#]*
## i.g., greptimedb_wal_topic_0, greptimedb_wal_topic_1.
## **It's only used when the provider is `kafka`**.
topic_name_prefix = "greptimedb_wal_topic"
## Expected number of replicas of each partition.
## **It's only used when the provider is `kafka`**.
replication_factor = 1
## Above which a topic creation operation will be cancelled.
## The timeout for creating a Kafka topic.
## **It's only used when the provider is `kafka`**.
create_topic_timeout = "30s"
# The Kafka SASL configuration.