* refactor(mito): retain region compaction and DDL lifecycle state
Signed-off-by: Lei, HUANG <ratuthomm@gmail.com>
* fix(mito): finalize opening batch in the worker on registration
Clean up the `opening_regions` entry and reply to its waiters in the same
serialized worker step that registers an opened region, instead of waiting
for the background opener to resume after the registration acknowledgement.
Previously the entry was only removed by the background opener, so a close
followed by a reopen could join the stale opening batch and receive a false
success after the region no longer existed. Replying from the worker closes
that window because registration and waiter cleanup are now atomic with
respect to the worker loop.
Add a test-only `EventListener::on_region_open_registered` hook and a
regression test that holds the first background opener to verify a reopen
does not wait for the previous open task. On failure the region is still
stopped and waiters receive `RegionClosed`.
Signed-off-by: Lei, HUANG <ratuthomm@gmail.com>
* fix(mito): unregister drop runtime state before WAL cleanup
The region leaves the region map before its WAL entries are obsoleted. A
failure in `wal.obsolete(...).await?` returned early and skipped
`cleanup_dropped_region_runtime_state`, so the resident compaction, flush,
and index scheduler entries were never unregistered. Since the region is no
longer in the map, a later close cannot clean them up either, leaking the
entries for the rest of the worker's lifetime.
Move the runtime-state cleanup to immediately after the region leaves the
map, before the fallible WAL cleanup. The schedulers only touch their own
state and do not depend on the region map or WAL, so the reordering is safe
and leaves the success path unchanged.
Signed-off-by: Lei, HUANG <ratuthomm@gmail.com>
* refactor(mito): tighten visibility of new compaction state items
Keep `ActiveCompaction::idle` private (it is only used inside `state.rs`)
and expose the other items added by the resident-region refactor as
`pub(crate)` instead of `pub(super)`, per the style guide. Also drop a
stale comment in `on_compaction_finished`: DDL handoff now happens in
`take_ready_ddls` before this block, which only runs when no DDL is
queued.
Signed-off-by: Lei, HUANG <ratuthomm@gmail.com>
* fix(mito): skip compaction scheduling while a region is truncating
`handle_flush_finished` dispatches queued DDLs before scheduling
compaction. A truncate queued behind a committing flush marks the region
as `Truncating` synchronously, but `schedule_compaction` only skipped
staging states, so it could schedule a new automatic compaction while the
truncate was still running.
Add `MitoRegion::is_truncating` and skip compaction scheduling for a
truncating region. Add a regression test that queues a full truncate
behind a committing flush and asserts no compaction is scheduled.
Signed-off-by: Lei, HUANG <ratuthomm@gmail.com>
* fix(mito): log DDL-fenced compaction rejection
The refactor dropped the log emitted when a region was skipped because of
pending DDL requests. Automatic triggers carry no waiter, so the rejection
became silent and could not be told apart from a scheduling interval that
had not elapsed. Log the queued and executing DDL state before cancelling.
Signed-off-by: Lei, HUANG <ratuthomm@gmail.com>
* docs(mito): document DDL reply fence invariants
Explain why fence_ddl_replies installs executing_ddl before handing off
the batch and why CompactionDdlComplete carries both a scheduler-wide
generation and a per-slot request_id. The generation fences replies from
superseded batches or region incarnations, while request_id makes
completion idempotent so only the last distinct reply releases the fence.
Signed-off-by: Lei, HUANG <ratuthomm@gmail.com>
* refactor(mito): rename DDL completion request_id to request_index
The value is the DDL's index within the handed-off batch, not a globally
stable request identifier; request_id suggested otherwise. Rename it to
request_index across the scheduler, worker dispatch, and tests.
Signed-off-by: Lei, HUANG <ratuthomm@gmail.com>
---------
Signed-off-by: Lei, HUANG <ratuthomm@gmail.com>
* ci: benchmark the object store WAL against S3 from an EC2 runner
Add a workflow_dispatch workflow that loads the o11ybench agent observability
dataset into a fresh GreptimeDB container per cell and compares three WAL
configurations that differ only in [wal]: Raft Engine with sync_write on the
instance store, and the object store WAL in durable and enqueued mode, with
the data files on S3 in the runner's region for all three. Rounds alternate
the configuration order. A summary reports rows per second, D/R and E/R as
the geometric mean over rounds, append acknowledgement p50 and p99, and the
object store WAL counters. A smoke mode runs the same chain on ubuntu-latest
against the MinIO fixture.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* ci: scrape WAL metrics on failed loads and keep logic out of the benchmark YAML
Scrape the object store WAL metrics before stopping a cell's server whether
the load succeeded or failed, so failed cells keep their stall, poison and
create failure counters. Build the summary from the plan so that cells which
never ran are listed and leave their ratios undefined. Move budget, region,
runtime image, manifest and credential scan logic from the workflow into the
helper scripts, and drive the cell lifecycle in the tests through fake docker
and curl binaries.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* ci: drop redundant required flags from benchmark inputs
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* ci: fail a benchmark cell whose metrics scrape fails and zero untouched WAL counters
A load that succeeds without its metrics scrape now fails the cell with exit
code 4 and records metrics_scraped in cell.json. The summary reports an
event counter that never registered as zero when the cell is an object store
WAL cell whose scrape succeeded and created objects, and as unknown otherwise.
Planned cells that never ran are read through the same path as recorded ones.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* ci: register the S3 benchmark workflow for dispatch from its branch
A workflow_dispatch workflow can only be dispatched once GitHub has
registered it, which happens on the default branch or when one of its
triggers fires. A push trigger limited to this branch registers it; validate
runs only for workflow_dispatch, so a push run skips every job.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* ci: delete the benchmark S3 prefix one object at a time
boto3 now sends a CRC checksum instead of Content-MD5 with DeleteObjects,
which the MinIO fixture of the smoke run rejects. DeleteObject needs neither
and the prefix of one run holds at most a few thousand objects.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* ci: default the S3 benchmark to c6id.2xlarge
c7id is not offered in the runner region ap-southeast-1, where DescribeInstanceTypes
rejects it. c6id is the compute optimized x86 family with a local NVMe instance
store that the region offers, with the same 8 vCPU and 16 GiB at 2xlarge.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* ci: drop the registration trigger of the S3 benchmark workflow
The workflow is registered now, so dispatches no longer need a push trigger
on its branch.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* ci: fail an S3 benchmark run on EC2 without a local instance store
Without an instance store the Raft Engine WAL of the R cells silently stayed
on the root EBS volume, so the run measured a different baseline. Only a smoke
run may keep it on the root volume now.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* ci: let the instance store check fail its workflow step
The step piped wal-disk into tee without pipefail, so a rejected host still
passed. wal-disk now writes its record itself and the step runs it alone.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
---------
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: add the object store WAL RFC
Propose an object store backed write-ahead log as a WAL provider next to
Raft Engine and Kafka: a datanode batches the entries of its regions into
immutable objects under one prefix, creates each object conditionally
under a monotonically increasing sequence, acknowledges a write once its
object is durable by default, and replays each region from its own
segments after listing the prefix. A reference implementation exists and
follows this RFC in a series of pull requests; the document describes it
and marks what is proposed on top of it.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: link the object store WAL RFC to its tracking issue
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: describe the predecessor chain and instance epoch of the object store WAL
Recovery replays only the chain of objects that ends at the latest
complete object, and every open starts a new epoch with an empty start
object, so an object whose create was reported as failed is never
replayed once a later append has been acknowledged. Update the object
format, recovery, writing, the failure matrix, the garbage collection
constraints, compatibility, evidence, drawbacks and the cluster mode
alignment accordingly.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: identify object store WAL objects by sequence and epoch
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: derive the object store WAL epoch from the start object sequence
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: separate the retryable and the conflicting start object outcomes
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: describe garbage collection of the object store WAL through the reclaim boundary
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: judge vanished objects after chain selection and fix the evidence references
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: record the stop-time conflict and the completed fault matrix in the object store WAL RFC
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: limit the stop-time conflict to objects of the same or a later epoch
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: keep the conflicting-content condition in the stop-time conflict sentence
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: report the S3 write path measurements of the object store WAL
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: qualify the seal timer findings and link the S3 runs
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* docs: state the serial durable cycle behind the one-batch-per-interval observation
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
---------
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
The `wal.flush_interval` comment claimed that under sustained load the
batch always seals on `max_batch_bytes` and the interval no longer
matters. Sustained-load runs on S3 sealed on the timer in every case
because the admitted bytes did not reach `max_batch_bytes` within one
interval. State the actual condition: the batch seals on size only when
the admitted bytes reach `max_batch_bytes` before the next timer tick;
otherwise it seals on the timer, whatever the load.
Regenerated `config/config.md` from the example TOMLs.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* test(mito2): cover regex inverted index pruning
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* build: keep build inputs checkout-relative for cross-worktree cache reuse
Two small changes so compilations are byte-identical across worktrees
of the same commit and can share a build cache (mbx, sccache):
- src/common/version/build.rs: derive the workspace root relative to
CARGO_MANIFEST_DIR instead of the CARGO_WORKSPACE_DIR injected by
.cargo/config.toml (an absolute path). The env var remains as an
explicit override for the submodule case; the [env] injection in
.cargo/config.toml is removed so the relative derivation is used by
default.
- Replace three env!("CARGO_MANIFEST_DIR") uses in test/bench code
(servers prom_decode bench + decode test, common-meta test_util)
with a runtime std::env::var lookup, so the compiled artifact no
longer embeds the checkout path.
No behavior change: same files are read, same version info emitted.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* build: address cross-worktree cache review feedback
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
---------
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(log-store): drive recovery, read, delete and stall failures through the fault matrix
Add the second half of the object store WAL fault matrix: the failures of
the object store access and of the store state that the conditional-create
rows did not drive. Each case asserts the failed waiters and their error,
whether the store is poisoned, the sequence, both indexed gauges at the
failure, the exact movement of every counter and what a read sees after.
Recovery: a listing, a trailer fetch and a footer fetch that fail, the
catalog rejecting an object after every footer was read, and an object
gone by the time it is fetched, below the boundary of the tip (ignored)
and at it (construction fails); each leaves the gauges and the recovery
metrics untouched. Reads: a segment fetch that fails, without a hole or a
read timer sample, and a later read of the region that succeeds.
Collection: a delete that fails and is attempted again at the next
collection, and a stop that waits for a delete in flight, with the catalog
agreeing with the listing in both outcomes. Store state: per-region entry
id exhaustion refusing the append without touching the batch it sealed,
and the durable stop paths with several creates in flight that succeed,
fail transiently or conflict after stop began. Stalled appends released by
admission, by a stop, by a terminal error and by the teardown of the actor,
the first two timed as a pair. The recovery and read timers observed around
parked fetches, twice each, with holds that cannot overlap.
The cases use the existing doubles, extended with a failing listing and a
failing range read at an offset on RecordingIo and parked range reads on
ParkedIo. Case::row is now optional, and the assertion that every row of
the failure matrix has a case is unchanged.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* test(log-store): lose a wide object at its header and footer reads in the fault matrix
The vanished-object case now runs three ways: a short object read whole,
and a wide object deleted before its header and tail window reads or
before its separate footer read, each below and at the boundary of the
tip. The entry id exhaustion case reads the other region of the sealed
batch back, the teardown case asserts the indexed gauges before and after
the actor is gone, and two comments that restated their names are gone.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* test(log-store): prove the stop-path ordering and share the wide-footer fixture in the fault matrix
The case of creates that succeed after stop began keeps object 1 parked
while objects 4 to 2 complete, round-trips a command through the actor and
asserts that no append is acknowledged and nothing is indexed before
object 1 is released. The stop-with-delete case replaces its yields with
the same command barrier, queued behind the stop command. The wide-footer
fixture moves to the tests module and is shared with the recovery test
that fetches a footer longer than the tail window. The gauge helper is
named for its side effect and the vanished-object stage enum for what it
names.
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* test(log-store): keep the four-create scenario doc on its fault matrix case
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
---------
Signed-off-by: jeremyhi <fengjiachun@gmail.com>
* feat(jaeger): support v3 HTTP query APIs
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(jaeger): preserve full traces with v3 time hints
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* docs(jaeger): clarify validation of v3 lookup hints
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(jaeger): bound v3 searches and preserve v1 binary attributes
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(jaeger): match resource attributes in v0 tag filters
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
---------
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): follow IEEE 754 float semantics in operators and aggregations
- Comparisons: Arrow orders floats totally (NaN == NaN, NaN > +Inf); PromQL
comparisons with NaN are false except !=.
- min/max: skip NaN unless the whole group is NaN. DataFusion's grouped
min/max let NaN replace and be replaced by neighbours, so the result
depended on input order.
- topk/bottomk rank NaN last.
- round() rounds ties up and returns NaN for a zero step, like Prometheus.
- sqrt() of a negative number is NaN instead of an error.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): round() without a step rounds to integers
The planner passed a zero step for single-argument round(), which now
yields NaN like Prometheus; pass 1 instead. Found by the compatibility harness.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): apply the sqrt guard to builtin functions and register min/max for plan decoding
sqrt resolves through the session's builtin functions, not the UDF branch.
Distributed plans are decoded with the serializer's session, which did not
know prom_min/prom_max.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test(promql): cover NaN and rounding semantics
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): keep the nested count rewrite for PromQL min/max
count(min(x) by (l)) was rewritten by count_nest_aggr only for the builtin
min/max names. prom_min/prom_max return NULL for the same groups, so the
rewrite stays valid. Update plan snapshots for the IEEE comparison, sqrt,
topk and min/max changes.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): register prom_min/prom_max in the function registry
Distributed plans split an aggregation into state and merge steps only for
functions whose state function exists in FUNCTION_REGISTRY; prom_min and
prom_max lived in the promql crate, so min/max stopped being pushed down to
datanodes. Move them to common-function and register them before the state
functions are derived.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test: update sqlness results for PromQL float semantics
Plans and column names now show prom_min/prom_max, the NaN guards of
comparisons and sqrt, and the NaN sort key of topk. round() results follow
Prometheus' formula. Distributed min/max keep the state/merge split.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): keep column names and NULLs in IEEE comparisons and sqrt
Rewriting comparisons as `a op b AND NOT (isnan(a) OR isnan(b))` changed
TQL output column names and turned NULL operands into true for `!=` and
false for the bool forms. Implement comparisons and sqrt as PromQL UDFs
that propagate NULL and report the column name of the expression they
replace; prom_min/prom_max report min/max column names the same way.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test: restore column names in sqlness results for PromQL float semantics
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test(promql): share the prom_min/prom_max groups accumulator with tests
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): keep distinct names for prom_min/prom_max
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
---------
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): default a stepless subquery to a one-minute step
`x[5m:]` evaluated the inner expression at the query step, which is 1s
for instant queries. Prometheus uses the evaluation interval, one minute
by default.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test(promql): cover the default subquery step
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test(promql): cover the default subquery step in a range query
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
---------
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): return range-vector results for instant queries
An instant query of a range selector or subquery (e.g. `up[5m]`) returned
the internal range arrays, which panicked when the frontend decoded them as
dictionaries. Plan such queries as the selected samples with their own
timestamps, and reject range-vector expressions in range queries like
Prometheus does.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test(promql): keep encode_substrait on a valid range query
A range query of a range-vector expression now fails like in Prometheus.
Wrap the subquery in max_over_time so the case still encodes both
RangeManipulate nodes through substrait.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): plan wrapped range-vector roots and reject a zero subquery step
TQL EXPLAIN, ANALYZE and AS alias wrap the root in an extension, which the
range-vector path did not unwrap. A zero subquery step divided by zero.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test(promql): cover wrapped range-vector roots and a zero subquery step
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): default subquery step and reject range vectors in range queries
A root subquery without a step uses the one-minute default instead of
the query step. The HTTP range query endpoint rejects range-vector
expressions even when start equals end.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test: cover root subquery default step
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* refactor(promql): align root subqueries with subquery_child_window
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): return millisecond timestamps for range-vector results
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* refactor(promql): keep the time index timezone and name its table in errors
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
---------
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* feat(datanode): execute nested MergeScan plans on the datanode
Give a datanode the ability to decode and execute a plan whose MergeScan
carries a nested plan (e.g. an inner MergeScan over a build table), so a
distributed sub-plan pushed to a datanode can itself fan out to other
datanodes. This is the execution/codec capability only; it is not wired
to any user-facing switch and produces no plan by itself.
Codec (src/query/src/query_engine/default_serializer.rs):
- MergeScanAwareSerializer keeps the request's SessionState and the
engine catalog manager, and decodes a MergeScan payload's tables by
(catalog, schema, table) name through the engine catalog rather than
the region-aware request catalog list (which binds every table to the
request's region and breaks on mismatched columns).
Datanode (src/datanode):
- New DatanodeRegionQueryHandler (region_query.rs) serves region queries
for the MergeScan nodes of a received plan, resolving region leaders
via PartitionRuleManager and dispatching to peer datanodes.
- Wire KvBackendCatalogManager + PartitionRuleManager + NodeClients into
the datanode's catalog/cache registry (with_dist_planner stays false:
a datanode never rewrites its own plan).
Bug fixes uncovered while wiring this up:
- cache registry: register the table/table_info/table_name/table_route/
partition_info caches the datanode's catalog manager and partition
rule manager require (previously missing -> panic on lookup).
- default_serializer: decode_sub_plan bridge no longer drives a future
on a current-thread runtime from within another executor (spawn a
dedicated thread + runtime when not on a multi-thread runtime).
Tested by query_engine::default_serializer unit tests: nested MergeScan
round-trip and engine-catalog resolution with mismatched column names.
No user-facing behavior change.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(query): keep GreptimeDB function bindings after rebuilding decode state
SessionStateBuilder::new_from_existing collects the function registry
(HashMap) into a Vec and build() re-registers each entry; register_udf
registers aliases before the function's own name and the Vec order is
unstable. Rebuilding the session state to install MergeScanAwareSerializer
after the GreptimeDB functions were registered could therefore rebind a
name/alias (e.g. date_format, an alias of DataFusion's to_char) back to
the built-in function. This path runs for every decoded plan, not only
nested MergeScan ones.
Extract the GreptimeDB function registration into register_greptime_functions
and re-apply it after the final build of both the top-level decode state and
the MergeScan payload state. Add regression tests asserting date_format stays
bound to the GreptimeDB implementation on both paths.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(datanode): address review on nested MergeScan capability
Three follow-ups from review of the nested MergeScan execution capability:
1. Layer the datanode caches (P2b). table_cache derives from
table_info_cache + table_name_cache, and partition_info_cache derives
from table_route_cache, but they were all registered in a single flat
CacheRegistry whose invalidate() runs concurrently with no ordering.
A derived cache could be invalidated and refilled from a base cache
that had not been invalidated yet, leaving stale metadata. Build the
base caches and the derived caches as separate layers of a
LayeredCacheRegistry (reusing the existing layering mechanism) so base
caches finish invalidating before derived ones. Both the production
path (cmd) and the test scaffolding now build the same layered
registry; the flat build_datanode_cache_registry is removed.
2. Don't let a nested MergeScan's internal region query compete for the
concurrency permit (P2a). An outer query holds its permit until its
stream is fully consumed, and a nested MergeScan makes the datanode
issue region queries to peer datanodes as execution stages of that
same query. If such an inner stage also took a permit, the outer query
waiting on it would deadlock against its own held permit (or time out
when max_concurrent_queries > 0). Mark inner DN->DN region queries as
internal stages: QueryRequest gains an flag carried over
the wire in the request header's query_context extensions (reserved
key, not injectable via hints), and the region server skips acquiring
a concurrency permit for internal-stage requests on both acquire
sites. Outer queries still acquire normally.
3. Add an end-to-end capability test
(tests-integration/.../nested_merge_scan_capability_test.rs): hand-build
Join(<local probe region>, MergeScan(build regions on both datanodes))
with mismatched column names and duplicate build keys, ship it to a
datanode over a real TCP Flight endpoint, consume the stream fully and
compare the row multiset against the frontend's reference join result,
assert a real cross-datanode DoGet via an RPC counter, and assert an
inner region failure fails the whole query. Also migrate the
with_real_datanode_grpc_addr + DatanodeRpcStats test scaffolding.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(query): adapt date_format regression test to the updated DataFusion fork
ScalarUDFImpl's trait bounds changed so its impl can no longer be
downcast via as_any(); assert the function equals the GreptimeDB
implementation instead, which to_char can never satisfy.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(datanode): cover the internal stage under the concurrency limit
Add an end-to-end regression test for the nested MergeScan capability
under max_concurrent_queries = 1: both datanodes' only permit is held by
the outer queries' streams while their inner MergeScans fetch the build
regions from the same datanodes over real Flight connections. The result
consumption is bounded by a timeout, so a broken link anywhere in the
chain (marking the inner request internal, carrying the marker in the
request header, honoring it at the receiving acquire sites) fails the
test instead of deadlocking.
Also add a minimal with_datanode_options_override hook to the cluster
builder, and fix the test module doc: the outer request is handed to
RegionServer::handle_remote_read in process; only the inner region
queries go over the wire.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(datanode): cover cold route loads under unrelated invalidations
The table route cache uses InitStrategy::VersionChecked, whose version
counter belongs to the whole CacheContainer rather than to a single
table: any TableId invalidation bumps the shared version, so a cold
load of one table's route retries when an unrelated table is
invalidated concurrently. Document this as a known limitation on the
region query handler, and pin the behavior with an integration test:
run the nested plans while a background task invalidates an unrelated
table id on every datanode, assert the results still match the expected
multiset (retries only add work, they never return a stale route), and
report the observed load amplification via the cache miss counter
(2 loads without invalidations, 8 over three runs with 24 unrelated
invalidations in this run). Also expose the datanode cache registries
on the test cluster so a test can drive the local invalidation path
that the heartbeat handler drives after a metasrv broadcast.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* docs(query): trim nested plan decoder commentary
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* refactor: remove redundant nested scan support plumbing
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(query): decode nested MergeScan payloads asynchronously
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(query): avoid unnecessary extension name allocation
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(query): retain extension decoding for manual session states
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* refactor(datanode): keep nested scan query admission unchanged
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(datanode): honor query handler factories and trim nested scan fixtures
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(datanode): retain required cache error import
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* refactor(query): reuse registered state for nested scan decoding
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(query): verify payload information schema table functions
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(query): retain nested scan runtime and cold-route guards
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* refactor(query): address nested scan review feedback
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(session): preserve read preference across region query hops
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(session): verify strict context rejection at Flow ingress
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(query): remove redundant serializer error case
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
---------
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(operator): unify unknown flow option rejection across OSS and enterprise builds
Remove the enterprise-gated passthrough that let unknown flow options
through SQL-layer validation. Both OSS and enterprise binaries now reject
unknown flow options with a single 1004(InvalidArguments) error at the
SQL layer, producing identical sqlness output.
The metasrv-side validate_flow_options remains as a defensive check.
Signed-off-by: discord9 <discord9@163.com>
* fix(flow): use DISTINCT output columns as sink key for streaming flows
When a streaming flow's source table has no primary key, sink key
inference produced an empty key. For SELECT DISTINCT v, the DISTINCT
establishes v as the output key, but the inference only looked at
source primary keys. Combined with the zero-valued placeholder
timestamp, all output rows shared the same storage identity and were
deduplicated into a single row.
Use all non-time-index output columns as the sink key when the plan
is a DISTINCT and no source primary key was inherited.
Signed-off-by: discord9 <discord9@163.com>
* fix(flow): preserve extension options and complete DISTINCT sink identity
Keep enterprise CREATE FLOW options available to the existing handler and report unknown ordinary options as InvalidArguments with the OSS SQL error text. Infer DISTINCT sink keys from every selected non-time-index output column, including when source key columns survive. Wait for asynchronous mirrors in show_create_flow and exercise 1000 distinct sink rows in the SQL regression.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(flow): align meta unknown option status assertions
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(flow): use valid aliases in DISTINCT sink regression SQL
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* test(flow): record complete DISTINCT sink SQL results
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
---------
Signed-off-by: discord9 <discord9@163.com>
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(promql): accept evaluation times before the Unix epoch
Prometheus accepts negative timestamps in query parameters; GreptimeDB
rejected them while parsing and would panic converting them in the planner.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test: update flow_tql error for a negative TQL end time
A negative end now parses and fails the end >= start check instead of
the timestamp conversion.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): reject evaluation times outside the millisecond range
The signed millisecond conversion cast u128 to i64 unchecked, so a far-past
time wrapped into the future. Reuse the checked conversion from
label_values and return an error instead.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* chore: drop an unused import
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): round query times to milliseconds like Prometheus
Numeric times round the fraction of a second to milliseconds, and
pre-epoch times floor sub-millisecond parts, matching Prometheus'
parseTime and timestamp.FromTime.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): round the float fraction before converting to nanoseconds
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test: skip the overflow case where SystemTime cannot hold it
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
---------
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* perf(promql): zero-copy timestamp payload for millisecond zero-offset ranges
RangeManipulate rebuilt the timestamp range payload per sample through
wide-precision i128 conversion even for millisecond input with zero
offset, where the input is already millisecond and nothing changes.
That turned a previously zero-copy path into a per-sample loop plus a
Vec allocation, regressing range-function queries.
Reuse the input timestamp array directly when the time index is already
millisecond and the offset is zero. Other units and non-zero offsets
keep the wide-precision conversion (required so ns/us data near the
native type limits stays correct under large offsets).
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
Signed-off-by: discord9 <discord9@163.com>
* perf(promql): vectorize timestamp payload with plain i64
RangeManipulate rebuilt the timestamp range payload per sample through
wide-precision i128 conversion, turning a hot path into a scalar loop
with a Vec allocation. Realistic timestamps and offsets stay far inside
i64, so the wide path is unnecessary.
- millisecond + zero offset reuses the input array with zero copy;
- other units / offsets shift in one vectorized i64 pass (no i128, no
per-sample Option, no overflow branch).
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
Signed-off-by: discord9 <discord9@163.com>
* perf(promql): check timestamp payload overflow via arrow try_unary
Replace the silent wrapping shift with try_unary + checked_mul/checked_add.
checked arithmetic lowers to a per-lane overflow flag so the single pass
still vectorizes over the dense i64 buffer; any overflow short-circuits to
an error instead of silently wrapping to a wrong timestamp. No i128, no
per-sample Option.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
Signed-off-by: discord9 <discord9@163.com>
* perf(promql): vectorize cursor scans with i64 fast path
RangeManipulate::calculate_range and InstantManipulate::manipulate
compared every sample through i128 wide-precision arithmetic. For
realistic timestamps and offsets the shifted values stay far inside
i64, so the wide path was pure overhead.
Split both scans into an i64 fast path (checked arithmetic precomputed
once per batch) and an i128 exact fallback. Any input that would
overflow i64 falls back to the i128 scan, keeping window membership
semantics identical.
Verified: 47/47 range_manipulate + instant_manipulate tests, 275/275
promql crate tests, 1612 randomized cases byte-identical to i128 scan.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
Signed-off-by: discord9 <discord9@163.com>
* refactor(promql): generalize cursor scans and share endpoint precheck
Address review comments on PR #9417:
- Generalize scan_ranges/scan_instants over Nanoseconds (i64/i128),
eliminating the duplicated scan bodies.
- Extract shared narrow_scan precheck used by both RangeManipulate and
InstantManipulate, with identical checked endpoint logic.
- Trim doc comments to one line; fix factual errors about fallback
trigger (2262 for ms/s, not native type limits).
Semantics unchanged: same cursor algorithm, same overflow fallback,
same i64-vs-i128 selection, same stale/null/tie handling.
Verified: 47/47 range_manipulate + instant_manipulate tests, clippy clean.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
Signed-off-by: discord9 <discord9@163.com>
* perf(promql): vectorize payload with try_unary and add scan carrier bench
Replace the per-sample i128 payload loop with a single arrow try_unary
pass: i128 intermediate holds any native timestamp shift, only the final
millisecond result must fit i64. This is semantically equivalent to the
original i128 implementation (no silent wrap, no per-sample branching).
Add cursor_scan_carriers A/B benchmark: i64 fast path vs i128 exact
fallback on four workloads. Add five deterministic differential tests
pinning scan_ranges::<i64> and scan_ranges::<i128> to be pointwise
equivalent on boundary shapes (sub-ms samples, fallback triggers,
nanosecond extremes, seeded precision matrix).
Bench (N=4096, 15s cadence):
- payload wall_time 20x15s: 74.94us -> 48.98us (-35%)
- payload wall_time 240x15s: 102.44us
- scan i64 fast path: 10.97us vs i128 19.84us (1.81x)
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* perf(promql): restore zero-copy payload for ms+offset=0 and fix comment
Restore the zero-copy fast path for millisecond timestamps with zero
offset, guarded by the input column's exact data type (Timestamp
(Millisecond, None)) rather than self.time_unit. The old time_unit guard
would accept a tz-carrying millisecond column and produce a dictionary
value type disagreeing with the output schema; the datatype guard is the
correct check.
Also fix the comment: arrow's try_unary is a fallible kernel that LLVM
cannot auto-vectorize; the win over the hand-written loop comes from
eliminating the per-sample Option/Vec and branches, not SIMD.
Add a regression test with a timezone-carrying millisecond index to
prove it falls back to the try_unary path.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* perf(promql): replace try_unary with unary + Cell overflow flag
Arrow's try_unary is a fallible kernel that LLVM cannot auto-vectorize.
Replace it with the infallible unary kernel: the closure is a plain
i64 -> i64 mapping that sets a Cell<bool> on overflow instead of
returning Err per sample. The flag is checked once after the loop.
This removes the per-lane Result construction and lets LLVM vectorize
the shift+truncate sequence. Bench (N=4096, 15s cadence):
- window4x15s: 42.12us -> 28.77us (-32%)
- window20x15s: 48.98us -> 27.68us (-43%)
- window240x15s: 102.44us -> 27.61us (-73%)
- regular_sparse: 90.04us -> 29.20us (-68%)
All cases now run at ~27-29us regardless of window size; the zero-copy
path dominates for ms+offset=0.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* perf(promql): dispatch payload conversion on null_count; drop wall-time bench
Null-free batches (dominant PromQL case) take arrow's vectorizable unary
kernel with a Cell<bool> overflow flag; nullable batches take try_unary so
a NULL slot's arbitrary underlying value cannot raise a false batch-wide
overflow error. Add a regression test exercising both paths.
Remove bench_range_manipulate_wall_time (all-ms zero-offset cases only
exercised the zero-copy path, measuring nothing about the conversion) and
its orphaned helpers; restore the cursor-scan bench helpers it shared.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
* fix(promql): keep nullable timestamp cursor scans on i128
NULL slots may contain unordered arbitrary values, so the endpoint-only
narrow-scan proof cannot bound their interior timestamp arithmetic. Both
range and instant scans now attempt i64 only for null-free timestamp arrays
and otherwise retain the existing exact i128 path.
Add focused interior-NULL regressions through both active entrypoints,
checking concrete selections and output against the existing wide scan.
The null-free fast path and timestamp payload conversion are unchanged.
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
---------
Signed-off-by: discord9 <55937128+discord9@users.noreply.github.com>
Signed-off-by: discord9 <discord9@163.com>
* fix(promql): keep metric name for comparison operators
`collect_metric_names` dropped the response `__name__` for every binary
operator except `and` / `or` / `unless`, so comparisons such as
`node_load1 > 5` came back without the metric name. Grafana rendered them
as `{instance="..."}` and alert templates reading `$labels.__name__` got
nothing.
Follow Prometheus' `shouldDropMetricName` / `resultMetric`: arithmetic
operators drop the name, comparisons keep the left-hand name, except when
the comparison returns a bool or is a one-to-one `on(...)` match. A
`group_left` / `group_right` modifier does not reduce the labels, so the
name is kept there.
Both the HTTP PromQL API and the gRPC gateway share
`retrieve_metric_name_and_result_type`, so one fix covers both.
Signed-off-by: Aumnertic <hznuzc@gmail.com>
* docs: document binary_keeps_metric_name
Signed-off-by: Aumnertic <hznuzc@gmail.com>
* fix(promql): handle on(__name__), scalar lhs and group_right
Signed-off-by: Aumnertic <hznuzc@gmail.com>
* chore(promql): correct result-side docs and drop unreachable String arm
Signed-off-by: Aumnertic <hznuzc@gmail.com>
* fix(promql): drop metric name when explicitly ignored
Handle ignoring(__name__) for one-to-one comparisons and cover
comparison, group_left, group_right and unless behavior.
Signed-off-by: Aumnertic <hznuzc@gmail.com>
---------
Signed-off-by: Aumnertic <hznuzc@gmail.com>
* fix(meta-srv): stop parking the heartbeat response on the region supervisor queue
Signed-off-by: lczllx <2181719471@qq.com>
* test(meta-srv): cover the heartbeat response path at the handler level
Signed-off-by: lczllx <2181719471@qq.com>
---------
Signed-off-by: lczllx <2181719471@qq.com>
The duplicate-label-set rejection must not fire when a label rewrite keeps
every series distinct through another label. Normalizing mixed histogram
boundary representations (`10.0` -> `10`) while `instance` still differs is
the false positive reported in #9444. The check already accounts for it
since it became a per-timestamp row count instead of a planning-time
rejection, but no fixture covered the positive case. Add one, together
with the aggregation variant that produces a genuine duplicate and must
still fail.
Signed-off-by: fengyue-xve <1759519137@qq.com>
* fix(promql): keep vector labels when a computed scalar is on the right
`metric + time()` or `metric * scalar(other)` planned the join with the
scalar side's context, which has no tags, so every series collapsed into
one unlabeled series. Use the vector side's context when the right operand
is scalar-typed.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test(promql): update results for labels kept with a computed scalar on the right
v * pi() and v + time() now keep the vector's labels and time index name,
as in Prometheus.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* fix(promql): qualify columns of an unnamed operand in vector joins
A comparison filter's projection has no table name, so after the join its
time index resolved unqualified and was ambiguous when the other operand had
the same column. Also keep the empty-metric branch unchanged for scalar
left operands.
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
* test(promql): cover comparison filters joined with a scalar
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
---------
Signed-off-by: Dennis Zhuang <killme2008@gmail.com>
WenyXu
Lei, HUANG
jeremyhi
Yingwen
liyang
shuiyisong
discord9
dennis zhuang
sun
fys
Dragon Roar
Aum
Niladri Adhikary
fengyue-xve