[Bug] Parallel dedup semi-join planning takes an exclusive lock on gp_segment_configuration
I maintainer di solito rispondono entro 2 giorni
Nessuno ha ancora preso questa issue.
Valutazione
- Difficoltà
- 4/5
- Tempo stimato
- 3-5 giorni
- Idoneità per principianti
- 53/100
Direzione di ricerca
Start with the parallel semi-join regression fixture in src/test/regress/sql/cbdb_parallel.sql and trace the two planner paths in src/backend/cdb/cdbpath.c to cdbcomponent_get_maxdbid() in src/backend/cdb/cdbutil.c. Check how the catalog is opened and closed, and identify a read-only lookup that preserves locking for topology-changing callers. Verify with EXPLAIN and the fixture that planning no longer holds an AccessExclusiveLock while topology changes retain their required locking.
Scritto dal modello di indicizzazione a partire dal testo della issue.
Descrizione
Apache Cloudberry version
Apache Cloudberry 2.2.0-incubating, build 13d3c58eaa0 (PostgreSQL 14.10)
What happened
While diagnosing intermittent connection delays on a multi-segment cluster, we observed ordinary read-only SELECTs holding a granted AccessExclusiveLock on pg_catalog.gp_segment_configuration. New connections were waiting for AccessShareLock on the same shared catalog. pg_blocking_pids() identified the SELECT backend as the blocker, including in a different database on the same coordinator.
On a separate three-segment development cluster running the same build, EXPLAIN without ANALYZE alone acquired the lock. In the on/off comparison below, the final plan was Parallel Hash Semi Join in both cases. The lock can therefore come from evaluating a candidate that is never executed.
What you think should happen instead
Planning a read-only query should not hold a transaction-duration AccessExclusiveLock on the shared topology catalog. The planner only needs to read the maximum segment DBID. Please preserve the locking semantics required by topology-changing callers.
How to reproduce
On the affected build in a development environment, use the parallel semi-join regression fixture as a starting point. We compared EXPLAIN (without ANALYZE) on an equivalent EXISTS query joining two distributed tables, with optimizer=off, enable_parallel=on, and two workers per gather.
| Planner settings | Current backend’s lock on gp_segment_configuration after EXPLAIN |
|---|---|
| Both parallel dedup semi-join options enabled | Granted AccessExclusiveLock |
enable_parallel_dedup_semi_join=off and enable_parallel_dedup_semi_reverse_join=off |
No lock observed |
We inspected pg_locks in each read-only test transaction and rolled back immediately after taking the sample.
Operating System
Enterprise Linux 9, x86_64
Anything else
The two parallel join paths in cdbpath.c call cdbcomponent_get_maxdbid() when the deduplicated input uses multiple parallel workers. The catalog implementation opens the relation with AccessExclusiveLock and closes it with NoLock, keeping the lock until transaction end.
A possible fix is to give the planner a separate read-only lookup with appropriately scoped locking. Turning off both dedup options at session scope avoided this path on the tested build while leaving ordinary parallel semi-joins available. We verified the behavior on 2.2.0 only.
Are you willing to submit PR?
- Yes, I am willing to submit a PR!
Code of Conduct
- I agree to follow this project's Code of Conduct.
- Lingua principale
- C
- Stelle
- 1.4k
- Fork
- 273
- Merge medio
- 4g 6h
- PR unite (30g)
- 105
Preparare l'ambiente
- Nessun Dockerfile né file Docker Compose
- Ha un modello di pull request
- Leggi la guida per i contributori
Come iniziare
- Leggi tutta la issue e poi la guida ai contributi del progetto.
- Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
- Fai un fork del repository e lavora su un branch.
- Apri una pull request che faccia riferimento al numero della issue.
Altre issue di apache/cloudberry
-
isolation2 lockmodes fails on cassert builds: extra AccessShareLock on partition rootForse già presa @tuhaihe l’ha presa oggi. Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 78/100
apache/cloudberry#2356 ·
I maintainer di solito rispondono entro 2 giorni
-
psql tab completion broken for storage servers, tags, profiles and resource groups (pre-PG15 query format)Forse già presa @allenvox l’ha presa 5 giorni fa. Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 72/100
apache/cloudberry#2316 ·
I maintainer di solito rispondono entro 2 giorni
-
[Bug] Expand fails to recognize user-defined tablespacesForse già presa @nix-oss l’ha presa 92 giorni fa. Apertatype: Bug
Difficoltà 2/5 1-3 ore Idoneità per principianti 76/100
apache/cloudberry#1885 · 2 reazioni ·
I maintainer di solito rispondono entro 2 giorni
-
gpexpand first stage fails: 'SyncPackages' object has no attribute 'ret' (extension sync never runs; regression)Forse già presa @talmacschen-arch l’ha presa 153 giorni fa. Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 86/100
apache/cloudberry#1825 ·
I maintainer di solito rispondono entro 2 giorni
-
Difficoltà 5/5 Più di una settimana Idoneità per principianti 15/100
apache/cloudberry#2348 · 1 commento ·
I maintainer di solito rispondono entro 2 giorni
Tutte le issue di apache/cloudberry
Issue simili
-
Difficoltà 2/5 1-3 ore Idoneità per principianti 78/100
I maintainer di solito rispondono entro 1 giorno
-
windows-artifacts: the win-arm64 job uses a Visual Studio generator missing from windows-11-armAperta
Difficoltà 1/5 Meno di un'ora Idoneità per principianti 86/100
-
Add Royalty funding badgeAperta
Difficoltà 1/5 1-3 ore Idoneità per principianti 78/100
crosspoint-reader/crosspoint-reader#4535 ·
I maintainer di solito rispondono entro 1 giorno
-
[core] Clients never advertise RNS_UD_CS_SUPPORT_ERRINFO_PDU, so a conforming server can't send them Set Error InfoForse già presa @akallabeth l’ha presa oggi. Aperta
Difficoltà 2/5 1-3 ore Idoneità per principianti 75/100
FreeRDP/FreeRDP#14230 · 1 commento ·
I maintainer di solito rispondono entro 1 giorno
-
Wrong release linkedAperta
Difficoltà 1/5 Meno di un'ora Idoneità per principianti 88/100
Genymobile/scrcpy#7178 ·
I maintainer di solito rispondono entro 3 giorni