[Bug] Parallel dedup semi-join planning takes an exclusive lock on gp_segment_configuration
Los mantenedores suelen responder en 2 días
Nadie ha tomado este issue todavía.
Evaluación
- Dificultad
- 4/5
- Tiempo estimado
- 3-5 días
- Aptitud para principiantes
- 53/100
Línea de trabajo
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.
Escrito por el modelo de indexación a partir del texto del issue.
Descripción
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.
- Lenguaje dominante
- C
- Estrellas
- 1.4k
- Forks
- 273
- Merge medio
- 4 d 6 h
- PR fusionados (30 d)
- 105
Preparar el entorno
- Sin Dockerfile ni archivo de Docker Compose
- Tiene una plantilla de pull request
- Leer la guía de contribución
Primeros pasos
- Lee el issue completo y luego la guía de contribución del proyecto.
- Comenta en el issue que vas a ocuparte — evita que dos personas hagan lo mismo.
- Haz un fork del repositorio y trabaja en una rama.
- Abre un pull request que haga referencia al número del issue.
Más de apache/cloudberry
-
isolation2 lockmodes fails on cassert builds: extra AccessShareLock on partition rootPosiblemente ocupada @tuhaihe la tomó hoy. Abierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 78/100
apache/cloudberry#2356 ·
Los mantenedores suelen responder en 2 días
-
psql tab completion broken for storage servers, tags, profiles and resource groups (pre-PG15 query format)Posiblemente ocupada @allenvox la tomó hace 5 días. Abierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 72/100
apache/cloudberry#2316 ·
Los mantenedores suelen responder en 2 días
-
[Bug] Expand fails to recognize user-defined tablespacesPosiblemente ocupada @nix-oss la tomó hace 92 días. Abiertotype: Bug
Dificultad 2/5 1-3 horas Aptitud para principiantes 76/100
apache/cloudberry#1885 · 2 reacciones ·
Los mantenedores suelen responder en 2 días
-
gpexpand first stage fails: 'SyncPackages' object has no attribute 'ret' (extension sync never runs; regression)Posiblemente ocupada @talmacschen-arch la tomó hace 153 días. Abierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 86/100
apache/cloudberry#1825 ·
Los mantenedores suelen responder en 2 días
-
Terminated by signal 6: AbortedAbierto
Dificultad 5/5 Más de una semana Aptitud para principiantes 15/100
apache/cloudberry#2348 · 1 comentario ·
Los mantenedores suelen responder en 2 días
Todos los issues de apache/cloudberry
Issues similares
-
Dificultad 2/5 1-3 horas Aptitud para principiantes 78/100
Los mantenedores suelen responder en 1 día
-
windows-artifacts: the win-arm64 job uses a Visual Studio generator missing from windows-11-armAbierto
Dificultad 1/5 Menos de una hora Aptitud para principiantes 86/100
-
Add Royalty funding badgeAbierto
Dificultad 1/5 1-3 horas Aptitud para principiantes 78/100
crosspoint-reader/crosspoint-reader#4535 ·
Los mantenedores suelen responder en 1 día
-
[core] Clients never advertise RNS_UD_CS_SUPPORT_ERRINFO_PDU, so a conforming server can't send them Set Error InfoPosiblemente ocupada @akallabeth la tomó hoy. Abierto
Dificultad 2/5 1-3 horas Aptitud para principiantes 75/100
FreeRDP/FreeRDP#14230 · 1 comentario ·
Los mantenedores suelen responder en 1 día
-
Wrong release linkedAbierto
Dificultad 1/5 Menos de una hora Aptitud para principiantes 88/100
Genymobile/scrcpy#7178 ·
Los mantenedores suelen responder en 3 días