Fix loss of root/meta blkno after self-invalidation; defer invalidation processing under page locks#55
Open
Serge-sudo wants to merge 104 commits into
Open
Conversation
Discussion: https://postgr.es/m/CAPpHfdua-YFw3XTprfutzGp28xXLigFtzNbuFY8yPhqeq6X5kg%40mail.gmail.com Reviewed-by: Aleksander Alekseev, Pavel Borisov, Vignesh C, Mason Sharp Reviewed-by: Andres Freund, Chris Travers
Author: Andres Freund Reviewed-by: Pavel Borisov Discussion: https://www.postgresql.org/message-id/flat/CAPpHfdua-YFw3XTprfutzGp28xXLigFtzNbuFY8yPhqeq6X5kg%40mail.gmail.com
Snapshot have two pairing heap nodes: for data and system undos.
* Added SearchCatCacheInternal_hook, SearchCatCacheList_hook * Added SysCacheGetAttr_hook
IsFatalError() have_backup_in_progress() SnapBuildNextPhaseAt() DoLocalLockExist()
Outline-atomics is a gcc compilation flag that enables runtime detection of CPU support for atomic instructions. Performance on CPUs that do support atomic instructions is improved, while compatibility and performance on CPUs without atomic instructions is not hurt. Discussion: https://postgr.es/m/flat/099F69EE-51D3-4214-934A-1F28C0A1A7A7%40amazon.com Author: Tsahi Zidenberg
They are allowed to stay during shutdown checkpointing and help checkpointer do its work.
To use curl during shared_preload_libraries initialization.
- added option --extension for pg_rewind - extracted SimpleXLogRead from extractPageMap for generic wal iteration in pg_rewind
pg_column_toast_chunk_id() is tied to PG TOAST pointer format and can't work with OrioledDB TOAST pointer. So make it safely return NULL in this case. PG17+ only Fixes orioledb/orioledb#690
Instead of interperting all the extension wait events as process blocker, only recognize "StopEvent" by its name.
Use tupleid for ROW_REF_TID as in upsteam. Use &context->tmfd.ctid for ROW_REF_ROWID as in OrioleDB.
…o_unset_syscache_hooks call Single-user mode may run recovery and then start a session in the same process. When OrioleDB is loaded, recovery initializes the catalog cache, which caused session startup to fail because it expected a non-initialized state. Allow the catalog cache to be already initialized when starting a session in single-user mode after recovery. Additionally, fix a missing o_unset_syscache_hooks call, which left OrioleDB syscache hooks installed across the recovery/session boundary and contributed to crashes and inconsistent backend state. This ensures that backend-global state is properly handled when recovery and normal session execution happen within the same process.
The latest actions/checkout@v6 has improved credential security.
Also remove some changes for pg_regress and regress tests
Add extra argument indicating commit/abort.
It's needed to count custom extension data for database that is stored in PG datadir or pg_tblspc dir but outside of <dbOid> dir. Size counted by a hook is added to database size counted by existing method
Serge-sudo
force-pushed
the
fixed-invalid-meta-blocknum
branch
from
82c6a25 to
b8d017d
Compare
…on processing under page locks Fix an issue where B-tree descriptor state (`metaPageBlkno` and `rootPageBlkno`) could be lost after processing an invalidation message within the same transaction. Previously, after `o_btree_load_shmem()` initialized the descriptor, a invalidation message could be processed before the descriptor was used. This caused the descriptor to be recreated, resetting `metaPageBlkno` and `rootPageBlkno` to `InvalidBlockNumber`, which led to assertion failures (e.g. in `btree_ctid_get_and_inc`). While investigating, another problem was identified: invalidation message processing may occur while holding page locks. Invalidation handlers access system trees, and if required pages are evicted, they trigger page loads. This leads to `load_page()` being called while locks are held, hitting assertions that forbid holding locks during page load. To address both issues: * Ensure descriptor state is not lost due to mid-operation self-invalidation * Introduce deferred invalidation handling: * Skip invalidation message processing when it is unsafe (e.g. while holding page locks) * Process pending invalidation messages at the next safe point This prevents descriptor corruption and avoids unsafe page loads under locks.
Serge-sudo
force-pushed
the
fixed-invalid-meta-blocknum
branch
from
b8d017d to
2c21d4d
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
11 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fix an issue where B-tree descriptor state (
metaPageBlknoandrootPageBlkno) could be lost after processing an invalidation message within the same transaction.Previously, after
o_btree_load_shmem()initialized the descriptor, a invalidation message could be processed before the descriptor was used. This caused the descriptor to be recreated, resettingmetaPageBlknoandrootPageBlknotoInvalidBlockNumber, which led to assertion failures (e.g. inbtree_ctid_get_and_inc).While investigating, another problem was identified: invalidation message processing may occur while holding page locks. Invalidation handlers access system trees, and if required pages are evicted, they trigger page loads. This leads to
load_page()being called while locks are held, hitting assertions that forbid holding locks during page load.To address both issues:
Ensure descriptor state is not lost due to mid-operation self-invalidation
Introduce deferred invalidation handling:
This prevents descriptor corruption and avoids unsafe page loads under locks.
downlink -> orioledb/orioledb#792