Debian Patches
Status for pg-ivm/1.15-2
| Patch | Description | Author | Forwarded | Bugs | Origin | Last update |
|---|---|---|---|---|---|---|
| 173.patch | Improve stability of SERIALIZABLE isolation tests When getLastUpdateXid() uses a latest catalog snapshot under SERIALIZABLE, whether an RW-conflict is recorded may depend on the HOT tuple layout. Previously, the function stopped after finding the expected tuple, so it could miss an invisible tuple that would trigger an RW-conflict check. Modified the function to call systable_getnext() once more after finding the expected tuple and confirm that no additional tuples are returned. This causes any remaining tuples to be checked for visibility, reducing the dependency on the HOT tuple layout and improving the stability of the isolation tests. |
Yugo Nagata <nagata@sraoss.co.jp> | no | 2026-07-24 | ||
| 163.patch | Fix "could not obtain lock" error In REPEATABLE READ / SERIALIZABLE, IVM_immediate_before() took the matview lock with a no-wait ConditionalLockRelationOid() and errored immediately when it was busy. But that lock is also held briefly by create_immv()/refresh_immv(), and commit releases locks one at a time, so a waiter can reach the matview lock before the committer has released it and fail with a spurious "could not obtain lock" error -- with no concurrent maintenance actually in progress. This made create_insert2/create_insert3 flaky under load. Just wait for the lock, as READ COMMITTED already did. Real anomalies are still caught by the existing last-update XID check right after, which is independent of how the lock was acquired. Isolation test expected outputs are updated accordingly. Fixes #153 |
Dzmitry Levankou <dzmitry.levankou@snowflake.com> | no | 2026-06-10 |
All known versions for source package 'pg-ivm'
- 1.15-2 (sid, forky)