Make XactLockTableWait() and ConditionalXactLockTableWait() interruptable more.
authorFujii Masao <fujii@postgresql.org>
Fri, 30 May 2025 15:08:40 +0000 (00:08 +0900)
committerFujii Masao <fujii@postgresql.org>
Fri, 30 May 2025 15:08:40 +0000 (00:08 +0900)
commit961553daf5d6087b175aa98f3031a46a8666cecf
tree8edde55eba9f952398eebfa04c9713210ad2f593
parentc3eda50b0648005281c2a3cf95375708f8ef97fc
Make XactLockTableWait() and ConditionalXactLockTableWait() interruptable more.

Previously, XactLockTableWait() and ConditionalXactLockTableWait() could enter
a non-interruptible loop when they successfully acquired a lock on a transaction
but the transaction still appeared to be running. Since this loop continued
until the transaction completed, it could result in long, uninterruptible waits.

Although this scenario is generally unlikely since XactLockTableWait() and
ConditionalXactLockTableWait() can basically acquire a transaction lock
only when the transaction is not running, it can occur in a hot standby.
In such cases, the transaction may still appear active due to
the KnownAssignedXids list, even while no lock on the transaction exists.
For example, this situation can happen when creating a logical replication
slot on a standby.

The cause of the non-interruptible loop was the absence of CHECK_FOR_INTERRUPTS()
within it. This commit adds CHECK_FOR_INTERRUPTS() to the loop in both functions,
ensuring they can be interrupted safely.

Back-patch to all supported branches.

Author: Kevin K Biju <kevinkbiju@gmail.com>
Reviewed-by: Fujii Masao <masao.fujii@gmail.com>
Discussion: https://postgr.es/m/CAM45KeELdjhS-rGuvN=ZLJ_asvZACucZ9LZWVzH7bGcD12DDwg@mail.gmail.com
Backpatch-through: 13
src/backend/storage/lmgr/lmgr.c