BUG #17591: elog(ERROR) cause SharedSnapshotLock deadlock
The following bug has been logged on the website:
Bug reference: 17591
Logged by: ma liangzhu
Email address: ma100@hotmail.com
PostgreSQL version: 14.5
Operating system: centos7
Description:
in lock.c, we can see code Release lock before return , e.g.
```c
996: LockAcquireExtended
{
LWLockAcquire(partitionLock, LW_EXCLUSIVE);
proclock = SetupLockInTable( );
if (!proclock)
{
LWLockRelease(partitionLock); --- Release lock before return
ereport(ERROR,)
}
}
```
bug we can see some code doesn't release the lock. Does it may cause
deadlock?
```c
LockRelease()
{
LWLockAcquire(partitionLock, LW_EXCLUSIVE);
lock = locallock->lock;
if (!lock)
{
lock = (LOCK *) hash_search_with_hash_value( );
if (!lock)
2126: elog(ERROR, "failed to re-find shared lock object"); -- exit
without release
}
```
Hi,
Didn’t see any bugs here, it report an error and the transaction will end and release all resources.
Regards,
Zhang Mingli
Show quoted text
On Aug 22, 2022, 14:58 +0800, PG Bug reporting form <noreply@postgresql.org>, wrote:
The following bug has been logged on the website:
Bug reference: 17591
Logged by: ma liangzhu
Email address: ma100@hotmail.com
PostgreSQL version: 14.5
Operating system: centos7
Description:in lock.c, we can see code Release lock before return , e.g.
```c
996: LockAcquireExtended
{
LWLockAcquire(partitionLock, LW_EXCLUSIVE);proclock = SetupLockInTable( );
if (!proclock)
{
LWLockRelease(partitionLock); --- Release lock before return
ereport(ERROR,)
}
}
```bug we can see some code doesn't release the lock. Does it may cause
deadlock?```c
LockRelease()
{
LWLockAcquire(partitionLock, LW_EXCLUSIVE);lock = locallock->lock;
if (!lock)
{
lock = (LOCK *) hash_search_with_hash_value( );
if (!lock)
2126: elog(ERROR, "failed to re-find shared lock object"); -- exit
without release
}
```
PG Bug reporting form <noreply@postgresql.org> writes:
in lock.c, we can see code Release lock before return , e.g.
...
bug we can see some code doesn't release the lock. Does it may cause
deadlock?
Transaction cleanup will automatically release all locks (and other
resources too). So the places that do a manual release before
elog(ERROR) are something of a waste of code space. It might be worth
doing that if you're concerned about holding the lock for the minimum
possible amount of time; but usually that's only a consideration for
very heavily-contended spinlocks or LWLocks.
regards, tom lane