Hi,
I just noticed this while reviewing patch [1]. It looks like this is caused by
a simple typo.
In ProcWakeup():
```
PGPROC *
ProcWakeup(PGPROC *proc, ProcWaitStatus waitStatus)
{
PGPROC *retProc;
/* Proc should be sleeping ... */
if (proc->links.prev == NULL ||
proc->links.next == NULL)
return NULL;
Assert(proc->waitStatus == PROC_WAIT_STATUS_WAITING);
/* Save next process before we zap the list link */
retProc = (PGPROC *) proc->links.next;
/* Remove process from wait queue */
SHMQueueDelete(&(proc->links));
(proc->waitLock->waitProcs.size)--;
/* Clean up process' state and pass it the ok/fail signal */
proc->waitLock = NULL;
proc->waitProcLock = NULL;
proc->waitStatus = waitStatus;
pg_atomic_write_u64(&MyProc->waitStart, 0); <== Here, it should operate
on proc
/* And awaken it */
SetLatch(&proc->procLatch);
return retProc;
}
```
Since this function is clearly operating on the parameter proc, the only
statement that touches MyProc looks suspicious. It should reset
proc->waitStart, not MyProc->waitStart.
As written, it will incorrectly clear the caller’s waitStart instead of the
awakened backend’s, potentially leaving a stale waitStart value in the target
PGPROC.
[1]
https://postgr.es/m/cahgqgwgw4lhnwogqt3nbw3uwy8gl94_mb4t39wfr4_vgopu...@mail.gmail.com
Best regards,
--
Chao Li (Evan)
HighGo Software Co., Ltd.
https://www.highgo.com/
