Movatterモバイル変換

[0]ホーム
URL:

Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly

 Sign up
Appearance settings

Commitd41a178

Browse files
committed
Fix waitpid() emulation on Windows.
Our waitpid() emulation didn't prevent a PID from being recycled by theOS before the call to waitpid(). The postmaster could finish uptracking more than one child process with the same PID, and confusethem.Fix, by moving the guts of pgwin32_deadchild_callback() into waitpid(),so that resources are released synchronously. The process and PIDcontinue to exist until we close the process handle, which only happensonce we're ready to adjust our book-keeping of running children.This seems to explain a couple of failures on CI. It had never beenreported before, despite the code being as old as the Windows port.Perhaps Windows started recycling PIDs more rapidly, or perhaps timingchanges due to commit7389aad made it more likely to break.Thanks to Alexander Lakhin for analysis and Andres Freund for trackingdown the root cause.Back-patch to all supported branches.Reported-by: Andres Freund <andres@anarazel.de>Discussion:https://postgr.es/m/20230208012852.bvkn2am4h4iqjogq%40awork3.anarazel.de
1 parentb081fe4 commitd41a178

File tree

1 file changed

+40
-30
lines changed

1 file changed

+40
-30
lines changed

‎src/backend/postmaster/postmaster.c

Lines changed: 40 additions & 30 deletions
Original file line numberDiff line numberDiff line change
@@ -4811,7 +4811,7 @@ internal_forkexec(int argc, char *argv[], Port *port)
48114811
(errmsg_internal("could not register process for wait: error code %lu",
48124812
GetLastError())));
48134813

4814-
/* Don't close pi.hProcess here -the wait thread needs access to it */
4814+
/* Don't close pi.hProcess here -waitpid() needs access to it */
48154815

48164816
CloseHandle(pi.hThread);
48174817

@@ -6421,36 +6421,21 @@ ShmemBackendArrayRemove(Backend *bn)
64216421
staticpid_t
64226422
waitpid(pid_tpid,int*exitstatus,intoptions)
64236423
{
6424+
win32_deadchild_waitinfo*childinfo;
6425+
DWORDexitcode;
64246426
DWORDdwd;
64256427
ULONG_PTRkey;
64266428
OVERLAPPED*ovl;
64276429

6428-
/*
6429-
* Check if there are any dead children. If there are, return the pid of
6430-
* the first one that died.
6431-
*/
6432-
if (GetQueuedCompletionStatus(win32ChildQueue,&dwd,&key,&ovl,0))
6430+
/* Try to consume one win32_deadchild_waitinfo from the queue. */
6431+
if (!GetQueuedCompletionStatus(win32ChildQueue,&dwd,&key,&ovl,0))
64336432
{
6434-
*exitstatus=(int)key;
6435-
returndwd;
6433+
errno=EAGAIN;
6434+
return-1;
64366435
}
64376436

6438-
return-1;
6439-
}
6440-
6441-
/*
6442-
* Note! Code below executes on a thread pool! All operations must
6443-
* be thread safe! Note that elog() and friends must *not* be used.
6444-
*/
6445-
staticvoidWINAPI
6446-
pgwin32_deadchild_callback(PVOIDlpParameter,BOOLEANTimerOrWaitFired)
6447-
{
6448-
win32_deadchild_waitinfo*childinfo= (win32_deadchild_waitinfo*)lpParameter;
6449-
DWORDexitcode;
6450-
6451-
if (TimerOrWaitFired)
6452-
return;/* timeout. Should never happen, since we use
6453-
* INFINITE as timeout value. */
6437+
childinfo= (win32_deadchild_waitinfo*)key;
6438+
pid=childinfo->procId;
64546439

64556440
/*
64566441
* Remove handle from wait - required even though it's set to wait only
@@ -6466,13 +6451,11 @@ pgwin32_deadchild_callback(PVOID lpParameter, BOOLEAN TimerOrWaitFired)
64666451
write_stderr("could not read exit code for process\n");
64676452
exitcode=255;
64686453
}
6469-
6470-
if (!PostQueuedCompletionStatus(win32ChildQueue,childinfo->procId, (ULONG_PTR)exitcode,NULL))
6471-
write_stderr("could not post child completion status\n");
6454+
*exitstatus=exitcode;
64726455

64736456
/*
6474-
*Handle is per-process, so we close it here instead of inthe
6475-
*originating thread
6457+
*Close theprocess handle. Only after this point canthe PID can be
6458+
*recycled by the kernel.
64766459
*/
64776460
CloseHandle(childinfo->procHandle);
64786461

@@ -6482,7 +6465,34 @@ pgwin32_deadchild_callback(PVOID lpParameter, BOOLEAN TimerOrWaitFired)
64826465
*/
64836466
free(childinfo);
64846467

6485-
/* Queue SIGCHLD signal */
6468+
returnpid;
6469+
}
6470+
6471+
/*
6472+
* Note! Code below executes on a thread pool! All operations must
6473+
* be thread safe! Note that elog() and friends must *not* be used.
6474+
*/
6475+
staticvoidWINAPI
6476+
pgwin32_deadchild_callback(PVOIDlpParameter,BOOLEANTimerOrWaitFired)
6477+
{
6478+
/* Should never happen, since we use INFINITE as timeout value. */
6479+
if (TimerOrWaitFired)
6480+
return;
6481+
6482+
/*
6483+
* Post the win32_deadchild_waitinfo object for waitpid() to deal with. If
6484+
* that fails, we leak the object, but we also leak a whole process and
6485+
* get into an unrecoverable state, so there's not much point in worrying
6486+
* about that. We'd like to panic, but we can't use that infrastructure
6487+
* from this thread.
6488+
*/
6489+
if (!PostQueuedCompletionStatus(win32ChildQueue,
6490+
0,
6491+
(ULONG_PTR)lpParameter,
6492+
NULL))
6493+
write_stderr("could not post child completion status\n");
6494+
6495+
/* Queue SIGCHLD signal. */
64866496
pg_queue_signal(SIGCHLD);
64876497
}
64886498
#endif/* WIN32 */

0 commit comments

Comments
 (0)
[8]ページ先頭
©2009-2025 Movatter.jp