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

Commit9f1c640

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 parent386a260 commit9f1c640

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
@@ -4901,7 +4901,7 @@ internal_forkexec(int argc, char *argv[], Port *port)
49014901
(errmsg_internal("could not register process for wait: error code %lu",
49024902
GetLastError())));
49034903

4904-
/* Don't close pi.hProcess here -the wait thread needs access to it */
4904+
/* Don't close pi.hProcess here -waitpid() needs access to it */
49054905

49064906
CloseHandle(pi.hThread);
49074907

@@ -6561,36 +6561,21 @@ ShmemBackendArrayRemove(Backend *bn)
65616561
staticpid_t
65626562
waitpid(pid_tpid,int*exitstatus,intoptions)
65636563
{
6564+
win32_deadchild_waitinfo*childinfo;
6565+
DWORDexitcode;
65646566
DWORDdwd;
65656567
ULONG_PTRkey;
65666568
OVERLAPPED*ovl;
65676569

6568-
/*
6569-
* Check if there are any dead children. If there are, return the pid of
6570-
* the first one that died.
6571-
*/
6572-
if (GetQueuedCompletionStatus(win32ChildQueue,&dwd,&key,&ovl,0))
6570+
/* Try to consume one win32_deadchild_waitinfo from the queue. */
6571+
if (!GetQueuedCompletionStatus(win32ChildQueue,&dwd,&key,&ovl,0))
65736572
{
6574-
*exitstatus=(int)key;
6575-
returndwd;
6573+
errno=EAGAIN;
6574+
return-1;
65766575
}
65776576

6578-
return-1;
6579-
}
6580-
6581-
/*
6582-
* Note! Code below executes on a thread pool! All operations must
6583-
* be thread safe! Note that elog() and friends must *not* be used.
6584-
*/
6585-
staticvoidWINAPI
6586-
pgwin32_deadchild_callback(PVOIDlpParameter,BOOLEANTimerOrWaitFired)
6587-
{
6588-
win32_deadchild_waitinfo*childinfo= (win32_deadchild_waitinfo*)lpParameter;
6589-
DWORDexitcode;
6590-
6591-
if (TimerOrWaitFired)
6592-
return;/* timeout. Should never happen, since we use
6593-
* INFINITE as timeout value. */
6577+
childinfo= (win32_deadchild_waitinfo*)key;
6578+
pid=childinfo->procId;
65946579

65956580
/*
65966581
* Remove handle from wait - required even though it's set to wait only
@@ -6606,13 +6591,11 @@ pgwin32_deadchild_callback(PVOID lpParameter, BOOLEAN TimerOrWaitFired)
66066591
write_stderr("could not read exit code for process\n");
66076592
exitcode=255;
66086593
}
6609-
6610-
if (!PostQueuedCompletionStatus(win32ChildQueue,childinfo->procId, (ULONG_PTR)exitcode,NULL))
6611-
write_stderr("could not post child completion status\n");
6594+
*exitstatus=exitcode;
66126595

66136596
/*
6614-
*Handle is per-process, so we close it here instead of inthe
6615-
*originating thread
6597+
*Close theprocess handle. Only after this point canthe PID can be
6598+
*recycled by the kernel.
66166599
*/
66176600
CloseHandle(childinfo->procHandle);
66186601

@@ -6622,7 +6605,34 @@ pgwin32_deadchild_callback(PVOID lpParameter, BOOLEAN TimerOrWaitFired)
66226605
*/
66236606
free(childinfo);
66246607

6625-
/* Queue SIGCHLD signal */
6608+
returnpid;
6609+
}
6610+
6611+
/*
6612+
* Note! Code below executes on a thread pool! All operations must
6613+
* be thread safe! Note that elog() and friends must *not* be used.
6614+
*/
6615+
staticvoidWINAPI
6616+
pgwin32_deadchild_callback(PVOIDlpParameter,BOOLEANTimerOrWaitFired)
6617+
{
6618+
/* Should never happen, since we use INFINITE as timeout value. */
6619+
if (TimerOrWaitFired)
6620+
return;
6621+
6622+
/*
6623+
* Post the win32_deadchild_waitinfo object for waitpid() to deal with. If
6624+
* that fails, we leak the object, but we also leak a whole process and
6625+
* get into an unrecoverable state, so there's not much point in worrying
6626+
* about that. We'd like to panic, but we can't use that infrastructure
6627+
* from this thread.
6628+
*/
6629+
if (!PostQueuedCompletionStatus(win32ChildQueue,
6630+
0,
6631+
(ULONG_PTR)lpParameter,
6632+
NULL))
6633+
write_stderr("could not post child completion status\n");
6634+
6635+
/* Queue SIGCHLD signal. */
66266636
pg_queue_signal(SIGCHLD);
66276637
}
66286638
#endif/* WIN32 */

0 commit comments

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