Re: Out of Memory errors are frustrating as heck! - Mailing list pgsql-performance

From Gunther
Subject Re: Out of Memory errors are frustrating as heck!
Date
Msg-id f6d4f3be-44f8-0c45-de0d-e68caef9bb60@gusw.net
Whole thread Raw
In response to Re: Out of Memory errors are frustrating as heck!  (Tomas Vondra <tomas.vondra@2ndquadrant.com>)
Responses Re: Out of Memory errors are frustrating as heck!
List pgsql-performance
On 4/20/2019 21:14, Tomas Vondra wrote:
Maybe. But before wasting any more time on the memory leak investigation,
I suggest you first try the patch moving the BufFile allocations to a
separate context. That'll either confirm or disprove the theory.

OK, fair enough. So, first patch 0001-* applied, recompiled and

2019-04-21 04:08:04.364 UTC [11304] LOG:  server process (PID 11313) was terminated by signal 11: Segmentation fault
2019-04-21 04:08:04.364 UTC [11304] DETAIL:  Failed process was running: explain analyze select * from reports.v_BusinessOperation;
2019-04-21 04:08:04.364 UTC [11304] LOG:  terminating any other active server processes
2019-04-21 04:08:04.368 UTC [11319] FATAL:  the database system is in recovery mode
2019-04-21 04:08:04.368 UTC [11315] WARNING:  terminating connection because of crash of another server process

SIGSEGV ... and with the core dump that I have I can tell you where:

Core was generated by `postgres: postgres integrator'.
Program terminated with signal SIGSEGV, Segmentation fault.
#0  0x00000000009f300c in palloc (size=8272) at mcxt.c:936
936             context->isReset = false;
(gdb) bt
#0  0x00000000009f300c in palloc (size=8272) at mcxt.c:936
#1  0x000000000082b068 in makeBufFileCommon (nfiles=1) at buffile.c:116
#2  0x000000000082b0f8 in makeBufFile (firstfile=73) at buffile.c:138
#3  0x000000000082b283 in BufFileCreateTemp (interXact=false) at buffile.c:201
#4  0x00000000006bdc15 in ExecHashJoinSaveTuple (tuple=0x1c5a468, hashvalue=3834163156, fileptr=0x18a3730) at nodeHashjoin.c:1227
#5  0x00000000006b9568 in ExecHashTableInsert (hashtable=0x188fb88, slot=0x1877a18, hashvalue=3834163156) at nodeHash.c:1701
#6  0x00000000006b6c39 in MultiExecPrivateHash (node=0x1862168) at nodeHash.c:186
#7  0x00000000006b6aec in MultiExecHash (node=0x1862168) at nodeHash.c:114
#8  0x00000000006a19cc in MultiExecProcNode (node=0x1862168) at execProcnode.c:501
#9  0x00000000006bc5d2 in ExecHashJoinImpl (pstate=0x17b90e0, parallel=false) at nodeHashjoin.c:290
...
(gdb) info frame
Stack level 0, frame at 0x7fffd5d4dc80:rip = 0x9f300c in palloc (mcxt.c:936); saved rip = 0x82b068called by frame at 0x7fffd5d4dcb0source language c.Arglist at 0x7fffd5d4dc70, args: size=8272Locals at 0x7fffd5d4dc70, Previous frame's sp is 0x7fffd5d4dc80Saved registers: rbx at 0x7fffd5d4dc60, rbp at 0x7fffd5d4dc70, r12 at 0x7fffd5d4dc68, rip at 0x7fffd5d4dc78

and I have confirmed that this is while working the main T_HashJoin with jointype JOIN_RIGHT.

So now I am assuming that perhaps you want both of these patches applied. So applied it, and retried and boom, same thing same place.

turns out the MemoryContext is NULL:

(gdb) p context
$1 = (MemoryContext) 0x0

all patches applied cleanly (with the -p1 option). I see no .rej file, but also no .orig file, not sure why that version of patch didn't create them. But I paid attention and know that there was no error.

-Gunther

pgsql-performance by date:

Previous
From: Tomas Vondra
Date:
Subject: Re: Out of Memory errors are frustrating as heck!
Next
From: Gunther
Date:
Subject: Re: Out of Memory errors are frustrating as heck!