BUG #13286: Core dumped during pg_terminate_backend call.

The following bug has been logged on the website:

Bug reference: 13286
Logged by: Maxim Boguk
Email address: maxim(dot)boguk(at)gmail(dot)com
PostgreSQL version: 9.4.1
Operating system: Centos Linux
Description:

Once per few days database getting core dumped during cron run of the
following query:

/usr/pgsql-9.4/bin/psql --no-psqlrc --single-transaction -d postgres -tx -c
"SELECT pg_terminate_backend(pid),now(),now()-xact_start as duration,* from
pg_stat_activity where (now() - pg_stat_activity.xact_start) > '1200
second'::interval and usename NOT IN ('postgres') and state<>'idle'"

I enabled collecting core and installed debug files for postgres.

There are gdb backtrace of the latest core:

Core was generated by `postgres: postgres postgres [local] SELECT
'.
Program terminated with signal 11, Segmentation fault.
#0 0x000000000066bf9b in BackendIdGetTransactionIds (backendID=<value
optimized out>, xid=0x7f47769a5538, xmin=0x7f47769a553c) at sinvaladt.c:426
426 xact =
&ProcGlobal->allPgXact[stateP->proc->pgprocno];

(gdb) bt
#0 0x000000000066bf9b in BackendIdGetTransactionIds (backendID=<value
optimized out>, xid=0x7f47769a5538, xmin=0x7f47769a553c) at sinvaladt.c:426
#1 0x00000000006287f4 in pgstat_read_current_status () at pgstat.c:2871
#2 0x0000000000628879 in pgstat_fetch_stat_numbackends () at pgstat.c:2342
#3 0x00000000006fb132 in pg_stat_get_activity (fcinfo=0x7fffaccf0ee0) at
pgstatfuncs.c:579
#4 0x000000000059dbc7 in ExecMakeTableFunctionResult (funcexpr=0x2899ae0,
econtext=0x2892bf0, argContext=<value optimized out>,
expectedDesc=0x289b0a8, randomAccess=0 '\000') at execQual.c:2193
#5 0x00000000005af622 in FunctionNext (node=0x2896cf0) at
nodeFunctionscan.c:95
#6 0x000000000059e69e in ExecScanFetch (node=0x2896cf0, accessMtd=0x5af370
<FunctionNext>, recheckMtd=0x5aeca0 <FunctionRecheck>) at execScan.c:82
#7 ExecScan (node=0x2896cf0, accessMtd=0x5af370 <FunctionNext>,
recheckMtd=0x5aeca0 <FunctionRecheck>) at execScan.c:167
#8 0x00000000005976e8 in ExecProcNode (node=0x2896cf0) at
execProcnode.c:426
#9 0x00000000005a87d2 in ExecHashJoin (node=0x2894790) at
nodeHashjoin.c:154
#10 0x0000000000597678 in ExecProcNode (node=0x2894790) at
execProcnode.c:457
#11 0x00000000005aea60 in ExecNestLoop (node=0x2892cb0) at
nodeNestloop.c:123
#12 0x0000000000597698 in ExecProcNode (node=0x2892cb0) at
execProcnode.c:449
#13 0x0000000000596252 in ExecutePlan (queryDesc=0x27e8670, direction=<value
optimized out>, count=0) at execMain.c:1486
#14 standard_ExecutorRun (queryDesc=0x27e8670, direction=<value optimized
out>, count=0) at execMain.c:319
#15 0x00007f4da708017b in pgss_ExecutorRun (queryDesc=0x27e8670,
direction=ForwardScanDirection, count=0) at pg_stat_statements.c:875
#16 0x0000000000686797 in PortalRunSelect (portal=0x2890930, forward=<value
optimized out>, count=0, dest=<value optimized out>) at pquery.c:946
#17 0x00000000006879c1 in PortalRun (portal=0x2890930,
count=9223372036854775807, isTopLevel=1 '\001', dest=0x7f4db1a68308,
altdest=0x7f4db1a68308, completionTag=0x7fffaccf1840 "") at pquery.c:790
#18 0x000000000068404e in exec_simple_query (query_string=0x2839090 "SELECT
pg_terminate_backend(pid),now(),now()-xact_start as duration,* from
pg_stat_activity where (now() - pg_stat_activity.xact_start) > '1200
second'::interval and usename NOT IN ('postgres') and st"...) at
postgres.c:1072
#19 0x00000000006856c8 in PostgresMain (argc=<value optimized out>,
argv=<value optimized out>, dbname=0x27a83f0 "postgres", username=<value
optimized out>) at postgres.c:4074
#20 0x0000000000632d7d in BackendRun (argc=<value optimized out>,
argv=<value optimized out>) at postmaster.c:4155
#21 BackendStartup (argc=<value optimized out>, argv=<value optimized out>)
at postmaster.c:3829
#22 ServerLoop (argc=<value optimized out>, argv=<value optimized out>) at
postmaster.c:1597
#23 PostmasterMain (argc=<value optimized out>, argv=<value optimized out>)
at postmaster.c:1244
#24 0x00000000005cadb8 in main (argc=3, argv=0x27a7490) at main.c:228

Error isn't repeatable but happens once week or so in random intervals.