search

2ndQuadrant / pglogical

Logical Replication extension for PostgreSQL 17, 16, 15, 14, 13, 12, 11, 10, 9.6, 9.5, 9.4 (Postgres), providing much faster replication than Slony, Bucardo or Londiste, as well as cross-version upgrades.
http://2ndquadrant.com/en/resources/pglogical/
Other
1.01k stars 153 forks source link

Segfault with triggers #229

Open slardiere opened 4 years ago

slardiere commented 4 years ago

Hi,

I use PostgreSQL 11.6 and pglogical 2.2.2 on debian, with deb packages. I'm using two instances in the same system.

As I need to filter DELETE on the subscriber side, I implement a simple plpgsql trigger function, on the subscriber side. The trigger is fired, then, in some case, return null to not apply the delete. At this moment, postgres segfault with this message in the logfile : 

LOG:  background worker "pglogical apply 32935:3371347705" (PID 95481) was terminated by signal 11: Segmentation fault

Then postgres restart, and loop over with the same problem. I can reproduce the situation, The stack trace is : 

#0  relation_close (relation=0x0, lockmode=lockmode@entry=0) at ./build/../src/backend/access/heap/heapam.c:1272
#1  0x00007fb43f899c25 in pglogical_relation_close (rel=rel@entry=0x55ba97094500, lockmode=lockmode@entry=0) at ./pglogical_relcache.c:200
#2  0x00007fb43f8925a8 in handle_delete (s=s@entry=0x7ffc4fc44610) at ./pglogical_apply.c:698
#3  0x00007fb43f8941f8 in replication_handler (s=0x7ffc4fc44610) at ./pglogical_apply.c:1139
#4  apply_work (streamConn=streamConn@entry=0x55ba97052f00) at ./pglogical_apply.c:1421
#5  0x00007fb43f894cd6 in pglogical_apply_main (main_arg=<optimized out>) at ./pglogical_apply.c:1976
#6  0x000055ba957f044d in StartBackgroundWorker () at ./build/../src/backend/postmaster/bgworker.c:834
#7  0x000055ba957fd258 in do_start_bgworker (rw=0x55ba96feb060) at ./build/../src/backend/postmaster/postmaster.c:5721
#8  maybe_start_bgworkers () at ./build/../src/backend/postmaster/postmaster.c:5947
#9  0x000055ba957fded5 in sigusr1_handler (postgres_signal_arg=<optimized out>) at ./build/../src/backend/postmaster/postmaster.c:5094
#10 <signal handler called>
#11 0x00007fb44562f5e3 in __select_nocancel () at ../sysdeps/unix/syscall-template.S:84
#12 0x000055ba955806e4 in ServerLoop () at ./build/../src/backend/postmaster/postmaster.c:1671
#13 0x000055ba957ff081 in PostmasterMain (argc=5, argv=0x55ba96f99e40) at ./build/../src/backend/postmaster/postmaster.c:1380
#14 0x000055ba95582624 in main (argc=5, argv=0x55ba96f99e40) at ./build/../src/backend/main/main.c:228

Attached script can reproduce the problem

Regards,

replilogique04.sql.gz

luss commented 4 years ago

Has anyone looked into this for you yet?

slardiere commented 4 years ago

I don't think so, no ; actually, I don't need it, because trigger was not the solution, but it's still a bug, afaik.

jcasanov commented 3 years ago

303 fixes this