Re: CREATE UNLOGGED TABLE seq faults when debug_discard_caches=1
От | Jakub Wartak |
---|---|
Тема | Re: CREATE UNLOGGED TABLE seq faults when debug_discard_caches=1 |
Дата | |
Msg-id | CAKZiRmxx_soeY3vQwz7yYLgZo3b-HoFAQ50NOwnzmV2qUMVa0A@mail.gmail.com обсуждение исходный текст |
Ответ на | Re: CREATE UNLOGGED TABLE seq faults when debug_discard_caches=1 (Tomas Vondra <tomas.vondra@enterprisedb.com>) |
Список | pgsql-hackers |
Hi hackers, Mr Lane, thank you for backporting this also to version 13. It seems to be occuring in the wild (without debug_discard_caches) for real user too when doing a lot of "CREATE INDEX i ON unlogged_table_truncated_after_crash (x,y)" which sometimes (rarely) results in SIGSEGV11. I've reproduced it also on 13.9 recently thanks to "break *btbuildempty / call InvalidateSystemCaches()". I'm leaving partial stack trace so that other might find it (note the: smgrwrite reln=0x0): #0 smgrwrite (reln=0x0, forknum=INIT_FORKNUM, blocknum=0, buffer=0xeef828 "", skipFsync=true) at smgr.c:516 #1 0x00000000004e5492 in btbuildempty (index=0x7f201fc3c7e0) at nbtree.c:178 #2 0x00000000005417f4 in index_build (heapRelation=heapRelation@entry=0x7f201fc49dd0, indexRelation=indexRelation@entry=0x7f201fc3c7e0, indexInfo=indexInfo@entry=0x1159dd8, #3 0x0000000000542838 in index_create (heapRelation=heapRelation@entry=0x7f201fc49dd0, indexRelationName=indexRelationName@entry=0x1159f38 "xxxxxxxx", indexRelationId=yyyyy..) #4 0x00000000005db9c8 in DefineIndex (relationId=relationId@entry=1804880199, stmt=stmt@entry=0xf2fab8, indexRelationId=indexRelationId@entry=0, parentIndexId=parentIndexId@entry=0 -Jakub Wartak. On Fri, Nov 25, 2022 at 9:48 AM Tomas Vondra <tomas.vondra@enterprisedb.com> wrote: > > > > On 11/18/22 15:43, Tom Lane wrote: > > David Geier <geidav.pg@gmail.com> writes: > >> On a different note: are we frequently running our tests suites with > >> debug_discard_caches=1 enabled? > >> It doesn't seem like. > > > > Hmm. Buildfarm members avocet and trilobite are supposed to be > > doing that, but their runtimes of late put the lie to it. > > Configuration option got lost somewhere? > > > > Yup, my bad - I forgot to tweak CPPFLAGS when upgrading the buildfarm > client to v12. Fixed, next run should be with > > CPPFLAGS => '-DCLOBBER_CACHE_ALWAYS', > > > regards > > -- > Tomas Vondra > EnterpriseDB: http://www.enterprisedb.com > The Enterprise PostgreSQL Company > > > >
В списке pgsql-hackers по дате отправления: