Fix regression failure in WAL redo process #152
Conversation
This error is happening in the 'pg_regress' test in the CI, but not on my laptop. Turn it into an ERROR, so that we get the error context and backtrace of it.
In the WAL redo process, even "permanent" buffers are stored in the local buffer cache. Need to pass RELPERSISTENCE_PERMANENT to smgropen() in that case.
| @@ -195,6 +195,10 @@ inmem_nblocks(SMgrRelation reln, ForkNumber forknum) | |||
| { | |||
| int nblocks = 0; | |||
|
|
|||
| /* | |||
There was a problem hiding this comment.
#0 inmem_nblocks (reln=0x5620695fc048, forknum=VISIBILITYMAP_FORKNUM)
at /home/knizhnik/zenith/tmp_install/build/../../vendor/postgres/contrib/zenith/inmem_smgr.c:197
#1 0x000056206894beba in smgrnblocks (reln=0x5620695fc048, forknum=forknum@entry=VISIBILITYMAP_FORKNUM)
at /home/knizhnik/zenith/tmp_install/build/../../vendor/postgres/src/backend/storage/smgr/smgr.c:564
#2 0x00005620686b5aba in vm_extend (rel=rel@entry=0x5620695dcdd8, vm_nblocks=vm_nblocks@entry=1)
at /home/knizhnik/zenith/tmp_install/build/../../vendor/postgres/src/backend/access/heap/visibilitymap.c:650
#3 0x00005620686b5c44 in vm_readbuf (rel=rel@entry=0x5620695dcdd8, blkno=blkno@entry=0, extend=extend@entry=true)
at /home/knizhnik/zenith/tmp_install/build/../../vendor/postgres/src/backend/access/heap/visibilitymap.c:576
#4 0x00005620686b5ea3 in visibilitymap_pin (rel=rel@entry=0x5620695dcdd8, heapBlk=, buf=buf@entry=0x7ffdf3e751f8)
at /home/knizhnik/zenith/tmp_install/build/../../vendor/postgres/src/backend/access/heap/visibilitymap.c:202
#5 0x000056206869dd6a in heap_xlog_insert (record=0x5620695e5928)
at /home/knizhnik/zenith/tmp_install/build/../../vendor/postgres/src/backend/access/heap/heapam.c:8904
#6 0x0000562068956e6d in ApplyRecord (input_message=0x7ffdf3e77270)
at /home/knizhnik/zenith/tmp_install/build/../../vendor/postgres/src/backend/tcop/zenith_wal_redo.c:612
The last commit here, "Fix errors in WAL redo about relpersistence mismatch" fixes the pg_regress failures from neondatabase/neon#1573. The others add debugging helpers that were useful to hunt it down. Let's add them permanently, too.