Instability of phycodorus in pg_upgrade tests with JIT

From: Michael Paquier <michael(at)paquier(dot)xyz>
To: Postgres hackers <pgsql-hackers(at)lists(dot)postgresql(dot)org>
Subject: Instability of phycodorus in pg_upgrade tests with JIT
Date: 2025-10-15 22:53:15
Message-ID: aPAl27_urHSODwRN@paquier.xyz
Views: Whole Thread | Raw Message | Download mbox | Resend email
Thread:
Lists: pgsql-hackers

Hi all,

I have spotted a couple of buildfarm failures for buildfarm member
phycodorus on REL_14_STABLE and REL_13_STABLE:
https://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=phycodurus&dt=2025-10-15%2009%3A12%3A36

These are sporadic, pointing at a backtrace with JIT in some cases on
REL_13_STABLE. Short extract with a broken free:
#18 0x00007fd9eed38242 in llvm::AttributeSet::addAttribute
(this=0x7fff3ec6af40, C=..., Indices=..., A=...) at
/home/bf/src/llvm-project-3.9/llvm/lib/IR/Attributes.cpp:882
#19 0x00007fd9eee51d42 in llvm::Function::addAttribute
(this=0x55714a9a01e8, i=4294967295, Attr=...) at
/home/bf/src/llvm-project-3.9/llvm/lib/IR/Function.cpp:377
#20 0x00007fd9eedbf113 in LLVMAddAttributeAtIndex (F=0x55714a9a01e8,
Idx=4294967295, A=0x55714a3f82e0) at
/home/bf/src/llvm-project-3.9/llvm/lib/IR/Core.cpp:1845
#21 0x00007fd9fbe2b393 in llvm_copy_attributes_at_index
(v_from=v_from(at)entry=0x55714a34ab28, v_to=v_to(at)entry=0x55714a9a01e8,
index=index(at)entry=4294967295) at
/home/bf/bf-build/phycodurus/REL_13_STABLE/pgsql.build/../pgsql/src/backend/jit/llvm/llvmjit.c:551
#22 0x00007fd9fbe2c2df in llvm_copy_attributes (v_from=0x55714a34ab28,
v_to=v_to(at)entry=0x55714a9a01e8) at
/home/bf/bf-build/phycodurus/REL_13_STABLE/pgsql.build/../pgsql/src/backend/jit/llvm/llvmjit.c:566
#23 0x00007fd9fbe34b28 in llvm_compile_expr (state=0x55714a3a80b8) at
/home/bf/bf-build/phycodurus/REL_13_STABLE/pgsql.build/../pgsql/src/backend/jit/llvm/llvmjit_expr.c:158
#24 0x00005571479f5448 in jit_compile_expr
(state=state(at)entry=0x55714a3a80b8) at
/home/bf/bf-build/phycodurus/REL_13_STABLE/pgsql.build/../pgsql/src/backend/jit/jit.c:177

REL_14_STABLE points at a crash, without a backtrace. It looks like
only this host is seeing such failures for the upgrade test, for only
these two branches. Is that something we'd better act on even for v13
which is going to be EOL soon?

Thanks,
--
Michael

Responses

Browse pgsql-hackers by date

  From Date Subject
Next Message Jacob Champion 2025-10-15 23:19:09 Re: Thoughts on a "global" client configuration?
Previous Message Masahiko Sawada 2025-10-15 22:47:36 Re: POC: enable logical decoding when wal_level = 'replica' without a server restart