Describe the bug
Exec: /rundir/erts-16.4.0.3/bin/erlexec -noinput +Bd -boot /rundir/releases/latest/start -mode embedded -boot_var SYSTEM_LIB_DIR /rundir/lib -config /rundir/releases/latest/sys.config -args_file /rundir/releases/latest/vm.args -- foreground
Root: /rundir
/rundir
sys/unix/sys_signal_stack.c:111:sys_thread_init_signal_stack(): Internal error: Failed to set alternate signal stack
This manifest on a patched (with #11249) version of 28.5.0.3 in GitHub Actions. We've had this problem for a while, very intermittent, probably because there is only a few CI runners with the "right" architecture.
To Reproduce
Build a container image (OTP-28.5.0.3 + PR #11249) and let it run something in Github Actions.
Note that the error is on line 111 (the unpatched version has the error on line 103) which should indicate that the patch was applied.
Expected behavior
A correctly sized alternate signal stack.
Affected versions
Tested on 28.5.0.3 + patch.
Additional context
The patch does:
ss_size = max(SIGSTKSZ, getauxval(AT_MINSIGSTKSZ))
If getauxval(AT_MINSIGSTKSZ) comes back as 0/missing on some runner variants, it falls back to musl’s fixed SIGSTKSZ=8192 , and on AVX512-class hosts that can still fail exactly like before. This is consistent with “intermittent across CI fleet” even with the patch applied.
Describe the bug
This manifest on a patched (with #11249) version of 28.5.0.3 in GitHub Actions. We've had this problem for a while, very intermittent, probably because there is only a few CI runners with the "right" architecture.
To Reproduce
Build a container image (OTP-28.5.0.3 + PR #11249) and let it run something in Github Actions.
Note that the error is on line
111(the unpatched version has the error on line103) which should indicate that the patch was applied.Expected behavior
A correctly sized alternate signal stack.
Affected versions
Tested on 28.5.0.3 + patch.
Additional context
The patch does:
If
getauxval(AT_MINSIGSTKSZ)comes back as 0/missing on some runner variants, it falls back to musl’s fixedSIGSTKSZ=8192, and on AVX512-class hosts that can still fail exactly like before. This is consistent with “intermittent across CI fleet” even with the patch applied.