Skip to content

Crash handler logs devel 2 - #25

Closed
nchaimov wants to merge 25 commits into
develfrom
crash-handler-logs-devel-2
Closed

nchaimov wants to merge 25 commits into
develfrom
crash-handler-logs-devel-2

Conversation

@nchaimov

Copy link
Copy Markdown
Member

Dummy PR to force CI to run

--crash-log enables logging of which ranks crashed at which callsites. --crash-log takes optional path to log location
When run in session mode, the log can span multiple jobs. Always include executable in crashsite key.
Returning from the signal handler for a "real" fault retries faulting
instruction and refaults, but returning from a synthetic signal, such as
from kill(getpid(), SIGSEGV), does NOT refault. Detect synthetic signals
and explicitly reraise.
If the signal was triggered by the user (e.g., kill(SIGSEGV)), there is
no address to check. Treat such a signal as recovered rather than try to
check an address we don't have.
SIGSEGV SIG_IGN is ignored for user-sent signals, handled as SIG_DFL for
kernel-sent signals
If a thread other than the forking one held a lock at the time of fork,
it would disppear holding the lock, leading to a deadlock the next time
Spindle tries to acquire it.

This branch was successfully deployed

1 active deployment
Spindle CI — e21a5b8e Deployed Sep 22, 2026 by nchaimov via Testsuite (Flux, Ubuntu) #66
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant