Skip to content

Crash handler pr fixes - #18

Open
nchaimov wants to merge 13 commits into
develfrom
crash-handler-pr-fixes
Open

Crash handler pr fixes#18
nchaimov wants to merge 13 commits into
develfrom
crash-handler-pr-fixes

Conversation

@nchaimov

@nchaimov nchaimov commented Jul 8, 2026

Copy link
Copy Markdown
Member

Dummy PR to force CI to run

nchaimov added 13 commits June 17, 2026 15:33
Adds server-side components of crash handling.
Client sends LDCS_MSG_CRASH_REPORT to sever.
Server propagates request. When a single rank is chosen
for a given crashsite, replies with LDCS_MSG_CRASH_RESPONSE.
Add the client-side crash handler. On a fatal signal, it chains to an
application-registered signal handler, if any, and, if the application
handler did not fix the fault, sends LDCS_MSG_CRASH_REPORT to the server
which selects a single process per crashsite. Non-selected crashsites
set their coredump limit to zero.
Adds --crash-dedup option to control whether crash deduplication is
enabled.
Adds --enable-crash-handler to control whether the crash handler is
compiled.
Adds --enable-crash-dedup to control whether crash deduplication is
enabled by default at runtime.
Adds a new script run_crash_tests_template.sh which runs various
configurations of the new crash_test.c testsuite. This tests various
kinds of crashes, numberes of distinct crashsites, etc. to validate that
the crash handler deduplicates them correctly. It also tests chaining to
application signal handlers that use the safepoint pattern to fix faults
and retry.
Run the new crash handler testsuite in CI. Configure CI runner to generate
coredumps, add shared filesystem to aggregate coredumps across nodes in
mutli-node tests. Grants CAP_SYS_RESOURCE so ulimit can be set in
containers.
autotools-generated files re-generated
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