Only the end-of-stream sentinel may stop the convergence output thread - #7267
Merged
Merged
Conversation
hnil
added
the
manual:irrelevant
hnil
force-pushed
the
fix-convergence-output-crash
branch
from
August 11, 2026 08:28
9093b40 to
9ebcef1
Compare
hnil
marked this pull request as ready for review
August 11, 2026 08:49
hnil
changed the title
hnil
force-pushed
the
fix-convergence-output-crash
branch
from
August 11, 2026 12:06
9ebcef1 to
91524cb
Compare
Member
|
jenkins build this please |
Contributor
There was a problem hiding this comment.
Pull request overview
This PR refines the lifecycle and shutdown semantics of the asynchronous convergence output writer to ensure only an explicit end-of-stream sentinel can terminate the writer thread, avoiding premature shutdown when a valid step produces an empty convergence report sequence.
Changes:
- Introduces an explicit
isFinalflag onOutputRequestand updates the writer loop to terminate only on that sentinel (not on “empty reports”). - Makes INFOITER header generation resilient by deriving columns from the first request that actually contains a convergence report, and skipping report-less requests.
- Adds shutdown backstops by making
endThread()idempotent and ensuring the convergence output thread is joined during destruction paths.
Reviewed changes
Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.
Show a summary per file
| File | Description |
|---|---|
| opm/simulators/flow/SimulatorFullyImplicit_impl.hpp | Flushes any remaining step reports on simulator destruction and ends the convergence output thread. |
| opm/simulators/flow/SimulatorConvergenceOutput.hpp | Adds a destructor that calls endThread() as a safety backstop. |
| opm/simulators/flow/SimulatorConvergenceOutput.cpp | Makes endThread() idempotent by resetting the stored thread after joining. |
| opm/simulators/flow/ExtraConvergenceOutputThread.hpp | Extends OutputRequest with an explicit isFinal end-of-stream marker. |
| opm/simulators/flow/ExtraConvergenceOutputThread.cpp | Updates writer logic to use isFinal sentinel, defers header creation until a real report exists, and skips empty-report requests. |
💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.
Member
|
Would you mind responding to Copilot's review comments, please? |
When a simulation aborted before completing any report step, the convergence output thread crashed with SIGSEGV: the end-of-stream marker enqueued by signalLastOutputRequest() is a request with an empty reports vector, and writeConvergenceHeader() unconditionally dereferenced reports.front() of the first request to lay out the metric columns. Three related problems fixed: - The shutdown sentinel is now an explicit isFinal flag on OutputRequest instead of "reports is empty". An ordinary request with no convergence reports (a step that aborted before its first convergence check) was previously indistinguishable from the sentinel and silently shut the writer thread down mid-run. - The header is laid out from the first request that actually carries a convergence report, and requests without reports are skipped instead of dereferenced. - The step reports accumulated up to the point of failure are now flushed to the INFOITER file before the writer thread is joined, and endThread() is idempotent (also called from a new SimulatorConvergenceOutput destructor as a backstop). Previously a failed run left a zero-byte INFOITER -- discarding the convergence history of exactly the runs whose convergence history is most needed. Reproducer: any deck that fails in its first report step with --output-extra-convergence-info=steps,iterations, e.g. opm-tests wconprod/WCONPROD-01.DATA (crash) or spe02/SPE02-SPIDER-01.DATA (zero-byte INFOITER); both now exit cleanly with diagnostics on disk. Healthy runs produce byte-identical INFOITER content.
- endThread() joins only a joinable thread. - Regression test: a request with no reports that is not the sentinel must not stop the output thread. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
hnil
force-pushed
the
fix-convergence-output-crash
branch
from
September 17, 2026 14:21
91524cb to
b4bcbda
Compare
Member
Author
|
jenkins build this please |
bska
approved these changes
Sep 17, 2026
bska
left a comment
bska
left a comment
Member
There was a problem hiding this comment.
Thanks a lot for the updates. This looks good to me now and as the build check is green, I'll merge this into the master branch to enable the new behaviour.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
#7257 has landed since this was opened and fixes the segfault it originally targeted. What is left is the semantics underneath that crash.
OutputRequestused "reports is empty" as the shutdown sentinel, butSimulatorConvergenceOutput::write()builds one request perStepReport— so a step that aborted before its first convergence check produces an ordinary request with no reports, indistinguishable from the sentinel. The writer thread then stops for the rest of the run. This makes the sentinel an explicitisFinalflag, lays the header out from the first request that actually carries a report, and skips report-less requests instead of dereferencing them.endThread()is idempotent and a destructor backstops it.Checked on current master:
wconprod/WCONPROD-01.DATAwith--output-extra-convergence-info=steps,iterationsexits cleanly, and its INFOITER is empty both with and without this PR — that run never completes a step, so there is nothing to write. The "zero-byte INFOITER" claim in the original description does not hold; INFOITER is flushed incrementally as steps complete. I have left that out.