End forked workers with _exit() instead of PHP's full teardown - #6320
Merged
Merged
Conversation
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
| - name: "Install bashunit" | ||
| uses: "TypedDevs/bashunit@94933c088b0719e0c4cf5e3147dad67f713b970d" # 0.44.0 | ||
| with: | ||
| directory: "e2e" |
| - name: "Build the turbo extension" | ||
| run: make -C turbo-ext -j"$(nproc)" | ||
|
|
||
| - name: "Build the fork-unsafe extension" |
| - name: "Build the fork-unsafe extension" | ||
| # a model of ext-grpc without grpc.enable_fork_support: its module | ||
| # shutdown waits for a thread that a forked child never has. phpize | ||
| # builds in place, so a copy keeps the build tree out of the checkout. |
Comment on lines
+1082
to
+1088
| with: | ||
| coverage: "none" | ||
| php-version: "8.5" | ||
|
|
||
| - uses: "ramsey/composer-install@65e4f84970763564f46a70b8a54b90d033b3bdda" # v4.0.0 | ||
|
|
||
| - name: "Install bashunit" |
ondrejmirtes
force-pushed
the
forked-worker-exit
branch
from
August 31, 2026 15:59
08274b3 to
914cc5c
Compare
A pcntl_fork()ed worker inherits every loaded extension of the main process but none of its threads. exit() runs each extension's module shutdown in the child, and one that waits for its threads to check out there - ext-grpc's grpc_shutdown() without grpc.enable_fork_support - waits forever: the worker has delivered its result, the parent keeps polling waitpid(), and the run hangs at 100%. A forked child that does not exec() has to end with _exit(), which PHP itself cannot do. The turbo extension now provides Runtime::exitImmediately(), and ForkedChildTerminator registers it as the child's last shutdown function, after the crash reporter's, so every way out - exit(), a fatal error, an uncaught exception - skips the destructors and module shutdowns while the crash report still reaches the parent. Fork mode from a phar already requires the extension; a source checkout forking without it keeps the plain exit(). turbo-ext/tests/exit-immediately.php models the wedged teardown in userland, and the new e2e job builds a minimal fork-unsafe extension (ext-grpc's shape) and runs fork mode with it loaded. Co-Authored-By: Claude Fable 5 <[email protected]> Claude-Session: https://claude.ai/code/session_01C7847gqpkLK9hdBztahv8c
ondrejmirtes
force-pushed
the
forked-worker-exit
branch
from
August 31, 2026 16:06
914cc5c to
aa90b57
Compare
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
2 participants
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.
A
pcntl_fork()ed worker inherits every loaded extension of the main process but none of its threads.exit()runs each extension's module shutdown in the child, and one that waits for its threads to check out there — ext-grpc'sgrpc_shutdown()withoutgrpc.enable_fork_support, as sampled by a user in phpstan/phpstan#15131 — waits forever: the worker has delivered its result, the parent keeps pollingwaitpid(), and the run hangs at 100% with no "Result cache is saved".PHPSTAN_TURBO=0means spawn mode, which is why it "works"; and the trigger is the machine's php.ini rather than the project, which is why no reproducer repository could show it.A forked child that does not
exec()has to end with_exit(), which PHP itself cannot do. The turbo extension now providesRuntime::exitImmediately()(_exit()with the engine's exit status), andForkedChildTerminatorregisters it as the forked child's last shutdown function — afterForkedChildCrashReporter's — in bothForkedProcessand the fixer'sForkedProcessPromise. That covers every way out (exit(), a fatal error, an uncaught exception), skips the destructors and every extension's module shutdown, and still lets the crash report reach the parent. Fork mode from a phar already requires the extension; a source checkout forking without it keeps the plainexit().Regression coverage:
turbo-ext/tests/exit-immediately.phpmodels the wedged teardown in userland (a control pass proves the child hangs with plainexit()), run in every phar.yml matrix next to the smoke tests.pcntl_fork()+exit().Verified locally with that model extension loaded: the shipped 2.2.11 phar hangs at 100% (the forked worker sampled in
php_module_shutdown → zm_shutdown_* → pthread_cond_wait), a phar built from this branch completes in the same setup (symfony php, restart with the dist.so,fork (pcntl_fork)), a worker dying of the memory limit still reports "PHPStan process crashed because it reached configured PHP memory limit", the full test suite passes with the extension loaded, and raw analysis output is identical with turbo on and off.Closes phpstan/phpstan#15138
Closes phpstan/phpstan#15142
🤖 Generated with Claude Code
https://claude.ai/code/session_01C7847gqpkLK9hdBztahv8c