Fix log stream hang after failed cached builds

[TorstenDittmann]

Summary

• terminate the active logs tail process when a runtime disappears while streaming logs
• add an E2E regression for failed cache-enabled v5 builds with concurrent /logs

Verification

• ./vendor/bin/pint --test --config pint.json src/Executor/Runner/Docker.php tests/e2e/ExecutorTest.php
• docker run --rm --network executor_runtimes -v "$PWD":/app -w /app executor-openruntimes-executor ./vendor/bin/phpunit --configuration phpunit.xml --filter testLogStreamClosesAfterFailedCachedBuild
• docker run --rm --network executor_runtimes -v "$PWD":/app -w /app executor-openruntimes-executor ./vendor/bin/phpunit --configuration phpunit.xml --testsuite=unit

Negative Control

• temporarily removed the termination fix, rebuilt the executor, and confirmed testLogStreamClosesAfterFailedCachedBuild failed because /logs stayed open for ~13.57s instead of closing promptly

[greptile-apps]

Greptile Summary

This PR fixes a log-stream hang that occurred when a runtime was removed (e.g., after a failed cached build) while a /logs SSE client was connected: the timer callback now flushes any buffered log chunk and terminates the tail -F process before returning, so the stream closes promptly instead of waiting for the full timeout to expire.

• Docker.php: In the timer tick that watches $activeRuntimes, a new branch is added for the "runtime no longer present" case — it writes any pending $logsChunk to the response, then calls proc_terminate on $logsProcess (when the handle is available), which unblocks the Console::execute call and allows Timer::clear to run.
• ExecutorTest.php: A new E2E regression test testLogStreamClosesAfterFailedCachedBuild starts a /logs stream and a failing cached build concurrently, then asserts the stream closes in under 10 seconds and that the build's output is present in the 400 error response.

Confidence Score: 4/5

The fix correctly closes the stream for the primary failure scenario; a fast-failing build that never emits timing data can still leave the stream open until the configured timeout.

The timer callback change is tightly scoped and backed by a negative-control regression test. The proc_terminate call is gated on !empty($logsProcess), meaning a build that completes (or fails) before the tail -F process has had a chance to emit any output — before $logsProcess is set — bypasses the termination path and the stream stays open until $timeout expires.

src/Executor/Runner/Docker.php — specifically the !empty($logsProcess) guard, which is the remaining gap in the fix.

Important Files Changed

Filename	Overview
src/Executor/Runner/Docker.php	Adds flush-then-terminate logic in the timer callback when the runtime disappears mid-stream; correct for the common case where the tail process handle is already set, but a narrow timing window remains if the process hasn't been initialized yet
tests/e2e/ExecutorTest.php	New regression test for failed cached-build log streaming; uses a random runtimeId but a hardcoded destination path and cacheKey that match the existing testLogStream test, creating a latent conflict if tests ever run in parallel

_{Reviews (2): Last reviewed commit: "fix: flush logs before terminating missi..." | Re-trigger Greptile}