Experiment formal browser owned cleanup source results 01 - kmosoti/FabricO11y GitHub Wiki
Browser owned cleanup source results, 2026-10-15
These are harness source controls, not desktop/browser or frozen-package acceptance. The actual production finalizer and binder/stopper execute against real bounded synthetic processes, retained pidfds and real adopted-child reaping. The cgroup placement/population/kill interfaces are modeled.
The previous 13-method registration retained one success, eight failures and four dependent skips. At a5c6ad6b693ad1b5ac85cd459d15b6911154b713, all 13 executed successfully without skips. A follow-up preserves Popen's actual leader status rather than consuming the leader with waitid. At 336949ee75f03b927a69b2eadddf1a9bc4da07ed, all 13 again executed successfully without skips.
The controls cover leader-already-exited, TERM-resistant and setsid children, adopted zombies, failed stops with other groups attempted, sampler-stop refusal, stale process/directory/descriptors, placement/delegation/capability refusal and bounded stuck population. Original tests and expected failures were unchanged. ResourceWarnings from synthetic process Popen bookkeeping remain in both logs; outer containment stopped, and independent fixture guardians handled refused groups. This does not establish native launch/late-fork completeness.
Native live_browser_check.py explicitly refuses execution pending actual delegated launch ownership and the registered later-fork/reaping strategy. No new browser acceptance command is admitted by these source results. Full fast checks have not run; normal documentation validation is recorded separately.
browser-owned-cleanup-controls-01
Revision a5c6ad6b693ad1b5ac85cd459d15b6911154b713. Exit 0. Outer stop confirmed: True.
env PYTHONPATH=tools/ui python3 -B -m unittest -v test_role_matrix.BrowserFixtureCleanupTests test_role_matrix.DelegatedBrowserOwnershipTests
Executed inside python3 -B tools/resource_group.py -- ..., admitted by the pinned run-step driver. Full raw command/receipt: /run/media/kmosoti/data/FabricO11y/results/service-opening-candidate/browser-owned-cleanup-controls-01.json. Log SHA-256 672f037c4ae937bc34ae8651d391df66d9bc08b77b70c6b939fd6c5a93d08941.
browser-owned-cleanup-controls-02
Revision 336949ee75f03b927a69b2eadddf1a9bc4da07ed. Exit 0. Outer stop confirmed: True.
env PYTHONPATH=tools/ui python3 -B -m unittest -v test_role_matrix.BrowserFixtureCleanupTests test_role_matrix.DelegatedBrowserOwnershipTests
Executed inside python3 -B tools/resource_group.py -- ..., admitted by the pinned run-step driver. Full raw command/receipt: /run/media/kmosoti/data/FabricO11y/results/service-opening-candidate/browser-owned-cleanup-controls-02.json. Log SHA-256 bf29bd55eac19c73d82d61e43f77cd018a1a2478945e6092cb8c215afbebe5cc.
Each cell reserved 1 GiB on the mounted data drive under the default 16/20 GiB high/max, no swap, 1800-second launcher. Heavy jobs were serialized. Failed historical inputs remain preserved; successful fixture-owned scratch was removed only after confirmed stops. No product/package/independent semantic oracle bytes changed.
Documentation command bun tools/docs/check.mjs, through the resource launcher: exit 0 at 361a326508c6e0952c2585bb00943489e9531498; outer stop confirmed. Raw receipt /run/media/kmosoti/data/FabricO11y/results/service-opening-candidate/browser-owned-cleanup-docs-01.json.