{"session": "aa7b0113-6875-4844-9251-b121d8830cf3", "env": "main-miss-lovelace", "command": "journal work log 14 \"Stored-back test, 8 runs each: -n 0 passed 8, -n 1 failed 1 of 8, -n 3 failed 1 of 8 (lighter load than before). With the earlier 14 in-process passes, in-process is 22/22 and xdist workers about 2 of 16 failing now (6 of 12 earlier at higher load): worker runs fail, in-process never, and a heavier machine raises the rate.\" 2>&1 | tail -1; journal work end 14 --how \"narrowed; nothing more to learn without reading stored.py's stamp paths\" 2>&1 | tail -1\njournal helper report \"Alfred: one more measurement on the stored-back flake, nothing changed in code. Same test, 8 runs each at lighter load (about 30): -n 0 (one process) 8 of 8 passed; -n 1 failed 1 of 8; -n 3 failed 1 of 8. With my earlier 14 passes in one process that is 22 of 22 in one process against 2 of 16 failing in an xdist worker now, and 6 of 12 failing earlier when the machine was busier. So: it only fails in an xdist worker, and a loaded machine makes it fail more often; it is a real race in the row store's stamp or index path (src/controllers/stored.py), made likely by slowness, not a starved wait. It still needs the owner of stored.py. I stop on it here.\" 2>&1 | tail -1", "parts": ["journal work log 14 \"Stored-back test, 8 runs each: -n 0 passed 8, -n 1 failed 1 of 8, -n 3 failed 1 of 8 (lighter load than before). With the earlier 14 in-process passes, in-process is 22/22 and xdist workers about 2 of 16 failing now (6 of 12 earlier at higher load): worker runs fail, in-process never, and a heavier machine raises the rate.\" 2>&1 | tail -1", "journal work end 14 --how \"narrowed; nothing more to learn without reading stored.py's stamp paths\" 2>&1 | tail -1", "journal helper report \"Alfred: one more measurement on the stored-back flake, nothing changed in code. Same test, 8 runs each at lighter load (about 30): -n 0 (one process) 8 of 8 passed; -n 1 failed 1 of 8; -n 3 failed 1 of 8. With my earlier 14 passes in one process that is 22 of 22 in one process against 2 of 16 failing in an xdist worker now, and 6 of 12 failing earlier when the machine was busier. So: it only fails in an xdist worker, and a loaded machine makes it fail more often; it is a real race in the row store's stamp or index path (src/controllers/stored.py), made likely by slowness, not a starved wait. It still needs the owner of stored.py. I stop on it here.\" 2>&1 | tail -1"]}