Batch doctor (#7555)

When the merge queue's re-gate hands a pull back red (the re-gate carries it alone, so a red on top of other pulls is the queue's own search, #4999), src/checks/doctor.ts classes the red (doctor_classify.ts) and acts, in crew/autonomy.ts's handBack:

Every act is a maintenance_run row (duty = batch-doctor): batch (base:head), culprit pull, class, action, probe (the first error line); the newest 200 per repository are kept (RUNS_KEPT). A sealed pull's number, batch and probe are told by no door. Read them on the flow panel (Batch doctor card), in the terminal Flow dialog, or as doctor in MCP queue_gates (all doctorRuns). A doctor's hold reason begins batch doctor: (HOLD_PREFIX, isDoctorHold). The infra class matches errno codes, no space left, disk quota and tar: lines only.

No single culprit (#11262): when the queue's search narrows a red batch to one pull that is red only on top of pulls of its candidate that landed first, the doctor classes the red and records a recheck run whose batch is base:head+#a,#b (the members it was red with). These rows are one per (pull, head, members) and share the 200 RUNS_KEPT cap with the rest, so a flapping batch adds one and no more, but many pulls can still fill it. This records; it does not re-gate: the singly re-gate already exists (a red neither side moved off is gated again alone after REGATE_RED_MS, #3155) and its reproduced red is what is held. Nothing is held here: the re-gate that carries the pull alone decides, and a functional red that reproduces there is held with the exact failing test as above.

↑↓ navigate ↵ open ⌘↵ its app esc close

Keyboard shortcuts

Docs