Found while checking the runner's docs against its code:
crates/knotless-run/src/queue.rs module doc still says batches share "the cache volumes one run at a time uses"; per-run overlays removed that wait. It also says a pull in a batch moving drops the batch, but a head moving mid-run is caught at landing instead.
lexicons/dev/knotless/check.json:
the key description covers push and pull keys only, not a batch's (<owner hash>.<repo key>.<hash of "queue refs/heads/<branch>">.<tip>.<base>.<workflow>), and baseCommit is described as a pull's only, though batches set it too;
cancelled is described as "a run stopped for a newer commit's run", but runs are also cancelled when a pull moves or closes, a batch is dropped, or the runner stops past --stop-wait;
a step's exitCode is described as absent only when it ran out of time; it is absent when cancelled too.
Descriptions only, so evolving the lexicon is just text.
chadtmiller.com/knotless
Found while checking the runner's docs against its code:
<owner hash>.<repo key>.<hash of "queue refs/heads/<branch>">.<tip>.<base>.<workflow>), andbaseCommitis described as a pull's only, though batches set it too;cancelledis described as "a run stopped for a newer commit's run", but runs are also cancelled when a pull moves or closes, a batch is dropped, or the runner stops past--stop-wait;exitCodeis described as absent only when it ran out of time; it is absent when cancelled too.Descriptions only, so evolving the lexicon is just text.