Newest run wins: a new run cancels older live runs of the same bookmark or proposal issue #267 closed: completed andymccall opened this 16 days ago · 0 comments

Andy's decision (2026-10-19): runs are piling up faster than the runners clear them ("we're essentially DoS'ing our runners"). This morning there were 46 live runs, most of them for proposals already landed or superseded, with proposal/N runs 3-4 deep for the same proposal.

Newest wins, everywhere. A new run cancels the older live runs (queued and running) of the same thing:

• the same bookmark (push runs already do this, flow::supersede);
• the same proposal: a new proposal/N run cancels older proposal/N runs;
• when a proposal's run starts, the source bookmark's older push runs on a state it now covers;
• plus proposal #264: landing cancels the proposal's runs that haven't started.

On by default. A per-repository switch (Settings › OxygenFlow, maintain role) keeps running ones finishing (only queued ones are cancelled). Each cancelled job says "superseded by run #N". Tests for each case.