Skip to main content
← Back to list
01Issue
BugOpenSwamp CLIPublic
AssigneesNone

Relationships

#4293 S3 and GCS: after a real commit by another writer, a serve stays off the pull fast path until it restarts

Opened by zack · 11/2/2026

Split from #4102, which fixes this for commits a service made itself.

Model-scoped and subdir-scoped pulls never re-arm the sidecar commitSeq, because neither has read the whole index (#2246), and the runs and pollers of a serve only make those pulls. So one commit by another writer (a peer, or a CLI push) leaves that serve on the slow path until a full pull, which in practice means a restart.

After #4102 this state no longer overwrites other models unpushed files, and scoped pulls list only their own entries, but each poll still reads every shard and lists the datastore. Deployments with real peers keep that cost. Related to #2998.

Wanted: a way to re-arm without reading every shard on every miss.

02Bog Flow
◉OPEN○TRIAGED○IN PROGRESS○SHIPPED

Open

11/2/2026, 3:47:53 PM

No activity in this phase yet.

03Sludge Pulse

Sign in to post a ripple.