You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 191a386
Browse filesBrowse the repository at this point in the historyBrowse files
fix(server): stop unbounded event replay and full-DB snapshot hydration
subscribeThread's stale-cursor catch-up read the entire global event log
(readEvents with Number.MAX_SAFE_INTEGER) and filtered per-thread in JS.
Reconnecting clients with cursors hundreds of thousands of events behind
forced the server to page and JSON-decode every intervening event's
payload, OOM-killing the backend on large databases. The replay is now
bounded to the projection head with the same gap cap subscribeShell
already uses; past the cap the client is reset with a fresh thread
snapshot, which it already handles.
GET /api/orchestration/snapshot and the offline project CLI hydrated
every message and activity payload in the database via getSnapshot().
The route's only consumer (the project CLI) reads the project list, so
both now use the lightweight getCommandReadModel() with the same wire
shape.
Co-Authored-By: Claude Fable 5 <[email protected]>
0 commit comments