A schedule now has a window. If nothing ran because a worker was down, the missed run fires inside the window instead of being skipped. Default window is one hour, set it per schedule.

The outbound pool ran out of connections during a spike on 11 September. Pool is now sized to three times peak and there is an alert at 70 percent.

Set a threshold per queue. Over it, you get paged. Under it, a daily digest at nine in the morning in your timezone.

The default backoff moved from 1, 2, 4, 8, 16 seconds to 1, 4, 16 seconds, 1 minute, 5 minutes. Existing queues keep their old schedule until you change it.

Point Gantry at a database you own. The control plane keeps counts and timings, never payloads. Team plan and above.

A migration held a lock for longer than the deploy timeout on 14 July. Migrations now run behind a two second lock timeout.

gantry worker --queues a,b is now gantry worker a b. The old flag warns until v2.0.0 and then goes away.

Same four verbs as Node, Python and Go. Ruby 3.2 and up.