- Shipped
- November 14, 2026 at 8:56 PM UTC
- Author
- Kamo
- Commit
- ff28b24
Stepping the prune (c740c1b) was not enough: on k1m1's shared disk a 3-5 GB --keep-storage step takes 2-2.5 min and a queued `docker rm` sat out each step (probe: 106-139 s per rm, vs 427 s under the old single prune). Run every 5 min and prune only when no Forgejo job container younger than 60 min is running, re-checking between steps and stopping as soon as a job starts. At >=55% volume usage it prunes regardless (still stepped). CI is idle ~33% of the time with 4+ idle windows of 10+ min a day, and the volume has ~130 GB headroom.