mirror of
https://github.com/kevinveenbirkenbach/computer-playbook.git
synced 2026-11-10 12:47:04 +00:00
kevinveenbirkenbachandClaude Opus 5.5
c1a9015f47
GitLab#3 of fork run 42305502796 died at the deploy cap (update pass rc=-15 after 7525 s; both passes may use 19020 s of the 20400 s budget). The round was storage-constrained by a hair (declared 92.1 GiB against 91.6 GiB free), so the prune handler ran after every compose flush: 55 prunes, 2926 s, up to 116 s each. The job log's host sampler (1928 free-disk samples) shows that those 55 prunes reclaimed only 13.2 GiB in total. prune.sh now reads the free space of the Docker data root and leaves Docker data alone at or above compose_prune_free_floor_mib (30720). A probe that fails falls through to the prune. Replayed on the sampler series, the floor fires 3 of the 55 prunes (302 s, saving 2624 s) and keeps the lowest free space at the observed 22.8 GiB. Across the eight red rows of the run, the largest growth between two flushes is 34.7 GiB, at the first image pulls only while ~90 GiB are free, and at most 15.3 GiB later. The unconstrained NC#7 already bottoms at 7.9 GiB without any prune. The handler reports changed only when it prunes. test_prune.py covers skip, prune, failed probes and a missing floor; against the old script the skip and the missing-floor cases fail. 🎯 Fix: 85 % 💥 Break: 5 % 💸 Cost: – ⏱️ Time: – 🛡️ Security: 100 % 🏆 Score: 81 🚦 Gate: ⛔ fix 85 % < 90 Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>