From 4f370518cb6611e82af505b9f0c0f1f6da953b79 Mon Sep 17 00:00:00 2001 From: Philip Bernaerts Date: Sat, 11 Jul 2026 18:06:09 +0200 Subject: [PATCH] docs(monitoring): record cAdvisor #3860 status and disable_metrics=disk fallback Note in docker-stats-textfile.sh header that only cAdvisor's disk-stats collection breaks on Docker 29 (still open at v0.60.5), that --disable_metrics=disk is a viable fallback, and why the script remains the deliberate choice (unprivileged, timer-paced, anon-memory metric). Upstream fix to watch: google/cadvisor#3892. Co-Authored-By: Claude Fable 5 --- monitoring/scripts/docker-stats-textfile.sh | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/monitoring/scripts/docker-stats-textfile.sh b/monitoring/scripts/docker-stats-textfile.sh index 905df64..053012c 100755 --- a/monitoring/scripts/docker-stats-textfile.sh +++ b/monitoring/scripts/docker-stats-textfile.sh @@ -3,6 +3,11 @@ # cAdvisor can't collect metrics on this host (Docker 29 + overlayfs snapshotter, # google/cadvisor#3860), so we sample the Docker stats API instead — it reads cgroup-v2 # usage directly and is independent of the storage-driver layer-resolution bug. +# Status 2026-07-11: #3860 still open at cAdvisor v0.60.5. Only disk-stats collection is +# broken; `--disable_metrics=disk` makes cAdvisor work (per-container CPU/mem/net, no +# container_fs_*). Deliberately staying with this script: no privileged container, no 15s +# sampling loop, and it exposes the anon-memory metric dashboards/alerts depend on. +# Upstream fix to watch: google/cadvisor#3892 (+ #3910). # # Metrics {name=...}: # docker_container_cpu_percent CPU% across all cores