Runner: size CPU for alpha-worker-0, the S3 runner node
Demo / hello (push) Successful in 49s

alpha-worker-0 is labelled for the runner (S3 option 2, chosen 2026-11-16)
for its 148 GB local disk. It has 1 CPU, and other services already request
about 85% of it, so the 500m DinD request would leave the pod Pending.

- DinD CPU request 500m -> 50m. The pod's three containers request 110m.
- DinD CPU limit "2" -> 750m, so a build can't starve Elasticsearch,
  Discourse or LocalAI on that node.

Storage placement (the capped local emptyDir at /var/lib/docker) is
unchanged. NFS can't back overlay2, so a PVC isn't an option.

Refs DevOps/software-infrastructure#36, #37

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
2026-11-16 21:52:42 -05:00
co-authored by Claude Opus 5.5
parent e7cd36c59a
commit 3974f603fc
+16 -7
View File
@@ -83,11 +83,20 @@ gitea_actions_prune_keep_images: >-
| map('regex_replace', '^[^:]+:docker://', '')
| unique | list }}
# Runner and DinD CPU and memory. Sized for a 2-CPU, 16 GB worker (S3's
# recommended node) running one job at a time. The runner process itself is
# light; DinD holds the job containers, so it gets nearly all of the budget.
# A job that exceeds the DinD memory limit is OOM-killed inside the pod and
# fails with a step error, instead of the node evicting the whole runner.
# Runner and DinD CPU and memory, for one job at a time. The runner process
# itself is light; DinD holds the job containers, so it gets nearly all of the
# budget. A job that exceeds the DinD memory limit is OOM-killed inside the pod
# and fails with a step error, instead of the node evicting the whole runner.
#
# CPU is sized for the current runner node, alpha-worker-0 (chosen for S3 on
# 2026-11-16 for its 148 GB disk). It has 1 CPU, and other services already
# request about 85% of it (Elasticsearch, Discourse, LocalAI's GPU), so:
# - the DinD request is small enough to schedule there (the pod's three
# containers request 110m in total);
# - the DinD limit stays below one core, so a build can't starve those
# services. Builds are slower as a result.
# On a dedicated 2-CPU node (S3 option 1, #45 or #78), go back to a 500m
# request and a "2" limit.
gitea_actions_runner_resources:
requests:
cpu: 50m
@@ -95,8 +104,8 @@ gitea_actions_runner_resources:
limits:
cpu: 500m
memory: 256Mi
gitea_actions_dind_cpu_request: 500m
gitea_actions_dind_cpu_limit: "2"
gitea_actions_dind_cpu_request: 50m
gitea_actions_dind_cpu_limit: 750m
gitea_actions_dind_memory_request: 2Gi
gitea_actions_dind_memory_limit: 8Gi