diff --git a/defaults/main.yml b/defaults/main.yml index 76a29bb..b71b51d 100644 --- a/defaults/main.yml +++ b/defaults/main.yml @@ -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