Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion .github/workflows/pr-test-rust.yml
Original file line number Diff line number Diff line change
Expand Up @@ -252,7 +252,7 @@ jobs:
reruns: ""
upload_benchmarks: true
parallel_opts: "" # No parallel for benchmarks (performance measurement)
ignore_opts: "--ignore=e2e_test/benchmarks/test_go_bindings_perf.py --ignore=e2e_test/benchmarks/test_nightly_perf.py --ignore=e2e_test/benchmarks/test_pd_perf.py" # Go and nightly benchmarks run in dedicated jobs
ignore_opts: "--ignore=e2e_test/benchmarks/test_go_bindings_perf.py --ignore=e2e_test/benchmarks/test_nightly_perf.py" # Go and nightly benchmarks run in dedicated jobs
runner: 4-gpu-h100
- name: agentic-apis
timeout: 30
Expand Down
13 changes: 12 additions & 1 deletion scripts/k8s-runner-resources/arc-runner-gpu.yaml
Original file line number Diff line number Diff line change
Expand Up @@ -44,7 +44,9 @@ spec:
- name: docker-sock
emptyDir: {}
- name: docker-storage
emptyDir: {}
emptyDir:
medium: Memory
sizeLimit: 4Gi
- name: dshm
emptyDir:
medium: Memory
Expand All @@ -70,9 +72,18 @@ spec:
image: fra.ocir.io/idqj093njucb/docker:dind
securityContext:
privileged: true # Required for DinD
Comment thread
key4ng marked this conversation as resolved.
resources:
requests:
cpu: "1"
memory: "2Gi"
limits:
cpu: "2"
memory: "4Gi"
Comment thread
key4ng marked this conversation as resolved.
Comment on lines +79 to +81

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P1 Badge Increase DinD memory limit to match tmpfs Docker storage

The new docker-storage volume is a memory-backed emptyDir with an 8Gi limit, but the DinD container is capped at memory: "4Gi"; for memory-backed emptyDir, writes are charged to the writing container, so Docker image/layer data in /var/lib/docker will hit the 4Gi cgroup limit first and can OOM-kill dockerd during larger pulls/builds. This makes the intended 8Gi cache unusable and can cause benchmark jobs on 4-gpu-h100 runners to fail as soon as Docker state exceeds ~4Gi.

Useful? React with 👍 / 👎.

env:
- name: DOCKER_TLS_CERTDIR
value: "" # Disables TLS for shared socket use
- name: DOCKER_DRIVER
value: overlay2
volumeMounts:
- name: docker-sock
mountPath: /var/run
Expand Down