[Chore][CI] Split k3 multiprocess tests into parallel pipeline steps#2914
[Chore][CI] Split k3 multiprocess tests into parallel pipeline steps#2914sammshen merged 1 commit intoLMCache:devfrom
Conversation
There was a problem hiding this comment.
Code Review
This pull request refactors the multiprocess CI tests into parallel Buildkite steps and introduces run-single-test.sh for individual test orchestration. Feedback identifies a likely typo in the model name, the omission of step priority in the new pipeline configuration, and a recommendation to use stricter shell options for script robustness.
| export VLLM_BASELINE_PORT="${VLLM_BASELINE_PORT:-9000}" | ||
| export MAX_WAIT_SECONDS="${MAX_WAIT_SECONDS:-300}" | ||
| export BUILD_ID="${BUILDKITE_BUILD_ID:-local_$$}" | ||
| export MODEL="${MODEL:-Qwen/Qwen3-14B}" |
There was a problem hiding this comment.
Qwen/Qwen3-14B appears to be a typo, as Qwen 2.5 is the current version and Qwen 3 has not been released. Please verify if Qwen/Qwen2.5-14B was intended to avoid CI failures during model download.
| export MODEL="${MODEL:-Qwen/Qwen3-14B}" | |
| export MODEL="${MODEL:-Qwen/Qwen2.5-14B}" |
| timeout_in_minutes: 30 | ||
| agents: { queue: "k8s" } |
There was a problem hiding this comment.
The priority: 1 setting from the original step has been omitted in the new parallel steps. This may cause these tests to be scheduled with lower priority than before. Consider adding it back to each step in the group to maintain consistent CI performance.
timeout_in_minutes: 30
priority: 1
agents: { queue: "k8s" }| # | ||
| # Each invocation is self-contained: launches servers, runs one test, cleans up. | ||
| # This mirrors the comprehensive tests' run-single-config.sh pattern. | ||
| set -o pipefail |
There was a problem hiding this comment.
For better robustness and consistency with run.sh, use set -euo pipefail. This ensures the script exits immediately if a command fails or an unset variable is used, which is particularly important for an orchestrator script.
| set -o pipefail | |
| set -euo pipefail |
…MCache#2914) [CI] Split k3 multiprocess tests into parallel pipeline steps Co-authored-by: Samuel Shen <slshen@uchciago.edu>
…MCache#2914) [CI] Split k3 multiprocess tests into parallel pipeline steps Co-authored-by: Samuel Shen <slshen@uchciago.edu>
No description provided.