Update Buildkite pipeline for the new JuliaGPU cluster#53
Conversation
Switch to the per-backend queues of the new JuliaGPU cluster. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The JuliaGPU cluster now has a macOS agent with an Apple Silicon GPU, so Metal jobs no longer need a separate pipeline on the ecosystem cluster. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
CPU test ran on the upload queue is that intentional? |
No, that shouldn't have happened. |
| agents: | ||
| queue: "juliagpu" | ||
| if: build.message !~ /\[skip tests\]/ | ||
| timeout_in_minutes: 10 |
There was a problem hiding this comment.
The JuliaGPU CI environment is not intended for CPU testing. Please remove.
| benchmark: "*" | ||
| env: | ||
| BENCHMARK_GROUP: CPU | ||
| JULIA_NUM_THREADS: 1 |
There was a problem hiding this comment.
The JuliaGPU CI environment is not intended for CPU testing. Please remove.
The JuliaGPU CI environment is not intended for CPU testing. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
|
Thanks! |
|
No problem! I will merge and close this once the build completes |
| queue: "metal" | ||
| if: build.message !~ /\[skip tests\]/ | ||
| timeout_in_minutes: 10 No newline at end of file | ||
| timeout_in_minutes: 10 |
There was a problem hiding this comment.
@Dale-Black Could you also add this environment variable? It'll prevent needless precompilation of packages after Pkg commands that won't end up getting used because the tests launch with --check-bounds=yes and the precompilation happened with --check-bounds=auto
| timeout_in_minutes: 10 | |
| timeout_in_minutes: 10 | |
| env: | |
| JULIA_PKG_PRECOMPILE_AUTO: false |
Tests launch with --check-bounds=yes, so precompilation from Pkg commands under --check-bounds=auto would go unused. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## master #53 +/- ##
===========================================
+ Coverage 33.90% 96.12% +62.21%
===========================================
Files 3 3
Lines 233 129 -104
===========================================
+ Hits 79 124 +45
+ Misses 154 5 -149 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
3 participants
The JuliaGPU Buildkite agents have moved to a dedicated cluster, where the single
juliagpuqueue has been split into per-backend queues. Steps now select agents usingqueue: "cuda",queue: "rocm"orqueue: "oneapi"instead ofqueue: "juliagpu"combined with acuda/rocm/inteltag. Steps that targetedqueue: "juliagpu"without a backend tag now use the cluster's default queue.The separate
benchmarkqueue is also gone: benchmark agents now live in the backend queues with abenchmarktag, so the CPU benchmark steps and the combine step now targetqueue: "cuda"withbenchmark: "*".The Metal steps (test and benchmark) target
queue: "juliaecosystem", which belongs to a different cluster and can no longer be referenced from pipelines running in the JuliaGPU cluster. They have been moved verbatim to.buildkite/pipeline-julia.yml, to be wired up to a separate ecosystem-cluster pipeline once it is back online. Until then, the combine-benchmarks step no longer depends onmetal-benchmarks, so the benchmark aggregation temporarily loses the Metal results.🤖 Generated with Claude Code