ci(checkpoint): mount shared model-cache PVC in DynamoCheckpoint tests - #12400
Conversation
WalkthroughCheckpoint deployment specifications and CI workflows now support an optional shared model-cache server, path, and PVC. The test helper mounts the PVC and adjusts ChangesShared model cache integration
Estimated code review effort: 3 (Moderate) | ~20 minutes 🚥 Pre-merge checks | ✅ 4 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (4 passed)
Comment |
There was a problem hiding this comment.
🧹 Nitpick comments (1)
tests/deploy/test_dynamocheckpoint.py (1)
276-279: 📐 Maintainability & Code Quality | 🔵 Trivial | ⚡ Quick winAdd a regression test for cache environment precedence.
Verify that a PVC removes container-level
HF_HOME, sets deployment-levelHF_HOMEto the mount path, and adds the PVC mount; also verify that the no-PVC path preserves TRTLLM’s snapshotHF_HOME.Also applies to: 293-296
🤖 Prompt for AI Agents
Verify each finding against current code. Fix only still-valid issues, skip the rest with a brief reason, keep changes minimal, and validate. In `@tests/deploy/test_dynamocheckpoint.py` around lines 276 - 279, Extend the deployment tests around the container environment filtering and mount_model_cache_pvc flow to cover both cache modes: with a model cache PVC, assert container-level HF_HOME is removed, deployment-level HF_HOME points to the mount path, and the PVC mount is present; without a PVC, assert TRTLLM’s snapshot HF_HOME remains unchanged.
🤖 Prompt for all review comments with AI agents
Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.
Nitpick comments:
In `@tests/deploy/test_dynamocheckpoint.py`:
- Around line 276-279: Extend the deployment tests around the container
environment filtering and mount_model_cache_pvc flow to cover both cache modes:
with a model cache PVC, assert container-level HF_HOME is removed,
deployment-level HF_HOME points to the mount path, and the PVC mount is present;
without a PVC, assert TRTLLM’s snapshot HF_HOME remains unchanged.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Path: .coderabbit.yaml
Review profile: CHILL
Plan: Enterprise
Run ID: 98c98466-4bd2-4626-9b3e-6bc21560cac9
📒 Files selected for processing (2)
.github/workflows/pr.yamltests/deploy/test_dynamocheckpoint.py
This comment has been minimized.
This comment has been minimized.
5980355 to
4c11947
Compare
|
/ok to test 4c11947 |
|
/ok to test 4c11947 |
4c11947 to
10a25da
Compare
|
/ok to test 10a25da |
10a25da to
1884688
Compare
|
/ok to test 1884688 |
|
/ok to test 1884688 |
1884688 to
3c7936f
Compare
|
/ok to test 3c7936f |
Checkpoint deploy CI was downloading Qwen via ModelExpress/HF on every run while regular deploy tests already used the Azure model-cache PVC. Wire the same provision + mount path through, and honor --model-cache-pvc in the checkpoint test harness (including skipping TRTLLM's container HF_HOME override so it cannot shadow the shared cache). Signed-off-by: Schwinn Saereesitthipitak <schwinns@nvidia.com>
3c7936f to
f2fbb78
Compare
|
/ok to test f2fbb78 |
Summary
model-cachePVC path regular deploy tests already use (setup-dynamo-operator+--model-cache-pvc).--model-cache-pvcintest_dynamocheckpoint.py, skipping TRTLLM containerHF_HOMEso it cannot shadow the shared cache.Validation
model-cacheand stop cold-fetching Qwen via HF/ModelExpress when the cache is populated.