Chen Sun
Chen Sun
Hi @droctothorpe feel free to contribute ? JFYI, here're some example code on how we tested the v2 Vertex client (which was moved out of KFP package in the master...
/lgtm /approve Thanks!
As you found out this is currently by design, I can see arguments from both side. We can probably keep this issue for seeking opinions.
Thank you for the fix. /lgtm and leave it to @IronPan for approval.
Hi @patrickmead, are you using KFP SDK v1.8.*? You can still set memory limit for components load from file. Example: https://github.com/kubeflow/pipelines/blob/d5bc8ddd6250d90b38cff5759e856f73e71e7d03/sdk/python/kfp/v2/compiler_cli_tests/test_data/pipeline_with_resource_spec.py#L25-L44 `set_memory_request` is not available/applicable for Vertex AI though.
I'm not entirely sure about v1 behavior, but in v2, we don't make any assumption on whether an artifact has content in its path. For instance, `Metrics` artifact has nothing...
No, I think it's by design. If users deletes artifacts, shouldn't that be considered user error? The solution can be run again with cache option disabled.
Discussed this in today KFP community meeting, and I want to summarize it here: >> In the past, the kfp project has also not considered this a user error and...
> I think the difference is that the user is not actively "relying on caching". For the user it just happens automatically when using the same component definition and arguments....
> > In summary, I'm supportive of the following options: > > ``` > > * Make artifact deleting a KFP feature so that KFP should do the full cleanup...