For live reload, retry if model is not ready #213
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
bolasim
approved these changes
Feb 8, 2023
| retry=( | ||
| retry_if_exception_type(ConnectionError) | ||
| | retry_if_exception_type(ModelNotReady) | ||
| ), | ||
| stop=stop_after_attempt(INFERENCE_SERVER_START_WAIT_SECS), |
There was a problem hiding this comment.
should we increase this timeout if we're also waiting from the model to be ready
There was a problem hiding this comment.
Good point. I think the tradeoff is that for some error cases it may increase the time to return an error. Let's start with this and then we can tune. This will not completely eliminate the issue, models can take any amount of time to load. In those cases it's fine to return the 503 error. Idea is that we retry a bit for the common simple cases.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
We were retrying if we couldn't connect to the inference server, but not when we could connect but the model wasn't ready. We do that now. The effect of this issue was that if model took a long time to load, then the requests until then would fail. This leads to a bad user experience when calling predict right after applying a patch for a slow loading model. This retry behavior is consistent with non-reloadable models, where, if the model is deploy via kserve or knative, the request would wait until the model is ready.
Please note that all files under
truss_container_fsare auto-generated and can be ignored for this review. Added a test for this case as well.