direct: fix persistent drift on model_serving_endpoints#5708
Open
pietern wants to merge 2 commits into
Open
Conversation
The backend always returns a defaulted config.traffic_config (100% to the served entity) even when the user does not specify one. #5693 dropped the suppression for it (and for budget_policy_id) assuming the new missing-in-remote auto-skip covered them, but both are present in RemoteType so the auto-skip never fires, and the direct engine reported a permanent update after every clean deploy. Suppress config.traffic_config as a backend_default (skip only when the user leaves it empty; real drift is still detected if it is set) and keep budget_policy_id under ignore_remote_changes (GET never echoes it). Make Changes.HasChange skip-aware so a suppressed-only change under a section no longer triggers a spurious section update (e.g. PUT /config on an unrelated email/tags update). Teach the testserver to default traffic_config like the backend so the running-endpoint acceptance test reproduces the drift locally. Co-authored-by: Isaac
Co-authored-by: Isaac
Contributor
Approval status: pending
|
Contributor
|
how did you find this? |
denik
reviewed
Jun 24, 2026
| ) | ||
|
|
||
| // HasChange checks if there are any changes for fields with the given prefix. | ||
| // HasChange checks if there are any actionable changes for fields with the given prefix. |
Contributor
There was a problem hiding this comment.
HasChange is used in a few places:
bind.go
dresources/app.go
dresources/cluster.go
dresources/model_serving_endpoint.go
dresources/sql_warehouse.go
dresources/vector_search_endpoint.go
are all of them safe to change this? should we PR this change separately (maybe we a dedicated regression test).
Collaborator
Integration test reportCommit: e52b568
28 interesting tests: 13 SKIP, 6 flaky, 6 RECOVERED, 3 FAIL
Top 17 slowest tests (at least 2 minutes):
|
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Changes
Restore suppression of the backend-defaulted
config.traffic_config(as abackend_default) andbudget_policy_id(ignore_remote_changes) onmodel_serving_endpoints, makeChanges.HasChangeskip-aware so a suppressed-only change doesn't trigger a spurious section update, and teach the testserver to defaulttraffic_configlike the backend so the drift is covered locally.Why
The backend always returns a defaulted
config.traffic_config(100% to the served entity) even when the user specifies none. #5693 dropped theignore_remote_changesentries fortraffic_configandbudget_policy_id, assuming the new "skip fields missing from RemoteType" auto-skip covered them — but both fields are present inRemoteType, so the auto-skip never fires and the direct engine reported a permanentupdateafter every clean deploy.traffic_configis a true backend default the user may also set, sobackend_default(skip only when the local value is empty; real drift still detected when set) is the right tool rather thanignore_remote_changes.budget_policy_idis never echoed by GET, so it stays underignore_remote_changes.Making the testserver faithful surfaced a latent over-update:
HasChangecounted skipped changes, so the suppressedtraffic_configtriggered a spuriousPUT /configon unrelated updates.HasChangeis now skip-aware (matchingHasChangeExcept).Tests
running-endpointnow reproduces the drift locally (fails without the fix). Passes on both engines on real cloud (aws-prod-ucws). Full./task test/fmt/lint/checkspass.This PR and its description were written by Isaac.