-
[OPIK-6810] [BE] add project_name to experiment_items_bulk to avoid Default Project fallback (#7021)
frostbyte_neo 发布于
2026-06-12 09:45:30 +00:00 - [OPIK-6810] [BE] add project_name to experiment_items_bulk to avoid Default Project fallback
Bulk-ingested items that use evaluate_task_result (no trace) had their
auto-created hidden traces hardcoded to the Default Project, while the
experiment lived in its own project. The async aggregation job then derived
the aggregate's project_id from those traces, and the find query served the
experiment from the aggregate path filtered by that project — so the
experiment disappeared from its own project view a few minutes after upload.- Add optional project_name to ExperimentItemBulkUpload; apply it to traces
that don't carry their own project (auto-created or project-less explicit
traces). - Keep the Default Project fallback when project_name is omitted, but return
an X-Opik-Deprecation header only when a fallback actually occurred. - Reject a mismatch between request project_name and an item trace's
project_name (422) via a bean-validation annotation. - Fix a pre-existing NPE in bulk validation when a trace is present with null
spans. - Experiment items now carry the resolved project_id (was null on the
no-trace path).
Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- [OPIK-6810] [BE] use project_name when bulk creates the experiment and dataset
The experiment built in the bulk path carried no project, so a bulk-created
experiment (and its auto-created dataset) landed with no project. Set
projectName on the experiment so ExperimentService.create resolves/creates the
project and passes its id to getOrCreateDataset — experiment and dataset are
now created in the requested project.Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- [OPIK-6810] [BE] make bulk spans inherit their trace's project
Spans in experiment_items_bulk only kept source=EXPERIMENT and their own
project_name, so a span could land in a different project than its trace
(project_name isn't even part of ExperimentItemBulkWriteView, so it is never
received on this path). Force each span into its trace's resolved project in
splitBatches, and drop the now-redundant span projectName assignment in
ExperimentItemBulkMapper (it still sets traceId for the trace-id match check).Also: validate project_name with @Pattern(NULL_OR_NOT_BLANK); drop unused
workspaceId param from validateExperimentConsistency; log projectName.Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- [OPIK-6810] [BE] reject bulk upload when project_name conflicts with existing experiment's project
Address PR review: when experimentId references an existing experiment,
validateExperimentConsistency now rejects (409) a non-blank request project_name
that doesn't match the experiment's project, preventing items from being split
across projects. Also strengthen the dataset test to assert the dataset is
created in the requested project (project_id), not just by name.Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- [OPIK-6810] [BE] dedupe bulk tests; document 409 as Conflict
- Extract shared test helpers in the BulkUpload suite (createDatasetWithItem +
BulkDataset record, newExperimentName, newProjectName) and apply them across
the bulk tests, removing the repeated dataset/item setup and name generation. - experiment_items_bulk 409 ApiResponse description: "Experiment dataset
mismatch" -> "Conflict" (now also covers project mismatch).
Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
- [OPIK-6810] [BE] move span/trace-id guard into hasMismatchingTraceId
Address review: fold the empty-spans guard into the helper (renamed
hasMatchingTraceId -> hasMismatchingTraceId since it returns true on a
mismatch) so validate() reads cleanly and the helper is null-safe on its own.Co-Authored-By: Claude Opus 4.8 (1M context) noreply@anthropic.com
Co-authored-by: Claude Opus 4.8 (1M context) noreply@anthropic.com
下载附件