Skip to content

fix(catalog): close workload state gaps found by e2e recording - #283

Draft
AviadHayumi wants to merge 1 commit into
e2e/flow-grovefrom
e2e/catalog-gaps
Draft

fix(catalog): close workload state gaps found by e2e recording#283
AviadHayumi wants to merge 1 commit into
e2e/flow-grovefrom
e2e/catalog-gaps

Conversation

@AviadHayumi

Copy link
Copy Markdown
Contributor

Part of the e2e flow recording stack. Reordered so flows independent of the catalog fix (#262) come before it.

Fourteen states operators actually write that the built-in definitions read
as Undefined, found by driving real workloads on live clusters and
classifying every observed CR from its own fields:

- batch Job: Completed misses SuccessCriteriaMet; Failed misses FailureTarget
- Deployment: just-created (Progressing before Available exists)
- StatefulSet: scale-down (ready above spec.replicas)
- JobSet: just-created and succeeded-pending windows; ready-flap in Running
- LeaderWorkerSet: startup before Available exists; scale-down replica lag
  (Running now keys on Available=True, the reason field is never extracted)
- RayCluster: no Initializing at all (provisioning and resume windows)
- RayJob: provisioning before jobStatus exists; the Suspending transient
- Knative Service: only Running was mapped; deploying and failed now read
- KServe InferenceService: the whole deploy phase before Ready is decided
- Milvus, DynamoGraphDeployment: the just-created empty phase
- NIMService: every non-terminal phase (empty, PVC-Created, ...)

Each fix was re-recorded on a live cluster and the replay golden asserts
every recorded frame through Karta. The full report with exact CR statuses
and kread reproductions is test/e2e/KARTA_GAPS.md on the e2e branch.

Signed-off-by: aviadh <aviad.hayumi@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant