test(InsightsTest): match retrySearchUntil threshold to asserted asset count

[sriram-atlan]

Summary

Two surgical fixes to make the daily Test (leangraph-test) workflow deterministic. Both target eventual-consistency races against the ES index that surface only under the workflow's parallel matrix on a shared tenant.

Test	Failure signature	Fix
InsightsTest.searchAssets	expected [3] but found [2] on typeName aggregation bucket count	Bump retrySearchUntil(index, 3L) → 4L so retry waits for all 4 assets to be indexed before evaluating the aggregation
SuggestionsTest.findSuggestions*	expected [1] but found [0] on ownerGroups / descriptions / assignedTerms suggestion lists	Extend awaitConsistency() to retry-search for the peer columns until their non-tag metadata is also visible in ES, not just tags

Together these address MS-1269 and MS-1270.

Root-cause walkthrough

InsightsTest.searchAssets

The test creates 4 assets and asserts that the typeName term-aggregation has 3 buckets (AtlanCollection, Folder, AtlanQuery). It also asserts entities.size() == 4 and approximateCount == 4L further down.

But it called retrySearchUntil(index, 3L) — retry until hits ≥ 3. When the 4th asset (typically the most-recently-created AtlanQuery) hasn't been indexed yet, the retry stops at 3 hits but only 2 distinct typeName values are present, so the bucket-count assertion intermittently fires. Aligning the retry threshold to the asset count closes the race.

SuggestionsTest.awaitConsistency

awaitConsistency() currently calls waitForTagsToSync(taggedAssetGuids, log), which covers __classificationNames and __traitNames but not ownerGroups, description, userDescription, or __meanings. The Suggestions API aggregates these additional fields from peer columns. Under parallel matrix load on a shared tenant, the ES outbox processor (30-second idle poll) hasn't drained the peer's non-tag updates by the time findSuggestionsDefault runs, so all aggregations come back empty.

Diagnostic evidence:

• Local isolated run: SuggestionsTest passes 24/24 in 1m 39s against leangraph-test
• CI parallel run on the same tenant: 0 aggregation buckets for owner groups
• [ms-1268-trace] server logs (separate diagnostic branch) confirm the peer column's __meanings and ownerGroups are correctly persisted in Cassandra and emitted into the ES bulk update body — they just hadn't arrived in ES yet when the test queried

The extension issues an explicit search that filters Columns named COLUMN_NAME1 with all four metadata fields existing, retrying until both metadata-bearing peers (t1c1 and v1c1) are visible. retrySearchUntil already encapsulates exponential backoff and bounded retries.

What this is not fixing

• MS-1267 (CategoryPreProcessor parent-anchor) — that was a real lean-graph product bug, already merged to switchable-graph-provider as #6721
• MS-1268 (term mutation response certificateStatus) — that was a cascade of MS-1267; resolved once MS-1267 was deployed

After this PR plus the existing PurposeTest / asset-import token-permission gaps that are environment-specific, the Test (leangraph-test) workflow should hold steady.

Test plan

• CI green on this PR (PR build / unit tests)
• After merge: re-run Test (leangraph-test) workflow, confirm Integration (InsightsTest) and Integration (SuggestionsTest) are green for 3 consecutive runs
• Confirm the non-leangraph nightly Test workflow still passes both jobs (no regression on the other tenant)

🤖 Generated with Claude Code