[sentry-junior]

Follow-up to #1512 and #1583.

#1512 fixed the counts sentry issue list --query 'release:"X"' displays: after building the listing, it refetches each issue's counts from the events endpoint. But in multi-project listings (DSN auto-detect, project search across orgs), the CLI merges the per-project results and trims them to --limit before that refetch runs. So --sort freq / --sort user picked the rows using period-wide counts. That could drop the issues with the most events in the release and keep ones that barely appear in it. The re-ranking afterwards only reorders rows that already survived the trim.

Root cause. The issues endpoint already returns release-scoped counts. StreamGroupSerializerSnuba builds conditions from the query's event-level search filters, and release isn't in skip_snuba_fields. When filtered isn't collapsed, it returns filtered.count / filtered.userCount computed with those conditions. The search executor's freq/user sort aggregates over the same filtered events. So each server page is already ordered by release-scoped counts. The CLI just never read those numbers, because buildIssueListCollapse always collapsed filtered.

Sentry sources checked: src/sentry/api/serializers/models/group_stream.py, group.py (GroupSerializerSnuba.__init__, SKIP_SNUBA_FIELDS), and src/sentry/search/snuba/executors.py (sort_strategies, aggregation_defs).

Change

• buildIssueListCollapse gains shouldCollapseFiltered. It defaults to true, so other callers like feedback are unchanged.
• When shouldScopeCounts is true (a release: filter, and the counts are shown or drive the sort), issue list stops collapsing filtered. It also keeps stats, because the server only returns filtered when stats aren't collapsed.
• applyQueryScopedCounts copies filtered.count/filtered.userCount into count/userCount right after each fetch: per target, org-wide, and cursor resume. That happens before any merge, sort, or trim. It then drops filtered, so the --json output shape doesn't change.
• Removes the per-issue events fan-out (issue-scoped-counts.ts, applyScopedCounts, and the project option on ExploreQueryOptions that only it used). That's one fewer request per listed issue.

Tests

• test/commands/issue/list.test.ts:
  • collapse params with and without release:
  • filtered counts promoted into count/userCount, and filtered: null falling back to the top-level counts
  • --sort freq --fields id,title keeping stats
  • a multi-project --sort freq --limit 2 merge/trim regression test
• The new tests fail on main (4 of 5; the no-release case passes either way) and pass with this change.
• test/lib/issue-collapse.property.test.ts: a property test for shouldCollapseFiltered.

Verification

• tsc --noEmit and pnpm run lint (oxlint + oxfmt): clean.
• pnpm run test:unit: 11188 passed and 6 failed. All 6 failures are permission tests (EACCES/EPERM, read-only DB, unreadable files) in completions, telemetry, upgrade, code-scanner, and file-changes. The same 6 fail on unmodified main in this sandbox.
• Not run: e2e tests and a live API check against a real org.

via BYK.

--

View Junior Session [Sentry]

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
sentry-local	Skipped		Oct 22, 2026 10:57am UTC