Skip to content

test: bring more of consumer/smoke and provider/smoke under the default smoke run - #98

Merged
saqibmanan merged 2 commits into
CIfrom
smoke-marker-coverage
Sep 17, 2026
Merged

saqibmanan merged 2 commits into
CIfrom
smoke-marker-coverage

Conversation

@saqibmanan

Copy link
Copy Markdown
Contributor

Option 1 from our discussion: fix the marker mismatch rather than fold everything into the default gate. Directory placement and the smoke marker had drifted apart significantly — this closes that gap for a specific, chosen set, not blanket-wide.

consumer/smoke/ — all of it, now

test_con_003/005/007 and test_publishers_flow (download-verification tests) were functional-only, so they sat in tests/consumer/smoke/ but never ran under the default -m smoke path. Added smoke alongside the existing functional — additive, so full/functional suite selections are unaffected.

Verified: 61/61 passed, -n 4, 4m18s against dev — every test in the directory now runs under -m smoke.

provider/smoke/ — 8 specific files, chosen deliberately

Added smoke (kept functional) to: test_prv_002_dashboard_components (new, 4 read-only visibility checks, no writes), 006, 006b, 007, 009, 011, 012 — the org-scoped create/edit flows, plus the pre-existing 001 login test.

Deliberately not included: the individual-scoped duplicates (002, 002b, 003), the two xfail chart tests (004, 008 — feature not fully built), 005 (edit profile — separately flagged, hardcodes a name assuming one specific account) and 010 (individual collaborative). This is a chosen subset, not derived from a rule.

Verified against dev, -n 3, the 8 files now selected: 9 passed, 2 failed (test_prv_006, test_prv_007). Both confirmed load-contention, not real bugs — each passes cleanly alone (99s and 145s respectively). This is the same class of flakiness already documented on the account-parallelism PR (#97): concurrent provider testing against the shared dev backend is measurably less reliable than serial, independent of which specific files are running that round — a different failing pair each time. Not something this PR fixes; a CI-concurrency or backend-capacity question, not a marker one. Flagging it here rather than silently shipping around it.

No workflow YAML change

run-smoke.yml's case-statement already scans the whole domain tree by marker (p="tests/consumer" / p="tests/provider", m="smoke") for every suite selection — the smoke/ folder name has never actually scoped anything on its own. So adding the marker to these specific tests is sufficient by itself; nothing else to wire up.

test_con_003/005/007 and test_publishers_flow were functional-only,
so they never ran under the default -m smoke path despite sitting in
tests/consumer/smoke/ alongside tests that do. Additive -- functional
stays, so the full/functional suite selections are unaffected.

Closes out consumer/smoke entirely: all 61 tests in the directory now
run under -m smoke. Verified: 61 passed, -n 4, 4m18s against dev.
Adds smoke (kept functional) to the org-scoped create/edit flows --
006, 006b, 007, 009, 011, 012 -- plus a new dashboard_components.py
(4 read-only visibility checks, already smoke-only, no writes).

Deliberately NOT included: the individual-scoped duplicates (002,
002b, 003), the two xfail chart tests (004, 008 -- feature not fully
built), and 005 (edit profile) and 010 (individual collaborative) --
selection is user's, not derived from a blanket rule.

Verified against dev, -n 3, the 8 files now selected under -m smoke:
9 passed, 2 failed (test_prv_006, test_prv_007). Both confirmed
load-contention, not real bugs -- each passes cleanly alone (099s and
145s respectively). Same class of failure already documented on the
consumer-account-parallelism PR: concurrent provider testing against
the shared dev backend is measurably flakier than serial, independent
of which specific files are running. Not fixed here -- a CI-config or
backend-capacity question, not a marker one.
@saqibmanan
saqibmanan marked this pull request as ready for review September 17, 2026 19:45
@saqibmanan
saqibmanan merged commit 784af80 into CI Sep 17, 2026
3 checks passed
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