fix(ci): restore E2E flakiness fixes for pgschema, docker-pull, and spec timing#1396
Merged
Conversation
…pec timing Three chronic Desktop E2E Integration failure modes in one pass: Bucket A (regression): restore PGSCHEMA_PLAN_* env vars at both pgschema apply sites. PR #1070 (e3736f0) added these vars so pgschema uses the already-running docker Postgres instead of downloading an embedded binary from Maven Central at test time. PR #963 (ff824a3, merged the next day) silently clobbered them via a stale-base rebase. The embedded-PG download is the transient 'no version found matching 17.5.0' flake that matches Will's 'flaky for ~2 weeks' window exactly. Bucket B (transient infra): wrap both 'docker compose up' service-start steps in a 3-attempt retry loop with backoff. A single Docker Hub 500 previously killed the entire job before any test ran. Bucket C (spec timing race): stabilize the 'Switch to Default' step in persona-env-vars.spec.ts. The radix dropdown re-renders mid-click (detaches from DOM) when switching back to Default after Anthropic. Extract the defaultOption locator and await expect(defaultOption).toBeVisible() before clicking — Playwright's auto-wait resolves the 'element is not stable' timeout that was failing all 3 retries. Co-authored-by: Will Pfleger <pfleger.will@gmail.com> Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
wesbillman
approved these changes
Jun 30, 2026
…etach race
The 'persona model options follow the selected LLM provider' test was
failing on the 'Switch to Default' step with 'element is not stable' /
'element was detached from the DOM'. The prior fix added
await expect(defaultOption).toBeVisible() before .click(), but
Playwright's .click() already auto-waits for visibility and stability,
so the assertion did not change the actionability window. A separate
visibility assertion can pass on one mounted instance, then the menu
item can detach before the subsequent click lands while Radix is still
re-rendering the collection after the provider change.
Fix: mirror the OpenAI and Anthropic steps in the same test, which both
use the bare page.getByRole('menuitemradio').click() pattern and pass
reliably. After await llmProvider.click(), call waitForAnimations(page)
to let Radix finish animating/re-mounting the menu collection, then
locate the 'Default' menuitemradio fresh at click-time. No stale handle
crosses the open/re-render boundary.
Co-authored-by: Will Pfleger <pfleger.will@gmail.com>
Signed-off-by: Will Pfleger <pfleger.will@gmail.com>
wpfleger96
pushed a commit
that referenced
this pull request
Jun 30, 2026
…work * origin/main: (25 commits) fix(thread): stop mid-scroll content jump in live threads (#1397) fix(ci): restore main to green — tauri fmt, personas.rs file-size split, Windows path test (#1399) fix(desktop): enable buzz-dev-mcp MCP server for Codex agents (#1394) fix(ci): restore E2E flakiness fixes for pgschema, docker-pull, and spec timing (#1396) fix(personas): persist pack-backed persona UI edits across reboot (#1392) fix(buzz-acp): clear steer_rx on all run_prompt_task exit paths (#1391) Restore channel date divider rule (#1395) Speed up profile wave action (#1379) Restore visible links for rich previews (#1378) Mobile channel list polish (#1367) style(desktop): unify corner radii to rounded-2xl (16px) (#1393) fix(desktop): skip keychain write when blob contents are unchanged (#1377) fix(desktop): stop clipping the agent-activity row under the composer (#1371) Constrain macOS overscroll to conversations (#1317) Mobile appearance foundation (#1366) chore(release): release Buzz Desktop version 0.3.38 (#1375) feat(desktop): provider-agnostic model selection + databricks discovery (#1307) release(helm): buzz chart 0.1.1 (#1374) Harden relay attack surfaces (#1369) ci(helm): publish chart to GHCR on chart-v* tags (#1372) ...
This was referenced Jun 30, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Fixes three distinct root causes behind the recurring "Desktop E2E Integration" CI failures. All three buckets in one commit on
duncan/e2e-flakiness-fixoff2f496debc.Bucket A — restore
PGSCHEMA_PLAN_*env block (.github/workflows/ci.yml)PR #1070 (
e3736f08b) added a 5-var block pointingpgschema applyat the already-running docker Postgres instead of downloading an embedded one from Maven Central. PR #963 (ff824a365) silently deleted those lines via a stale-base rebase. Restored at both pgschema-apply sites (ci.yml:396-400and547-551). Eliminates thefailed to start embedded PostgreSQL: no version found matching 17.5.0flake deterministically.Bucket B — retry loop for
docker compose up(.github/workflows/ci.yml)Both "Start integration services" steps ran
docker compose up -d postgres redis minio minio-initwith no retry. A single Docker Hub 500 killed the job pre-test. Wrapped both sites in a 3-attempt retry with 5s/10s backoff. Persistent failure still exits 1 after 3 attempts.Bucket C — stabilize radix detach race (
desktop/tests/e2e/persona-env-vars.spec.ts)The
persona model options follow the selected LLM providertest failed on the "Switch to Default" step withelement is not stable/element was detached from the DOM. The prior approach built a filtereddefaultOptionlocator before opening the menu and held it across the open→re-render boundary — Radix was still re-mounting the collection when the click landed.Fix mirrors the OpenAI and Anthropic steps in the same test: after
llmProvider.click(), callwaitForAnimations(page)to let Radix finish animating/re-mounting, then locatemenuitemradiofor "Default" fresh at click-time via the barepage.getByRolepattern. No stale handle crosses the open/re-render boundary.Gate
Buckets A/B/C are not exercised by
just desktop-tauri-test(828/828 passing locally). Real proof is the live E2E Integration CI shard.