Repository navigation
Scheduler: migrate the e2e tests to Playwright - #36357
Open
ajivanyandev wants to merge 16 commits into
Open
ajivanyandev wants to merge 16 commits into
ajivanyandev wants to merge 16 commits into
Conversation
In month view the test renders 11,600 appointments, and on a loaded CI agent each render keeps the page's main thread busy for 5-6s. Nothing can be checked in the page during that time, so the default 5s expect timeout could run out before the first check: in waitForViewportRender after scrollTo, or in toHaveScreenshot on the start step, when the screenshot began just before the appointments rendered. Give waitForViewportRender a 20s timeout and wait for the appointments to render before the start screenshot.
The Material themes import Roboto from Google Fonts and fall back to the Roboto they bundle. What Google serves is outside our control: in the failing CI runs it returned one variable font file for every weight, and the weight-500 text of the appointment form toolbar (the title and the Save and Cancel buttons) rendered a few pixels differently. The "appointment form with labelMode=static" screenshot in the material job failed this way in 5 of 123 runs, always with the same image. Answer the Google Fonts stylesheet requests with an empty stylesheet, so the tests do not depend on the network and always use the bundled font.
Now and then Chrome on the CI agents never paints a page it has just opened: the page runs scripts and answers evaluate calls, but produces no frame, so every wait that needs one times out - "waiting for element to be stable" in a screenshot, about a second per mouse move in a drag. The traces of 13 such failures since the Scheduler migration show no frame after the navigation, no failure screenshot, and no pending or slow request; the browser rendered the tests before them in the same worker. It is about one page in 30,000, so roughly every tenth run of the Scheduler jobs failed on it. Retry a failed test once on CI. Playwright runs the retry in a new worker with a new browser, and lists a test that passes then as flaky. This mirrors the TestCafe jobs, which rerun a single failed test in the quarantine job. A test with a missing etalon would now pass on its retry against the file it has just written, so the job collects new etalons on every run and fails when there are any. Traces are uploaded on every run too, so a test that passed on its retry keeps the trace of its failed run. A run where many tests fail runs them twice, so the job gets 15 minutes instead of 10.
…boto The etalon was recorded with Roboto from Google Fonts. With the bundled Roboto, the "Grand Total" header (weight 500) differs in a few pixels; the rest of the image is the same. CI recorded the new etalon.
- Remove the fluent-next etalons: 26_1 has no fluent-next theme. - Remove the hierarchical grouping test: resources.parentIdExpr is 26_2-only. - Remove toolbar.allowKeyboardNavigation from the toolbar tests: the option is 26_2-only, and 26_1's toolbar behaves as if it were false. - Restore 26_1's expectations in the vertical grouping resize tests: the fix from DevExpress#34805 is not on 26_1. - Remove the four TestCafe Scheduler etalons that only 26_1 had.
Their TestCafe counterparts differ between 26_1 and main because of 26_2-only changes: DevExpress#35089 (appointment AA contrast, 460 etalons), DevExpress#34753 (multilevel grouping, 93), DevExpress#36007 (focused resource-colored appointments, 8) and DevExpress#35843 (1). CI records the 26_1 ones.
- 557 etalons dropped in the previous commit, as CI recorded them. - 79 etalons that main recorded after DevExpress#35089 (appointment text contrast) and that 26_1 renders without it. TestCafe's tolerance hid the change, so their TestCafe etalons were never re-shot on main. Every diff is in appointment text color only; both attempts rendered each of them pixel for pixel the same.
16 more etalons with only DevExpress#35089's appointment text color in them, from the tests that stopped at their first mismatch in the previous run. Both attempts rendered each one pixel for pixel the same.
The 5 grouping-by-date timeline etalons, and 12 more with only DevExpress#35089's appointment text color in them. Both attempts rendered each re-shot one pixel for pixel the same.
6 more etalons with only DevExpress#35089's appointment text color in them. Both attempts rendered each one pixel for pixel the same.
The Scheduler tests were 26_1's only TestCafe tests marked unstable. Without them, --onlyQuarantined matches nothing, TestCafe fails with "No tests match your filter" and the quarantine job goes red. Ports the guard that DevExpress#36088 added on main: check-should-run reports whether any test is quarantined, a check-quarantine job skips the quarantine job when there are neither quarantined tests nor flaky reports, and the quarantined run is skipped when nothing is quarantined.
ajivanyandev
force-pushed
the
e2e/scheduler-playwright-26_1
branch
from
November 18, 2026 10:56
4aa884e to
57a8df5
Compare
ajivanyandev
marked this pull request as ready for review
November 18, 2026 10:58
This branch has not been deployed
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
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.
Moves the Scheduler e2e tests from TestCafe to Playwright on 26_1, as #36071, #36196 and #36212 did on main, together with the flaky-test fixes from #36310.
Stacked on #36356. Until it merges, this PR also shows its 3 commits.
Cherry-picks
26_1 changes
resources.parentIdExpris 26_2-only.toolbar.allowKeyboardNavigation: falseremoved from the toolbar tests. The option is 26_2-only, and 26_1's toolbar already behaves that way.unstable, so without the guard--onlyQuarantinedmatches nothing and the job fails with "No tests match your filter".Etalons
All differences come from 26_2-only changes that aren't on 26_1, so 26_1 renders those screenshots differently:
All other etalons are main's and pass unchanged on 26_1.
Test plan
lintandtype-checkpass fore2e/playwrightagainst 26_1's typings; the Scheduler specs use no other 26_2-only Scheduler or Toolbar options.