Raise the e2e timeout to 75 minutes

[sscarduzio]

The problem

timeout_minutes was 45, and it is per attempt. The suite needs more. Run 33853781633 was killed on spec 27 of 27.

The change

75 minutes, at all three retry steps: both jobs in all-e2e-tests.yml and the one in targeted-e2e-tests.yml.

What the number is based on

The nick-fields/retry action waits after every failed attempt, the last one included. With max_attempts: 2 a double timeout therefore costs 2×45+4 = 94 min at the old cap, and 2×75+4 = 154 at the new one.

45 was too short, but it is not what made these legs red. Of the 82 E2E legs that failed between 26 Aug and 5 Sep, exactly one reached the 94-minute floor. The other 81 finished below it, so at most one of their attempts timed out. The common shape is the 75–78 minute cluster: attempt 1 fails on its own at about 29 min, then attempt 2 is cut off at 45. Only the last attempt's error is reported, which is why sampling recent failures showed Timeout of 2700000ms hit — it hides why attempt 1 failed, and it denies attempt 2 the chance to finish.

Sizing: Tenancy.cy.ts alone measured 19:35, and the ES repo runs this same 27-spec suite under a 120-minute cap. Cost: each job is 4 versions × 3 envs, so the worst case moves from 94 to 154 min per leg, up to +720 runner-minutes per run. Neither job sets timeout-minutes, so two attempts fit inside the 360-minute default. max_attempts stays 2 — the retry is for the transients named at each step, not for the timeouts.

This does not turn the matrix green

It makes the suite report the real failure instead of a timeout, and lets a second attempt run to completion. The redness is spec flakiness, addressed in #118 and made visible by #120.

Today's runs show both halves of that, side by side. #124 carries develop into master, so its legs still run the 45-minute cap:

🔬 E2E Tests (released plugins) (9.4.6, docker)
##[error]Final attempt failed. Timeout of 2700000ms hit

This branch, at 75, reports the actual failure instead:

✖  Tenancy.cy.ts     02:07     17     1     4     -     12
✖  1 of 27 failed (4%)     30:25     82     52     4     14     12
error Command failed with exit code 4

Same leg, same day. One says "timeout", the other names the spec and the four tests. That is the whole point of the change.

Review notes

• Base branch is now master. docs/dev/branching.md in [RORDEV-2220] add branching, code style and writing style docs #125, case 4: a pipeline, workflow or runner change targets master, because master runs the nightly and the full matrix. Rebased onto master, so this PR carries only its own commits.
• The comments moved to the code they govern. The file used to open with an essay covering the history of the change and the drift between the two old copies. docs/dev/code-style.md asks for the current state at the code that enforces it, and leaves the story of the change to the commit message and the PR description — which is where the measurements above now live. Each retry step states only what the reader cannot see from the line: the cap is per attempt, the retry waits after the last attempt too, and the eck envs need the larger number.

🤖 Generated with Claude Code

Summary by CodeRabbit

• Chores
  • Increased automated end-to-end test time limits to accommodate slower test environments and longer-running test suites.
  • Updated workflow guidance to clarify expected execution times and retry behavior.

[Dzuming]

LGTM

[sscarduzio]

Merge order: this one first

#124 is red right now, and this PR is what unblocks it.

#124 carries develop into master, so its head still has the 45-minute cap — this raise is an open PR, not on either long-lived branch yet. Both of its failing legs are 9.4.6, docker, and they fail in the two different ways this PR is about:

leg	ends with
🔬 released plugins	Final attempt failed. Timeout of 2700000ms hit
🧪 pre-build plugins	4 of 27 failed (15%), exit code 4

2,700,000 ms is 45 minutes. So one leg reports the cap and says nothing about what went wrong, while the other, which got far enough to finish, names the specs.

Merging this PR to master first, then re-running #124, turns the first leg into a real result — a pass, or the name of the spec that failed. Merging #124 first leaves both branches on 45 and we learn nothing new from the re-run.

To be clear about what this does not do: Tenancy.cy.ts fails for its own reasons on the pre-build leg, and this PR does not fix that. It only stops the cap from hiding the evidence. The redness is spec flakiness, addressed in #118 and made visible by #120.