Run heavy color blocks in a serial Nala lane (fix shard-1 contention)

[hadobe]

Draft — implements Option 1 from the flakiness discussion.

Problem

The color-shared blocks (color-contrast-checker, color-explore, color-extract, color-headline, color-search-marquee, color-wheel) are heavy Spectrum canvas/image-processing blocks. Playwright shards by contiguous test order, so the alphabetically-adjacent color-* blocks all land in one shard and run 3-at-a-time (workers: 3), where they starve each other's decoration and time out in waitReady() (30s wait for data-block-status="loaded").

Reproduced locally: they pass solo and serial, fail clustered — pure concurrency contention, not a block bug. They already have martech=off, so it isn't trackers either.

Fix — isolate color into its own serial lane

• Each of the 4 shard lanes now excludes @color (NALA_EXTRA_GREP_INVERT), so the non-color suite (215 tests) partitions cleanly 4 ways with no color cluster.
• A new 5th matrix lane runs only @color with workers: 1 (serial), so the heavy color blocks never run concurrently.

Workers stay 3 on the shard lanes — no per-shard worker tuning. The color lane is part of the same run-nala-tests matrix, so the existing aggregate gate (needs: run-nala-tests) covers it with no branch-protection change.

pr.run.sh

• Collapse the two --grep-invert flags into one combined regex — Playwright only honours the last --grep-invert, so nopr was effectively dead; now nopr|@monitoring[|@color] all apply. Extendable via NALA_EXTRA_GREP_INVERT.
• Add NALA_GREP for a positive lane-restricting grep (the color lane).

Verified locally (against stage)

• Color lane (-g @color --workers=1, all 3 browsers): 24 passed / 6 skipped, ~1.5 min — the exact blocks that were failing under concurrency.
• Shard lanes exclude color: 0 color tests, 215 non-color split 4×54.
• Workflow YAML valid (5 lanes); pr.run.sh syntax OK; both lane commands construct correctly.

Cost

The color lane runs in parallel with the 4 shards (~2–4 min serial), so it doesn't add to gate wall-clock. Independent of #704 (tracker rollout) — different files, merge in any order.

Follow-up option (not here): now that the heavy color blocks are out of the shards, the shard lanes could likely go back to workers: 4 for speed — worth trying once this lands.

🤖 Generated with Claude Code

[aem-code-sync]

Hello, I'm the AEM Code Sync Bot and I will run some actions to deploy your branch and validate page speed.
In case there are problems, just click a checkbox below to rerun the respective action.

• Re-run all PSI checks
• Re-run failed PSI checks
• Re-sync branch

Commits

• dcbcb8a ✅ (latest)