[CI] e2e matrix (e2e / e2e-case *-sync-compose / e2e-storage) is flaky and blocks unrelated approved PRs #7768

Summary

The e2e aggregator and its e2e-case / e2e-storage children fail across many unrelated PRs. Because e2e is a required check, otherwise-approved PRs show red (BLOCKED / UNSTABLE) and cannot be merged.

Evidence

Failing checks aggregated across currently-approved open PRs:

Root cause 1 — artifact cascade from build-docker-images

e2e-case fails with (e.g. #7091):

##[error]Unable to download artifact(s): Artifact not found for name: shenyu-images

because the separate build-docker-images job failed (see the Maven-Central issue).

Root cause 2 — config-sync timing race

e2e-storage (#7668) creates a divide selector and immediately calls /anything, but the gateway has not yet synced the selector:

ERROR org.apache.shenyu.plugin.api.utils.WebFluxResultUtils - can not match selector data: divide , path is /anything

The selector is created and then deleted while assertions are still in flight.

Root cause 3 — external service / MQ timing

• e2e-storage assertions target the external httpbin.org.
• e2e-case (e2e-logging-rocketmq-compose) / ...-logging-rabbitmq-... flake on log-consumption timing (cf. closed [BUG] RocketMQ E2E can time out before its consumer acquires the access-log queues #7417).

Affected open PRs (sample)

#7038, #7080, #7088, #7091, #7183, #7188, #7189, #7191, #7192, #7198, #7236, #7387, #7513, #7666, #7668, #7699.

Proposed mitigations

• Decouple e2e from build-docker-images (build images inline or gate explicitly) so a Maven flake does not cascade.
• Replace the external httpbin.org dependency with the in-repo local httpbin fixture for all e2e suites (as introduced in [ISSUE #6461] Follow advertised WebSocket master addresses #7661).
• Ensure a unique RocketMQ/RabbitMQ consumer group per run + seekToBegin.
• Make the e2e aggregator non-required (the individual cases already carry signal) or auto-retry flaky cases.

Relates to #6281 (flaky-test tracking).