e2e: fix test resource isolation and make standalone parallelism configurable

[glours]

What I did

Several e2e tests used fixed host ports (8070, 8080, 8090, etc.), global buildx
builders registered with --use, and a hardcoded IPC container name. Under
parallel execution these daemon-level resources caused port conflicts,
buildkitd container-name collisions, and intermittent failures. Separately,
make e2e-compose-standalone hardcoded -parallel=1, forcing the entire
standalone suite to run serially even though the same tests run in parallel in
plugin mode.

Changes:

• Convert fixed host ports in four fixtures (network-test, sentences,
  build-test, volume-test) to ephemeral bindings. Add a ServicePublishedPort
  helper to framework.go to resolve the actual mapped port at test runtime.
• Give each buildx-builder test a unique daemon-scoped name via a BuilderName
  helper, preventing container-name collisions between parallel goroutines.
  Covers TestBuildPlatformsWithCorrectBuildxConfig, TestBuildPrivileged,
  TestBuildBuilder, TestBuildEntitlements, TestBuildTLS.
• Replace -parallel=1 in make e2e-compose-standalone with
  E2E_STANDALONE_PARALLEL?=4, aligning standalone with plugin mode on
  4-core GitHub Actions runners. E2E_PARALLEL_PLUGIN?=4 makes the
  plugin-mode knob explicit too. Both are overridable (E2E_STANDALONE_PARALLEL=1
  restores serial local debugging).
• Raise the poll delay in watch_test.go long-running polls (90 s / 120 s
  timeout) from the default 100 ms to 1 s; bump cascade_test.go and
  publish_test.go from 100 ms to 500 ms. Timeouts are unchanged.

The logging-driver fixture keeps its fixed port 24224: the Docker fluentd
log driver encodes fluentd-address at service-start time, so ephemeral
resolution would require a multi-step compose up. TestLoggingDriver is
serial, so there is no collision risk in practice.

Related issue

No related issue — this is a test-infrastructure improvement.

(not mandatory) A picture of a cute animal, if possible in relation to what you did

A panda who has learned not to share its bamboo with parallel neighbours 🐼

[docker-agent]

Assessment: 🟢 APPROVE

This PR correctly implements two waves of e2e CI optimisations. The diff was reviewed end-to-end across all 17 changed files.

Key changes verified:

• ServicePublishedPort — Correctly omits -f in its compose port call. docker compose port resolves the project purely from the daemon by the com.docker.compose.project label when -p is provided (toProjectName returns early, backend.Port() never reads a compose file). The net.SplitHostPort parse handles both IPv4 and IPv6 address formats produced by Compose. t.Fatalf calls runtime.Goexit() so the zero-value return after it is unreachable — no silent zero-port risk.

• BuilderName — Sanitisation and 60-char truncation are correct. All five call sites pass a top-level t (t.Name() = function name, e.g. TestBuildPlatformsWithCorrectBuildxConfig = 40 chars): the longest composed name is 56 chars, well below the truncation threshold. The resulting names are globally unique across parallel tests, compatible with Docker container-name and context-name character rules.

• IPC fixture — ipc: "container:ipc_e2e-src" matches exactly what the test creates (projectName+"-src" = "ipc_e2e-src"). The --rm flag on the source container and the final rm -f in cleanup coexist safely because the latter's result is discarded.

• Makefile — -parallel=$(E2E_PARALLEL_PLUGIN) and -parallel=$(E2E_STANDALONE_PARALLEL) are correctly positioned before ./pkg/e2e (test flags before package). --tags=standalone is also before the package path. TEST_FLAGS is only appended with -run/-skip, so no double -parallel issue in practice.

• Ephemeral ports — All removed fixed-port bindings (8070, 8090, 9090, 90, 95, 8080 in fixtures) have corresponding call-site updates to ServicePublishedPort or relaxed assertions (0.0.0.0: vs 0.0.0.0:8080). No residual hardcoded port numbers remain for the changed fixtures.

• Poll-delay calibrations — Raising delays from 100 ms → 500 ms–1 s reduces compose exec storm during long-running watch polls without changing test semantics.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[docker-agent]

🟢 No issues found — LGTM! View logs.