Remove `common.PORT` usage from `parallel` tests

[Trott]

• Version: master
• Platform: all
• Subsystem: test

Tests in parallel that use common.PORT risk getting EADDRINUSE if another test in parallel uses port 0 (to get an open port assigned by the operating system) at the same time the test runs. This appears to have happened recently. (See #12363 (comment).)

IMO, all instances of common.PORT in parallel should either be changed to use port 0 (if possible in the context of the test) or else moved to sequential (if using port 0 is not possible).

Here are the current tests that use common.PORT:

• test/parallel/test-cluster-basic.js (test: remove common.PORT from test-cluster-basic #12377)
• test/parallel/test-cluster-bind-twice.js (test: use dynamic port in test-cluster-bind-twice #12418)
• test/parallel/test-cluster-dgram-1.js (test: dynamic port in cluster worker dgram #12487)
• test/parallel/test-cluster-dgram-2.js (test: dynamic port in cluster worker dgram #12487) (test: replace common.PORT in test-cluster-dgram-2 #12492)
• test/parallel/test-cluster-dgram-reuse.js (test: replace port in cluster dgram reuse test #12901)
• test/parallel/test-cluster-disconnect-leak.js (Migrate cluster tests from common.PORT #12441)
• test/parallel/test-cluster-disconnect-race.js (Migrate cluster tests from common.PORT #12441)
• test/parallel/test-cluster-disconnect.js
• test/parallel/test-cluster-eaddrinuse.js
• test/parallel/test-cluster-inspector-debug-port.js
• test/parallel/test-cluster-ipc-throw.js
• test/parallel/test-cluster-master-error.js (tests: remove common.PORT from shared,master,rr tests #12451)
• test/parallel/test-cluster-master-kill.js (tests: remove common.PORT from shared,master,rr tests #12451)
• test/parallel/test-cluster-message.js
• test/parallel/test-cluster-net-send.js (tests: remove common.PORT from shared,master,rr tests #12451)
• test/parallel/test-cluster-process-disconnect.js (Migrate cluster tests from common.PORT #12441)
• test/parallel/test-cluster-rr-domain-listen.js (tests: remove common.PORT from shared,master,rr tests #12451)
• test/parallel/test-cluster-rr-ref.js (tests: remove common.PORT from shared,master,rr tests #12451)
• test/parallel/test-cluster-send-deadlock.js (tests: remove common.PORT from cluster send tests #12472)
• test/parallel/test-cluster-send-handle-twice.js (tests: remove common.PORT from cluster send tests #12472)
• test/parallel/test-cluster-server-restart-none.js
• test/parallel/test-cluster-server-restart-rr.js
• test/parallel/test-cluster-shared-handle-bind-error.js
• test/parallel/test-cluster-shared-leak.js (tests: remove common.PORT from shared,master,rr tests #12451)
• test/parallel/test-cluster-worker-disconnect-on-error.js (tests: remove common.PORT from cluster-worker-disconnect-on-error #12457)
• test/parallel/test-cluster-worker-disconnect.js (tests: remove common.PORT from cluster worker disconnect, exit and kill #12443)
• test/parallel/test-cluster-worker-exit.js (tests: remove common.PORT from cluster worker disconnect, exit and kill #12443)
• test/parallel/test-cluster-worker-kill.js (tests: remove common.PORT from cluster worker disconnect, exit and kill #12443)
• test/parallel/test-cluster-worker-no-exit.js (tests: remove common.PORT from shared,master,rr tests #12451)
• test/parallel/test-cluster-worker-wait-server-close.js (tests: remove common.PORT from cluster worker wait server close #12466)
• test/parallel/test-debugger-repeat-last.js (test: moved test-debugger-repeat-last.js from parallel to sequential #12470)
• test/parallel/test-dgram-bind-shared-ports.js (test: use dynamic port in test-dgram-bind-shared-ports #12452)
• test/parallel/test-dgram-close-in-listening.js
• test/parallel/test-dgram-close-is-not-callback.js
• test/parallel/test-dgram-close.js
• test/parallel/test-dgram-exclusive-implicit-bind.js
• test/parallel/test-dgram-implicit-bind-failure.js
• test/parallel/test-dgram-oob-buffer.js
• test/parallel/test-dgram-send-address-types.js (test: replace port in dgram send address types test #13007)
• test/parallel/test-dgram-send-callback-buffer-empty-address.js (test: replace port in dgram cd address empty test #12929)
• test/parallel/test-dgram-send-callback-buffer-length-empty-address.js (test: replace port in dgram cb address test #12944)
• test/parallel/test-dgram-send-callback-buffer-length.js (test: replace port in dgram cb length test #12943)
• test/parallel/test-dgram-send-callback-buffer.js (test: replace port in dgram cb test #12942)
• test/parallel/test-https-connect-address-family.js (test: replace port in https address family test #12915)
• test/parallel/test-net-better-error-messages-port-hostname.js (Remove common.PORT usage from parallel tests  #12473)
• test/parallel/test-net-better-error-messages-port.js
• test/parallel/test-net-connect-handle-econnrefused.js
• test/parallel/test-net-connect-immediate-destroy.js
• test/parallel/test-net-connect-immediate-finish.js
• test/parallel/test-net-connect-local-error.js
• test/parallel/test-net-listen-shared-ports.js
• test/parallel/test-net-localerror.js
• test/parallel/test-net-localport.js
• test/parallel/test-net-options-lookup.js
• test/parallel/test-net-reconnect-error.js
• test/parallel/test-net-server-bind.js (test: refactor test-net-server-bind #13273)
• test/parallel/test-net-socket-destroy-twice.js(Remove common.PORT usage from parallel tests  #12473)
• test/parallel/test-regress-GH-5051.js
• test/parallel/test-regress-GH-5727.js
• test/parallel/test-tls-client-abort.js (being worked on by @ahmed-taj) (tests: remove common.PORT from client abort #12461)
• test/parallel/test-tls-client-abort2.js (being worked on by @ahmed-taj) (tests: remove common.PORT from client abort #12461)
• test/parallel/test-tls-client-default-ciphers.js
• test/parallel/test-tls-connect.js
• test/parallel/test-tls-ticket-cluster.js

[richardlau]

IMO, all instances of common.PORT in parallel should either be changed to use port 0 (if possible in the context of the test) or else moved to sequential (if using port 0 is not possible).

Maybe update the description of common.PORT in test/README.md with this recommendation and/or (if possible) introduce a linting rule after the list is resolved?

[evanlucas]

For a few of these, (test-net-better-error-messages-*) they don't actually have a server, they try to connect to a server that is not running. Are those ones that you think we should still change @Trott?

[orafaelfragoso]

@evanlucas @richardlau I would love to contribute to this one, but I don't feel comfortable doing so. Can I count on you to guide me on my first contributions? Anything I should read on OS port attribution?

[thefourtheye]

So we are slowly removing common.PORT?

[gibfahn]

I would love to contribute to this one, but I don't feel comfortable doing so. Can I count on you to guide me on my first contributions?

@rafaelfragosom go for it! You should have plenty of people willing to help you if you have any issues (count me in).

[orafaelfragoso]

@gibfahn I'm making progress so far. Thank you!

[Trott]

So we are slowly removing common.PORT?

@thefourtheye From tests in parallel only. It's fine in sequential. And if we find tests in parallel that cannot be re-written to avoid common.PORT, then moving those tests to sequential is probably the way we'll go.

Once we get common.PORT out of parallel, we can probably alter common.js and maybe test.py to simply hardcode common.PORT to a single value. (In other words, there's special logic in there to make sure that parallel tests have different common.PORT values. That won't be necessary anymore.)