Re-enable MQTT and Postgres tests

[itowlson]

Both of these are currently ignored out for being flaky.

We should figure out why they are flaky. And then we should fix it and re-enable them.

[PhantomInTheWire]

I tried to reproduce this, and they no longer appear to be flaky. any way to reproduce this?

cc: @kate-goldenring

[PhantomInTheWire]

Following up with more context on my attempt to reproduce this:

• OS: macOS 15.5(Apple Silicon, arm64)

• Steps Taken:

  • I enabled the previously ignored tests by modifying tests/runtime.rs as follows:

  @@ -10,8 +10,8 @@ mod runtime_tests {
       ignore: [
           // This test is flaky. Often gets "Connection reset by peer" errors.
           // https://github.com/spinframework/spin/issues/2265
  -        "outbound-postgres",
  -        "outbound-postgres-variable-permission"
  +        // "outbound-postgres",
  +        // "outbound-postgres-variable-permission"
       ]
   );
  
  @@ -28,7 +28,7 @@ mod runtime_tests {
  
       #[test]
       fn conformance_tests() -> anyhow::Result<()> {
  -        let config = conformance_tests::Config::new("canary").ignore("outbound-mqtt");
  +        let config = conformance_tests::Config::new("canary");
           let conclusion = conformance_tests::run_tests(config, move |test| {
               conformance::run_test(test, &spin_binary())
           })?;

• Command Run: cargo test runtime

• Outcome: MQTT and postgres tests passed without any failures or flaky behavior on my setup.

Let me know if there are additional environment variables, external dependencies, or setup steps I should be using to trigger the failure.

[itowlson]

@PhantomInTheWire Thanks for looking into this! My memory of this issue is that the tests were generally reliable locally, but were flaky enough in GitHub Actions to be annoying (although I can't remember exactly how flaky).

So I don't think there's anything more you can test locally - probably the next step is to send a PR with the tests turned back on, re-run the PR actions a few times, and see if we see errors. Not a very lovely way to investigate the issue, I know...!

[PhantomInTheWire]

Thanks, @itowlson — that makes sense! I’ved opened a PR with the tests re-enabled and monitor the CI runs for any flakiness.