Intermittent failures in io.vertx.ext.dropwizard.MetricsTest.testHttpMetricsResponseCode about vertx-dropwizard-metrics

[tsegismont]

It appears this is due to:

    // A little clunky, but it's possible we got here after closed has been called
    if (closed) {
      removeAll();
    }

in io.vertx.ext.dropwizard.impl.TCPMetricsImpl#disconnected

When this is commented out, the test passes.

The reason the test fails now is that it starts/closes an http client/server pair several times in a row, using the same host and port. Sometimes, after the client (or the server) has been closed, the disconnected (channel inactive) event is fired, and then all metrics for the host:port are removed. But they have been created again in the meantime, for a new server which is not closed yet...

I don't understand yet why we need to remove all the metrics if closed has been called. Because if closed has been called, the metrics have been removed already. Does someone have any idea?

[cescoffier]

The removeAll is required as some metrics are not removed beforehand. Unfortunately I'm not able to reproduce it locally. Any hint ?

[tsegismont]

IIRC I found this bug while checking CI.

To reproduce it I believe I have marked the method with the Repeat annotation

[cescoffier]

I used the repeat annotation, 100 times and still no bug... Definitely related to speed. I need to check if we are not reusing the same metrics instance.

[tsegismont]

@cescoffier can you have a look? Thx

[vietj]

LGTM