bug: tests.lib.log: race condition

[katcipis]

bug: tests.lib.log: race condition

[katcipis]

The problem is here:

	var errs []error
	for {
		// buffered channel avoid goroutine leak
		result := make(chan error, 1)

		go func() {
			l.Printf("retrier: executing work: %s", name)
			result <- work()
		}()

		select {
		case res := <-result:
			{
				if res == nil {
					return nil
				}
				l.Printf("%s: got error: %s", name, res)
				errs = append(errs, res)
			}
		case <-ctx.Done():
			{
				l.Printf("retrier: %s: timeouted, returning all errors", name)
				return append(
					errs,
					errors.New("retrier timeout"),
				)
			}
		}
		// Sometimes we geet an error like: Number of write requests
		// for subscription 'x'
		// exceeded the limit of '1200' for time interval '01:00:00'.
		// Please try again after '5' minutes
		// Before trying again lets backoff a little.
		time.Sleep(backoff)
	}

To prevent Close being called at the same time as a Write I would have to wait all pending work jobs that are running before cancelling the test (and eventually calling tearDown).

The problem is that I do this on purpose exactly to guarantee that if some work goes haywire I guarantee that our teardown function WILL be executed, guaranteeing that resources are ALWAYS cleanedup, even when something locks up forever.

Now I can't think on a better solution then implementing a concurrency safe logger that prevents Write/Read being called concurrently with Close.

There is a discussion about this, and solving this on the os.File don't make sense, but on our particular case it makes sense:

golang/go#10001

[katcipis]

After talking with @tiago4orion it seems like a good idea to just leak the fd and not close it at all :-)