mesos-go additional testing for zk detection

Was thinking about the implementation some more, and am concerned that the unit tests are still not adequate. for example:

what if the client's embedded connection flaps between connected/disconnected - what does that look like to the detector (different from #99, which tests the whole connector flapping)?
what about other failure modes of the embedded connection, should client.watchChildren react more intelligently?

Feb 27 '15 04:02 jdef

@jdef with your integration of this with Kubernetes, do you still feel uncomfortable, should we revisit ?

Mar 23 '15 21:03 vladimirvivien

christos had mentioned that he experienced some problems with the detector when ZK was killed and started. I tried with a single-master system (on GCE) and the mesos master died too - which I thought was strange. I haven't tested on a multi-ZK cluster yet but I wanted to start a framework and then take down all the ZK nodes, then bring them back up and see what happened. And from there, perhaps derive some additional automated tests.

Mar 23 '15 21:03 jdef

additional testing:

https://github.com/mesos/mesos-go/commit/049b57bd9bd0550056f26b5ced76f36a820243eb

May 03 '15 16:05 jdef

@jdef: Is this covered in #162?

Sep 10 '15 12:09 tsenart

No, but #162 may be seen as an enabler because it's MUCH thinner than the fat client shim that was there before and the interface is cleaner.

On Thu, Sep 10, 2015 at 8:33 AM, Tomás Senart [email protected] wrote:

@jdef https://github.com/jdef: Is this covered in #162 https://github.com/mesosphere/mesos-go/pull/162?

— Reply to this email directly or view it on GitHub https://github.com/mesosphere/mesos-go/issues/100#issuecomment-139221787 .

Sep 10 '15 13:09 jdef

This is another candidate issue to be moved to an independent github.com/mesosphere/detector

Sep 10 '15 14:09 tsenart