Closed florianschmidt1994 closed 6 years ago
Update: Currently it looks like after cleaning my setup and rebuilding the jar, the local benchmark works only from the second time on that it is run, but consistently afterwards.
Hi! Can you please confirm, that you do not have a local installation of the HOBBIT platform? SDK conficts with it when it is running. If no platform is running, then the problem can be because of rabbitMQ may contain old queues from the previous run (by default it is not stopped/killed by the SDK after execution). To check that you may check whether no rabbit container is running when you start the checkHealth(). If it will help, then you may set .useCachedContainer(false) for the rabbitMqDockerizer in test file to force its killing before every run.
P.s.: under which OS and Java version you are running?
I built and installed a local version of the java sdk with mvn clean install
, is that not the way to do it?
When running it with useCachedImages = false
it never succeeds, for some reason it seems like something is not initialized correctly during the first run.
I'm runnning MacOS 10.13.3 with jdk1.8.0_131
Seems, I reproduced the problem. Trying to find a solution.
I have updated the SDK to 1.1.4 (packaged jar is under lib folder, you may install in by doing the mvn validate
).
I also updated images for benchmark-controller and eval-storage. Delete please the old ones on your machine by docker rmi <imageName>
, the new-one will be pulled by the SDK automatically.
I found the problem of not finishing execution only if multiple benchmark-controller containers were executed at the moment (run two executions in parallel, or old containers were not killed then I interrupted execution manually). This is not directly your problem, but probably updates solution should help. If not, then provide please full logs again, now they should contain some additional information for debugging.
Recently I have fixed some other bug with TaskGenerator. Please try the updated code from the repository.
It now seems to be working, thanks!
Hey everyone,
I have a problem where the checkHealth test sometimes does not seem to execute the SystemAdapter code at all, but instead gets stuck in a loop of
A already put together a script to clean up all remnants of previous runs, but so far I was not able to see in which cases this works / in which cases it doesn't.
My cleanup steps
Please note that sometimes it does work (meaning I also see the accuracy score, logs, ...), it's just that I can't figure out what is necessary to have it work every time. I attached the full logs of that run below. Any help would be very much appreciated