> 17 авг. 2015 г., в 9:28, Adam Bordelon <a...@mesosphere.io> написал(а): > > ++ dev@myriad mailing list > > Evgeniy, > > Thanks for trying out Myriad. Sorry for my delayed response. Just got back > from vacation. > Have you "flexed up" to add any NodeManagers to your YARN cluster? > See > https://github.com/mesos/myriad/blob/phase1/docs/API.md#put-apiclusterflexup > <https://github.com/mesos/myriad/blob/phase1/docs/API.md#put-apiclusterflexup> > Yes, that was the reason task was stuck in SCHEDULED state. Now I was able to run wordcount example successfully. But to make it complete, I had to flexup with 1 ‘large’ instance. I think its a bit overkill for such simple task with 2 map jobs. As I understand, YARN operates in terms of vCores, and each Job consumes at least 1 vCore + 1 for AM. When I run pure YARN, I can declare NodeManager has vCores = 2xpCores, so I can allow more parallel jobs which are not CPU hungry. With Myriad it looks like 1 pCore from Mesos == 1 YARN vCore and I have to allocate so many CPUs from Mesos, is it so? Is there a way to set how many YARN vCores maps to 1 Mesos CPU?
Thanks for your help! Eugene P.S. Can I subscribe to dev@myriad list too? Couldn’t figure it out from Myriad Incubator page..
smime.p7s
Description: S/MIME cryptographic signature