Mesos slaves ignore messages coming from Marathon

[Emilien Kenler]

Hello,

I often encounter the following issue.

All the deployments I make on Marathon are frozen and I can't scale or deploy any app.
All my slaves send the following messages:

Ignoring updating pid for framework 20150403-051742-16777343-5050-29135-0000 because it does not exist

Sometime, I can fix it by restarting the mesos master, but it doesn't work all the time.

It happens after configuration change and restart, mesos/marathon update or without any actions on the cluster.

Does anyone know what happens and how to solve it?

Thank you.

--

	Emilien Kenler Server Engineer | Wizcorp Inc. TECH . GAMING . OPEN-SOURCE WIZARDS
+ 81 (0)3-4550-1448|Website|Twitter|Facebook|LinkedIn

[Dario Rexin]

Hi Emilien,

could you check if there's more than one framework id for Marathon? Go to the "Frameworks" tab in the Mesos UI and check if there are multiple Marathon entries. There is a race in fetching the id from Zookeeper at a fresh installation, so when you start up several Marathon instances simultaneously, they will all start without a framework id and on failover register as a fresh instance. This is fixed in current master and will be released with Marathon 0.8.2.

Cheers,

Dario

--
You received this message because you are subscribed to the Google Groups "marathon-framework" group.
To unsubscribe from this group and stop receiving emails from it, send an email to marathon-framew...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Emilien Kenler]

Hi Dario,

Thank you for the help.
Looks like it was the issue.

I just noticed that frameworkId is null in the UI and at /v2/info, is that normal?

I'm using marathon 0.8.1 from the mesosphere repo on CentOS 7.

I also opened an issue about servicePorts and ports.
https://github.com/mesosphere/marathon/issues/1365

Regards,