jobmanager rpc inside kubernetes

[Chris Latko <ch...@latko.org>]

when trying to submit a job to my k8s cluster using `flink run -m
localhost:6123 app.jar` i get the following error in jobmanager log:

-----
dropping message [class akka.actor.ActorSelectionMessage] for
non-local recipient [Actor[akka.tcp://flink@localhost:6123/]] arriving
at [akka.tcp://flink@localhost:6123] inbound addresses are
[akka.tcp://flink@sis-flink-cluster-jobmanager.platform.svc.k8s.local:6123]
-----

there are several workarounds but i'm absolutely not going to bundle
my jar into a taskmanager pod just so i can deploy. i would rather
have the jar on ceph and deploy from tm or jm:
- kubectl exec -ti pod-replica-version bash
- flink run blah blah blah...

i could also run akka in unsafe mode, what are the majority of people
doing regarding this issue?

i have an up to date helm chart that is nearing production readiness
that i'm willing to share if people are interested (and my company
lets me).

Re: jobmanager rpc inside kubernetes

[Sampath Bhat <sa...@gmail.com>]

It would be helpful if you provide the complete CLI logs. Because even I'm
using flink run command to submit jobs to flink jobmanager running on K8s
and its working fine. For remote execution using flink CLI you should
provide flink-conf.yaml file which contains job manager address, port and
SSL/HA related config if the same is enabled in your flink cluster to which
the job had to be submitted.

On Fri, Apr 27, 2018 at 7:24 AM, Chris Latko <ch...@latko.org> wrote:

> when trying to submit a job to my k8s cluster using `flink run -m
> localhost:6123 app.jar` i get the following error in jobmanager log:
>
> -----
> dropping message [class akka.actor.ActorSelectionMessage] for
> non-local recipient [Actor[akka.tcp://flink@localhost:6123/]] arriving
> at [akka.tcp://flink@localhost:6123] inbound addresses are
> [akka.tcp://flink@sis-flink-cluster-jobmanager.platform.
> svc.k8s.local:6123]
> -----
>
> there are several workarounds but i'm absolutely not going to bundle
> my jar into a taskmanager pod just so i can deploy. i would rather
> have the jar on ceph and deploy from tm or jm:
> - kubectl exec -ti pod-replica-version bash
> - flink run blah blah blah...
>
> i could also run akka in unsafe mode, what are the majority of people
> doing regarding this issue?
>
> i have an up to date helm chart that is nearing production readiness
> that i'm willing to share if people are interested (and my company
> lets me).
>