GitHub user ssaavedra opened a pull request:
https://github.com/apache/spark/pull/22392
[SPARK-23200] Reset Kubernetes-specific config on Checkpoint restore
Several configuration parameters related to Kubernetes need to be
reset, as they are changed with each invokation of spark-submit and
thus prevents recovery of Spark Streaming tasks.
## What changes were proposed in this pull request?
When using the Kubernetes cluster-manager and spawning a Streaming
workload, it is important to reset many spark.kubernetes.* properties that are
generated by spark-submit but which would get rewritten when restoring a
Checkpoint. This is so, because the spark-submit codepath creates Kubernetes
resources, such as a ConfigMap, a Secret and other variables, which have an
autogenerated name and the previous one will not resolve anymore.
In short, this change enables checkpoint restoration for streaming
workloads, and thus enables Spark Streaming workloads in Kubernetes, which were
not possible to restore from a checkpoint before if the workload went down.
## How was this patch tested?
This patch needs would benefit from testing in different k8s clusters.
This is similar to the YARN related code for resetting a Spark Streaming
workload, but for the Kubernetes scheduler. This PR removes the initcontainers
properties that existed before because they are now removed in master.
For a previous discussion, see the non-rebased work at:
apache-spark-on-k8s#516
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/ssaavedra/spark fix-checkpointing-master
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/spark/pull/22392.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #22392
----
commit f457d023e8e488b89b97f5b3b9936831d7fc9bb6
Author: Santiago Saavedra <santiagosaavedra@...>
Date: 2018-09-11T07:58:40Z
Reset Kubernetes-specific config on Checkpoint restore
Several configuration parameters related to Kubernetes need to be
reset, as they are changed with each invokation of spark-submit and
thus prevents recovery of Spark Streaming tasks.
----
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
