Github user tgravescs commented on the pull request:
https://github.com/apache/spark/pull/7943#issuecomment-128126795
So Yarn Resourcemanager, Nodemanager and MapReduce shuffle handler all use
level db. We have been running them for quite while and doing rolling upgrades
without problems so I'm pretty confident in levelDB at this point.
Also on the stopApplications, the MapReduce shuffle handler is relying on
that and we run tens to hundreds of thousands of jobs a day and haven't seen
any issues. I'll take another look at the MR code to make sure I didn't miss
anything special they were doing. That combined with machines going down
frequently and us doing rolling upgrades makes me pretty confident its working
there too.
One of the main things I want to do is make sure whatever we do if
compatible with various versions of Spark. The nodemanagers are only going to
be running one version of the shuffle handler so it has to work across versions
(Spark 1.4, 1.5, 16, etc). At some point we may have to change the format
and we can do this by storing some sort of version and schema of the data store
and checking for compatibility when upgrading. We wouldn't have to do that now
as long as the next version that changes handles it. The MR shuffle handles has
code in there for this now but I don't think its ever changed versions at this
point.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
