Github user srowen commented on the pull request:
https://github.com/apache/spark/pull/5065#issuecomment-82244546
Yes this needs to be update. There are some small pre-existing problems
with this text though, which should be fixed here too.
MiB is a mebibyte, but the value is actually in MB, or megabytes. I imagine
that a lot of code/text elsewhere maybe talks about MB when it means MiB, but
here this should match Mesos docs which indicate that passing 384 will mean
384MB.
More importantly, I realized that the logic is slightly different here
between Mesos and YARN. In YARN, the overhead is always completely overrideable
by setting `spark.yarn.executor.memoryOverhead`. It merely defaults to the
greater of 384, or 10% of the executor memory. In the Mesos `MemoryUtils`, yes,
it's the max of 10% of executor memory, or `spark.mesos.memoryOverhead` which
defaults to 384.
I think this is an unintentional difference that should be rectified? right
now the caller has no way of overriding the 10% minimum, which doesn't seem
right.
I'm OK to commit this (I can change "MiB") since it would correctly reflect
the logic now, but I'd welcome it if you follow up by reviewing the argument
above and look at changing `MemoryUtils` in a separate issue.
---
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]
