Github user JoshRosen commented on the pull request:
https://github.com/apache/spark/pull/7004#issuecomment-116163279
Hey @JDrit,
Serialization performance is a big interest of mine and I'd be happy to
help with review of this patch. One question first, though: I've noticed that
this adds a new Avro dependency, but it seems to be scoped to compile-only. I
assume that this is because we don't want to introduce a hard dependency on
Avro to Spark itself, since doing so might create dependency conflicts with
user code. However, I'm worried about what happens if we run Spark without
_any_ version of Avro on the classpath: will we get ClassNotFoundExceptions
when KryoSerializer ties to create a GenericAvroSerializer?
If we can't come up with a clean way to handle this dependency issue in
Spark, the next best solution might be to release this Avro serialization code
as a third-package (e.g. via http://spark-packages.org or your own preferred
distribution channel). I think that this might be possible by packaging the
AvroSerializer code into its own JAR, then writing a custom Kryo registrator
and instructing users on how to configure `spark.kryo.registrator` to use it.
To provide a nice experience for end-users, you could even create a custom
"builder" class that users configure then apply to a SparkConf object in order
to set the appropriate settings for Avro / Kryo.
---
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]
