Github user JoshRosen commented on the pull request:
https://github.com/apache/spark/pull/9624#issuecomment-156922650
I did some more digging and it looks like this whole locking situation is a
much bigger mess than I originally estimated:
- SBT coordinates access to the Ivy cache using a lock file that is shared
by all running SBT processes:
https://github.com/sbt/sbt/blob/v0.13.8/ivy/src/main/scala/sbt/Ivy.scala#L65.
[By
default](https://github.com/sbt/sbt/blob/v0.13.8/ivy/src/main/scala/sbt/Ivy.scala#L119),
this file is named ".sbt.ivy.lock". This will handle coordination among SBT
processes but does not guard against other processes which might also use the
Ivy cache.
- Ivy's built-in locking support seems hard to configure and also seems
prone to issues related to lock files not being cleaned up when processes
crash: https://issues.apache.org/jira/browse/IVY-1388.
- At http://jira.pentaho.com/browse/BISERVER-4809, someone points out that
although Ivy's artifact cache can be guarded via locks, its resolution cache is
not guardable by any of Ivy's locking strategies.
@brkyvz, given all of this, I wonder whether Spark's default Ivy cache
location should be different than `~/.ivy2`.
To help clear this patch out of the queue, though, I'll just go ahead and
implement the flag as originally discussed (off by default and using shared
cache only in Jenkins, where it happens to be safe because each build workspace
has its own isolated Maven and Ivy cache directories which are preserved across
builds).
---
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]
