Github user mridulm commented on a diff in the pull request:
https://github.com/apache/spark/pull/887#discussion_r13062446
--- Diff: core/src/main/scala/org/apache/spark/storage/BlockManager.scala
---
@@ -772,7 +772,7 @@ private[spark] class BlockManager(
/**
* Replicate block to another node.
*/
- var cachedPeers: Seq[BlockManagerId] = null
+ @volatile var cachedPeers: Seq[BlockManagerId] = null
--- End diff --
Even though we can create explicit storage levels with different
replication, the stock impl has only two : no and one.
Spark codebase relies on this implicitly.
We had sort of discussed it in the checkpoint to blocks jira (forgot jira
id and can't search right now)
On 27-May-2014 10:39 am, "Aaron Davidson" <[email protected]> wrote:
> In core/src/main/scala/org/apache/spark/storage/BlockManager.scala:
>
> > @@ -772,7 +772,7 @@ private[spark] class BlockManager(
> > /**
> > * Replicate block to another node.
> > */
> > - var cachedPeers: Seq[BlockManagerId] = null
> > + @volatile var cachedPeers: Seq[BlockManagerId] = null
>
> @rxin <https://github.com/rxin> Looking at this code, is this right? When
> we create this (5 lines down), we seem to give it a specific
> level.replication, so we'll only end up with a certain number of peers,
> which we will then *cache*. It seems to me that every block thereafter
> will be forcibly stored at the same replication level as the first.
>
> This bug may be possible, because it would only show up if you're using 2
> separate replication levels above 1, which is probably uncommon. Am I
> missing something, though?
>
> â
> Reply to this email directly or view it on
GitHub<https://github.com/apache/spark/pull/887/files#r13062338>
> .
>
---
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.
---
