Github user squito commented on a diff in the pull request:
https://github.com/apache/spark/pull/6990#discussion_r33398416
--- Diff: core/src/main/scala/org/apache/spark/storage/BlockManager.scala
---
@@ -833,8 +833,10 @@ private[spark] class BlockManager(
logDebug("Put block %s locally took %s".format(blockId,
Utils.getUsedTimeMs(startTimeMs)))
// Either we're storing bytes and we asynchronously started
replication, or we're storing
- // values and need to serialize and replicate them now:
- if (putLevel.replication > 1) {
+ // values and need to serialize and replicate them now.
+ // Should not replicate the block if its StorageLevel is
StorageLevel.NONE or
+ // putting it to local is failed.
+ if (!putBlockInfo.isFailed && putLevel.replication > 1) {
--- End diff --
oops, I thought the purpose of this patch was something else earlier.
Sorry to make you explain this again, but I am not certain I understand
what is wrong w/ the current behavior. If we fail to put locally, with the
current behavior, (1) the block will get replicated anyway (2) the remote peer
will cache the block in the originally requested storage level if it can (eg.,
it will store the block in memory) since [the replicate request uses the
original
`putLevel`](https://github.com/apache/spark/blob/a56516fc9280724db8fdef8e7d109ed7e28e427d/core/src/main/scala/org/apache/spark/storage/BlockManager.scala#L853)
and (3) the remote node will tell `BlockManagerMaster` that it has the block,
b/c
[`putBlockData`](https://github.com/apache/spark/blob/a56516fc9280724db8fdef8e7d109ed7e28e427d/core/src/main/scala/org/apache/spark/storage/BlockManager.scala#L316)
just calls
[`putBytes`](https://github.com/apache/spark/blob/a56516fc9280724db8fdef8e7d109ed7e28e427d/core/src/main/scala/org/apache/spark/storage/BlockManager.scala#L676)
which defaults to `tellMast
er = true`.
Is my explanation of the current behavior incorrect, or is there something
wrong w/ the current behavior that I'm not seeing?
---
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]
