Github user squito commented on a diff in the pull request:
https://github.com/apache/spark/pull/7927#discussion_r36297450
--- Diff: core/src/main/scala/org/apache/spark/storage/BlockManager.scala
---
@@ -590,10 +590,21 @@ private[spark] class BlockManager(
private def doGetRemote(blockId: BlockId, asBlockResult: Boolean):
Option[Any] = {
require(blockId != null, "BlockId is null")
val locations = Random.shuffle(master.getLocations(blockId))
+ var failTimes = 0
for (loc <- locations) {
logDebug(s"Getting remote block $blockId from $loc")
- val data = blockTransferService.fetchBlockSync(
- loc.host, loc.port, loc.executorId,
blockId.toString).nioByteBuffer()
+ val data = try {
+ blockTransferService.fetchBlockSync(
+ loc.host, loc.port, loc.executorId,
blockId.toString).nioByteBuffer()
+ } catch {
+ case e: IOException if failTimes < locations.size - 1 =>
+ // Return null when IOException throw, so we can fetch block
+ // from another location if there still have locations
+ failTimes += 1
+ logWarning(s"Try ${failTimes} times getting remote block
$blockId from $loc failed:", e)
--- End diff --
I wonder if we should save the last exception, and throw it if we don't
find the block in any of the locations. that would be more consistent with the
old behavior (throw an exception), but then again, the api returns an option,
so maybe we should just swallow all exceptions and return `None` as you have
here. Also now I see what Sean meant earlier when he suggested pushing this
down into `fetchBlockSync`. It looks as if this method is expecting
`fetchBlockSync` to return null when it encounters an error, so maybe we should
just change the behavior there after all. @aarondav any thoughts? I'm leaning
towards pushing it into `fetchBlockSync` -- have it swallow the exception and
return null. (this is the only use of `fetchBlockSync`.)
---
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]
