[
https://issues.apache.org/jira/browse/SPARK-16746?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Hongyao Zhao updated SPARK-16746:
---------------------------------
Description:
I wrote a spark streaming program which consume 1000 messages from one topic of
Kafka, did some transformation, and wrote the result back to another topic. But
only found 988 messages in the second topic. I checked log info and confirmed
all messages was received by receivers. But I found a hdfs writing time out
message printed from Class BatchedWriteAheadLog.
I checkout source code and found code like this:
{quote}
/** Add received block. This event will get written to the write ahead log
(if enabled). */
def addBlock(receivedBlockInfo: ReceivedBlockInfo): Boolean = {
try {
val writeResult = writeToLog(BlockAdditionEvent(receivedBlockInfo))
if (writeResult) {
synchronized {
getReceivedBlockQueue(receivedBlockInfo.streamId) +=
receivedBlockInfo
}
logDebug(s"Stream ${receivedBlockInfo.streamId} received " +
s"block ${receivedBlockInfo.blockStoreResult.blockId}")
} else {
logDebug(s"Failed to acknowledge stream ${receivedBlockInfo.streamId}
receiving " +
s"block ${receivedBlockInfo.blockStoreResult.blockId} in the Write
Ahead Log.")
}
writeResult
} catch {
case NonFatal(e) =>
logError(s"Error adding block $receivedBlockInfo", e)
false
}
}
{quote}
It seems that ReceiverTracker tries to write block info to hdfs, but the
write operation time out, this cause writeToLog function return false, and
this code "getReceivedBlockQueue(receivedBlockInfo.streamId) +=
receivedBlockInfo" is skipped. so the block info is lost.
The spark version I use is 1.6.1 and I did not turn on
spark.streaming.receiver.writeAheadLog.enable.
I want to know whether or not this is a designed behaviour.
was:
I wrote a spark streaming program which consume 1000 messages from one topic of
Kafka, did some transformation, and wrote the result back to another topic. But
only found 988 messages in the second topic. I checked log info and confirmed
all messages was received by receivers. But I found a hdfs writing time out
message printed from Class BatchedWriteAheadLog.
I checkout source code and found code like this:
```
/** Add received block. This event will get written to the write ahead log
(if enabled). */
def addBlock(receivedBlockInfo: ReceivedBlockInfo): Boolean = {
try {
val writeResult = writeToLog(BlockAdditionEvent(receivedBlockInfo))
if (writeResult) {
synchronized {
getReceivedBlockQueue(receivedBlockInfo.streamId) +=
receivedBlockInfo
}
logDebug(s"Stream ${receivedBlockInfo.streamId} received " +
s"block ${receivedBlockInfo.blockStoreResult.blockId}")
} else {
logDebug(s"Failed to acknowledge stream ${receivedBlockInfo.streamId}
receiving " +
s"block ${receivedBlockInfo.blockStoreResult.blockId} in the Write
Ahead Log.")
}
writeResult
} catch {
case NonFatal(e) =>
logError(s"Error adding block $receivedBlockInfo", e)
false
}
}
```
It seems that ReceiverTracker tries to write block info to hdfs, but the
write operation time out, this cause writeToLog function return false, and
this code "getReceivedBlockQueue(receivedBlockInfo.streamId) +=
receivedBlockInfo" is skipped. so the block info is lost.
The spark version I use is 1.6.1 and I did not turn on
spark.streaming.receiver.writeAheadLog.enable.
I want to know whether or not this is a designed behaviour.
> Spark streaming lost data when ReceiverTracker writes Blockinfo to hdfs
> timeout
> -------------------------------------------------------------------------------
>
> Key: SPARK-16746
> URL: https://issues.apache.org/jira/browse/SPARK-16746
> Project: Spark
> Issue Type: Bug
> Components: Streaming
> Affects Versions: 1.6.1
> Reporter: Hongyao Zhao
> Priority: Minor
>
> I wrote a spark streaming program which consume 1000 messages from one topic
> of Kafka, did some transformation, and wrote the result back to another
> topic. But only found 988 messages in the second topic. I checked log info
> and confirmed all messages was received by receivers. But I found a hdfs
> writing time out message printed from Class BatchedWriteAheadLog.
>
> I checkout source code and found code like this:
>
> {quote}
> /** Add received block. This event will get written to the write ahead
> log (if enabled). */
> def addBlock(receivedBlockInfo: ReceivedBlockInfo): Boolean = {
> try {
> val writeResult = writeToLog(BlockAdditionEvent(receivedBlockInfo))
> if (writeResult) {
> synchronized {
> getReceivedBlockQueue(receivedBlockInfo.streamId) +=
> receivedBlockInfo
> }
> logDebug(s"Stream ${receivedBlockInfo.streamId} received " +
> s"block ${receivedBlockInfo.blockStoreResult.blockId}")
> } else {
> logDebug(s"Failed to acknowledge stream ${receivedBlockInfo.streamId}
> receiving " +
> s"block ${receivedBlockInfo.blockStoreResult.blockId} in the Write
> Ahead Log.")
> }
> writeResult
> } catch {
> case NonFatal(e) =>
> logError(s"Error adding block $receivedBlockInfo", e)
> false
> }
> }
> {quote}
>
> It seems that ReceiverTracker tries to write block info to hdfs, but the
> write operation time out, this cause writeToLog function return false, and
> this code "getReceivedBlockQueue(receivedBlockInfo.streamId) +=
> receivedBlockInfo" is skipped. so the block info is lost.
> The spark version I use is 1.6.1 and I did not turn on
> spark.streaming.receiver.writeAheadLog.enable.
>
> I want to know whether or not this is a designed behaviour.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org
For additional commands, e-mail: issues-h...@spark.apache.org
