I'm not sure I understand your question but I'll be the first to admit this is not fool proof.
That said here are a couple inherent risks I am taking. Assume FlumeEventA is one of 1000 events in a batch. If FlumeEventA makes it to FlumeAgent1 but the batch fails it is entirely possible when the batch is resent it goes to FlumeAgent2. Now this event is on 2 separate file channels, separate jvms and separate servers. It is possible but extremely unlikely that FlumeEventA is processed at the exact same time in FlumeAgent1 and FlumeAgent2. Both agents pop the event off the channel, pull the UUID off the header and check if it is in HBase. Both do not find it so both write to HDFS and we have a duplicate. Considering the archetecture we believe the odds of this are incredibly small and we are OK with the risk. Since the write to HDFS is in a transaction if it fails I don't do a HBase put of the UUID, the transaction rolls back and we try again. I did a fair amount studying the sink and bucketwriter code at the time to understand what the fail conditions are when writing to HDFS. If I remember right it could fail creating the file, writing to the file, closing the file and renaming the file. We all have or own SLAs to meet. After a pretty thorough review and amount of testing we were comfortable this met our SLA better than a mapreduce job to dedupe 90 billion log lines per day. Joey Echeverria <[email protected]> wrote: What happens if the write to HDFS succeeds before the HBase put? -Joey On Wed, Dec 3, 2014 at 2:35 PM, Mike Keane <[email protected]> wrote: > We effectively mitigated this problem by using the UUID interceptor and > customizing the HDFS Sink to do a check and put of the UUID to HBase. In the > customized sink we check HBase to see if we have seen the UUID before, if we > have it is a duplicate and we log a new duplicate metric with the existing > sink metrics and throw the event away. If we have not seen the UUID before > we write the Event to HDFS and do a put of the UUID to hbase. > > Because of our volume to minimize the number of check/puts to HBase we put > multiple logs in a single FlumeEvent. > > > -Mike > > ________________________________________ > From: Guillermo Ortiz [[email protected]] > Sent: Wednesday, December 03, 2014 4:15 PM > To: [email protected] > Subject: Re: Deal with duplicates in Flume with a crash. > > I didn't know anything about a Hive Sink, I'll check the JIRA about it, > thanks. > The pipeline is Flume-Kafka-SparkStreaming-XXX > > So I guess I should deal in SparkStreaming with it, right? I guess > that it would be easy to do it with an UUID interceptor or is there > another way easier? > > 2014-12-03 22:56 GMT+01:00 Roshan Naik <[email protected]>: >> Using the UUID interceptor at the source closest to data origination.. it >> will help identify duplicate events after they are delivered. >> >> If it satisfies your use case, the upcoming Hive Sink will mitigate the >> problem a little bit (since it uses transactions to write to destination). >> >> -roshan >> >> >> On Wed, Dec 3, 2014 at 8:44 AM, Joey Echeverria <[email protected]> wrote: >>> >>> There's nothing built into Flume to deal with duplicates, it only >>> provides at-least-once delivery semantics. >>> >>> You'll have to handle it in your data processing applications or add >>> an ETL step to deal with duplicates before making data available for >>> other queries. >>> >>> -Joey >>> >>> On Wed, Dec 3, 2014 at 5:46 AM, Guillermo Ortiz <[email protected]> >>> wrote: >>> > Hi, >>> > >>> > I would like to know if there's a easy way to deal with data >>> > duplication when an agent crashs and it resends same data again. >>> > >>> > Is there any mechanism to deal with it in Flume, >>> >>> >>> >>> -- >>> Joey Echeverria >> >> >> >> CONFIDENTIALITY NOTICE >> NOTICE: This message is intended for the use of the individual or entity to >> which it is addressed and may contain information that is confidential, >> privileged and exempt from disclosure under applicable law. If the reader of >> this message is not the intended recipient, you are hereby notified that any >> printing, copying, dissemination, distribution, disclosure or forwarding of >> this communication is strictly prohibited. If you have received this >> communication in error, please contact the sender immediately and delete it >> from your system. Thank You. > > > > > This email and any files included with it may contain privileged, > proprietary and/or confidential information that is for the sole use > of the intended recipient(s). Any disclosure, copying, distribution, > posting, or use of the information contained in or attached to this > email is prohibited unless permitted by the sender. If you have > received this email in error, please immediately notify the sender > via return email, telephone, or fax and destroy this original transmission > and its included files without reading or saving it in any manner. > Thank you. > -- Joey Echeverria This email and any files included with it may contain privileged, proprietary and/or confidential information that is for the sole use of the intended recipient(s). Any disclosure, copying, distribution, posting, or use of the information contained in or attached to this email is prohibited unless permitted by the sender. If you have received this email in error, please immediately notify the sender via return email, telephone, or fax and destroy this original transmission and its included files without reading or saving it in any manner. Thank you.
