There are three file roll defaults on the HDFS sink, rollInterval, rollSize and rollCount. Their defaults are 30s, 1024B and 10 ecents, respectively. You need to set each one as desired or disable them.
My mail client chopped up your config, but I did a search and didn't see any of those properties set. I would start there. Hope that helps, Paul Chavez -----Original Message----- From: Josh Myers [mailto:[email protected]] Sent: Monday, June 17, 2013 6:47 AM To: [email protected] Subject: Flume events rolling file too regularly Hi guys, We are sending JSON events from our pipeline into a flume http source. We have written a custom multiplexer and sink serializer. The events are being routed into the correct channels and consumed OK by the sinks. The custom serializer takes a JSON event and outputs a csv. Files are being written to s3 ( using s3n as hdfs ) but rather than appending the written csv file, each event seems to be generating it own csv. The output is what I would expect using rollCount 1, however we do occasionally get several events ( maybe 4 ) written per csv. Please see below for config. Ideally we want to use rollInterval of 24 hours, to generate a new .csv file every 24 hours, but have events pretty quickly flushed to the csv file after being sent. So one csv' per day that is consistently appended with whatever events we throw in. We found however that with a rollInterval of 24 hours the events weren't being flushed often enough... Any help would be hugely appreciated! Thanks. Josh
