----------------------------------------------------------- This is an automatically generated e-mail. To reply, visit: https://reviews.apache.org/r/5880/#review9025 -----------------------------------------------------------
Juhani, Thanks for the patch. As such I am ok with this patch. Considering it is an exec source, it is a given that it is a fire and forget source. The only issue I see with this patch, is that it is possible that if Flume fails/node crashes, uto bufferCount lines of output may not be persisted to the channel, though it is a simple extension of the fact that 1 event may not have been persisted in the current code as well. So maybe we should mention that in the java docs. Also we should mention the default. - Hari Shreedharan On July 10, 2012, 9:56 a.m., Juhani Connolly wrote: > > ----------------------------------------------------------- > This is an automatically generated e-mail. To reply, visit: > https://reviews.apache.org/r/5880/ > ----------------------------------------------------------- > > (Updated July 10, 2012, 9:56 a.m.) > > > Review request for Flume. > > > Description > ------- > > Added a new configuration setting "bufferCount" to allow for multiple lines > read from the exec output to be committed in a single channel transaction > > > This addresses bug FLUME-1361. > https://issues.apache.org/jira/browse/FLUME-1361 > > > Diffs > ----- > > flume-ng-core/src/main/java/org/apache/flume/source/ExecSource.java dbf79e0 > > flume-ng-core/src/main/java/org/apache/flume/source/ExecSourceConfigurationConstants.java > 73c985e > > Diff: https://reviews.apache.org/r/5880/diff/ > > > Testing > ------- > > Unit tests pass though new ones were not added. > > Further, tested this on our log generating servers and it alleviated the > incredibly poor performance of file channel caused by the constant commits > coming from ExecSource > > > Thanks, > > Juhani Connolly > >
