Increasing number of retries and/or retry.backoff.ms will help reduce the data loss. Figure out how long NLFPE occurs (this happens only as long as metadata is obsolete), and configure below props accordingly.
message.send.max.retries=3 (default) retry.backoff.ms=100 (default) > On Aug 12, 2016, at 1:59 AM, sunil kalva <kalva.ka...@gmail.com> wrote: > > We are seeing data loss, whenever we see "NotLeaderForPartitionException" > exception, > We are using 0.8.2 java client publisher API with callback, when i get > callback with the error i am logging them in a file and retrying them > later. > So number of errors = number of logged events are matching but overal,l few > events are missing. > Observations > Timestamp of missing events looks like just before the timestamp of the > error occurrence, meaning every time when we get this error we are loosing > couple events which are sent just before the error time. > > Is there any way to handle this, should it be handle at client side or some > tuning required at kafka side. > > t > SunilKalva ____________________________________________________________ Fit Mom Daily Australians Shocked: Don't Use Botox, Do This Instead http://thirdpartyoffers.netzero.net/TGL3231/57ae0e55f24a2e555d79st01vuc
