To followup on my earlier post, given this configuration: exec config src3143 'syslogUdp(3143)' 'agentSink("<host>", 35895)' exec config thecollector 'collectorSource(35895)' 'collectorSink("file:///logs/", "3143files")'
It seems like 'src3143' will try and connect to 'thecollector' only the first time and give up if that try fails? Is using the 'lazyOpen' decorator on src3143 like so required to have src3143 keep on retrying instead? exec config src3143 'syslogUdp(3143)' 'lazyOpen agentSink("<host>", 35895)' Thanks, On Thu, Jan 26, 2012 at 12:53 PM, Henry Larson <newtofl...@gmail.com> wrote: > The current config I use is as follows: > > exec config src3143 'syslogUdp(3143)' 'agentSink("<host>", 35895)' > > i. How do I view what 'agentSink()' gets expanded to? > ii. How do I pass insistentOpen as a decorator to agentSink? > > This is an attempt to resolve issues I am having right now where the agent > does not connect to the collector and has to be restarted to get that going. > > This is how I setup my test environment: > > 1. Bring up the flume collector first (null sink and source) : 'flume node > -n thecollector' > 2. Bring up the flume agent next (null sink and source) : 'flume node -n > src3143' > 3. Send this configuration to flume master through shell: > > exec config src3143 'syslogUdp(3143)' 'agentSink("<host>", 35895)' > exec config thecollector 'collectorSource(35895)' > 'collectorSink("file:///logs/", "3143files")' > > 4. Most of the time, the agent keeps in spooling files in the 'writing' > directory, unless I restart that agent, after which it connects to the > collector and sends over the data > > It seems to me that carrying out the above steps in the following order > decreases the chances of the agent failing to connect to the collector: 3 > first, followed by 1, then 2 > > However, my assumption is that the order of execution of steps should not > matter? > > Suggestions to help mitigate this issue would be welcome. >