Re: [pmacct-discussion] stamp_inserted and sql_history
Hello Jamie, On Tue, Aug 16, 2005 at 05:07:33PM +1000, Jamie Wilkinson wrote: Actually, I'm still not clear: the FAQ suggests that only sql_history is required, though sql_history_roundoff is advised. Is that true? True. sql_history writes the stamp fields; sql_history_roundoff enables the roundings. sql_history_roundoff alone (ie. supplied without sql_history) is silently ignored. Cheers, Paolo
Re: [pmacct-discussion] stamp_inserted and sql_history
This one time, at band camp, Paolo Lucente wrote: On Sat, Aug 13, 2005 at 02:14:00AM +1000, Jamie Wilkinson wrote: Ok. Does this mean that unless the config options 'sql_history' and 'sql_history_roundoff' exist, then pmacctd will not write time stamps to the database? yes. Thanks for the clarification. I've done so, but I've also added these two config options back to my config file, and I'm seeing a lot of 'We are missing data' errors in the syslog. Such error shouldn't be related in any way with the stamps. It signals that the shared memory segment between the Core Process (which collects packets from the network) and the Plugin (which writes flows into the DB) is full. (*) It's very likely that you have not enabled bufferization. Try adding to your configuration the two following lines (then tune the parameters in order to fit your scenario): === plugin_pipe_size: 8192000 plugin_buffer_size: 4096 === Ok, I already have a plugin_pipe_size, but no buffer_size. Anyway, there were only a couple of 'We are missing data' messages at startup so I suspect they are transient. Thanks.
Re: [pmacct-discussion] stamp_inserted and sql_history
On Sat, Aug 13, 2005 at 02:14:00AM +1000, Jamie Wilkinson wrote: Ok. Does this mean that unless the config options 'sql_history' and 'sql_history_roundoff' exist, then pmacctd will not write time stamps to the database? yes. I've done so, but I've also added these two config options back to my config file, and I'm seeing a lot of 'We are missing data' errors in the syslog. Such error shouldn't be related in any way with the stamps. It signals that the shared memory segment between the Core Process (which collects packets from the network) and the Plugin (which writes flows into the DB) is full. (*) It's very likely that you have not enabled bufferization. Try adding to your configuration the two following lines (then tune the parameters in order to fit your scenario): === plugin_pipe_size: 8192000 plugin_buffer_size: 4096 === CONFIG-KEYS and docs/INTERNALS give further hints and some details about them and how communication between Core Process and Plugins works. Give them a try and let me know how things get evolved. Cheers, Paolo (*) note that since versions 0.8.x, shared memory support is compiled in by default. This was not the case of previous versions that were requiring the './configure --enable-mmap'. This *might* explain why you didn't get such messages before.
