In <a96b0c030709030814s3936cd3dt2cfb133316744320 at mail.gmail.com> "Werner - Google" <wxxx333 at gmail.com> writes:
I didn't start do play with the new hobbit beta version, who good improvements about the disk I/O limitations on huge load systems. I have a specific question about what's hobbit behavior when disk cache for rrd is active, and i request the graph for a specific metric. Will i not see the last (around) 30 minutes of info on the graph? Or is there some kind "trap" activated when i request a specific graph, who'll request that the info for that host/service should be written to the RRD before the graph is generated?
The program that generates the graph will request a cache flush before doing the graph. There's no synchronization between the two, but since hobbitgraph.cgi and hobbitd_rrd are running on the same physical server, it should be Ok if hobbitgraph.cgi just waits "a while" - which I've randomly chosen to be 0.3 seconds.
Also, the way the caching works doesn't simply delay all updates for 30 minutes (that would make the server extremely busy while flushing all of the cached data every 30 minutes, and then mostly idle for a long time), it actively tries to distribute the updates over the entire period. So the average amount of data cached when you request a graph will be 15 minutes.
Regards, Henrik