[collectd] updates per second limit using rrdcached?

[Brian O'Neill]

So, we've been using rrdcached for a while with collectd. We have over the past few months started getting more and more drops, up to 3k per second. We are currently collecting approximately 250k statistics every 10 seconds, so, we're doing about 25k/sec. We're writing to the rrdcached engine using unix sockets. Based on the rrdcached operations stats it is only able to receive around 23k/sec. Obviously brining in 25k/sec and writing 23k/sec isn't good, so we're hitting the queue limit sizes.

We've tried upping the limits and changing the number of write threads, but none of it has been able to make a positive impact.

Anyone out there doing more than 25k stats per second with rrdcached? If so, what sort of CPU are you using? I have a feeling there issue is our use of an old X5660 at 2.8Ghz and that rrdcached is single threaded in its reading from the unix socket.

Anyone have any thoughts or suggestions?

Thanks,
Brian

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.verplant.org/pipermail/collectd/attachments/20181029/78d61ec7/attachment.html>