graphite-dev team mailing list archive

[bhardy <question215146@xxxxxxxxxxxxxxxxxxxxx>]

New question #215146 on Graphite:
https://answers.launchpad.net/graphite/+question/215146

I have a few questions on how to prevent the carbon-relay queue from filling up.  To simplify things, lets say I have a single carbon-relay running on one server that sends data to multiple carbon-caches on other servers.

CarbonClientFactory(10.236.83.34:2004:a) send queue is full (5000000 datapoints)

If one of the downstream caches is not accepting data, does the relay stop sending to all of them?  This looks to be what is occurring especially during the "warm up" of brand new metrics.  

Are my only options to choose between True/False on the USE_FLOW_CONTROL option on carbon-relay?   Specific sockets stop until send queue decreases or drop metrics when queue is full.

Is the cache simply filled with new metrics?  If so, should I simply have an abnormally high creates setting until this is done?  Carbon-relay does does not seem to like a very large max queue size?(I recently increased it to 20000000 and have not seen the "full" error.)  The downstream carbon-cache looks to iterate between writing all new metrics or writing all the updates.  During this period, the healthy carbon-caches, which are not getting new metrics, function normally until carbon-relay stops sending.

Maybe I'm mistaken but what is mentioned in the link below looks to still occur.  I am using the last stable 0.9.10 release from 5/31/2012.  Any ideas would help.  I imagine I can wait out the creation of new metrics but I have the same concern as Drew in the link.  Will all of this happen again when I take a carbon-cache down for maintenance???

https://answers.launchpad.net/graphite/+question/186672

-- 
You received this question notification because you are a member of
graphite-dev, which is an answer contact for Graphite.