Open pawellesniewski opened 7 years ago
What's your metric throughput?
Around 1 mln / min, 6x cyanite nodes.
Sorry, just to make sure: it'd be around 16K a second, also they're split between 6 Cyanite nodes?
Do you see the gap also growing over the time?
Yes, 16k per second, 6 cyanite nodes - just for testing purposes.
No gaps over the time.
The load doesn't seem too high. My suspicion (if Cyanite doesn't start lagging behind more and more) is this somehow related to the timer being faster than ingestion.
Well, it looks like i didn't get how exactly Cyanite works. It awaits for incomming metric values for whole datapoint period, calculates max/mean/min/sum and then it flushes it to Cassandra. So it is not possible to read data directly from Cyanite cache (and this is how i always get those nulls i think).
I think I'm starting to understand. Do you have a configuration file at hand? Wondering what rollups you have defined.
Do I understand it correctly: you would like to have a datapoint for current period, even if it's not yet flushed (in-memory).
I'm still uncertain why it returns nulls, usually they should be filtered out though, didn't have a chance to test yet.
Hi there,
my metric relay clones traffic into Graphite and Cyanite. Unfortunately Cyanite is always two datapoints behind Graphite. Is there any possibility to speed this up?
Engine rule:
Graphite results:
Cyanite results:
and after a while:
this testing metric is sent every one minute.