Brad Lhotsky
Brad Lhotsky
I read "Awesome Split Keyboards" as a review. The Matias Ergo Pro is awesome in theory. In practice, it's a terrible keyboard because of the reliability issues. I do not...
Seems a rather large omission to not allow the user to edit a new or existing MR description in their editor.
I think I'm hitting this bug as well. In 0.12.3, I was able to handle a massive query load without throwing any 500s, 100% of my internal monitoring system's query...
I tweaked concurrency and other settings. It still appears to be having issues, namely making too many requests to carbonserver: ``` 2021-10-28T19:13:20.837Z WARN zipper errors occurred while getting results {"type":...
Well, carbonserver is in fact listening.. a connection refused can also happen when the listen backlog is overflowing.. which appears to be the case. It seems like something changed and...
### TCP statistics If I attempt 100 simultaneous clients to carbonapi, I will wind up over 1,000 new connection opens per minute per carbonserver backend.. These appear to be shortlived...
More data, looking at zipper render requests, it's at **4 million** for the same duration: * zipper requests => groupByNodes(nonNegativeDerivative(carbon.api.graphite*.zipper.render_requests), 'max', 3)
OK, I installed Booking's carbonapi/carbonzipper on a host in the cluster.. I ran my brutal query test against Booking's fork and 0.15.4: 0.15.4 fell over at 100 concurrent clients, and...
Confirmed that switching to Booking's fork allows me to cut-over our monitoring stack to using carbonapi. If this gets addressed, I can retry this version and see if it scales...
FWIW, this is still a thing. I setup a script on a storage node to watch the "ActiveOpens / PassiveOpens" as reported by `/proc/net/snmp`: I set: ```yaml concurrency: 2048 maxBatchSize:...