The folks at Elastic have written up a really good howto on using Logstash. If you’re not that familiar with Logstash and are considering an ELK stack, add this to your reading list.
The folks at Salesforce engineering are talking about how they approach monitoring of their microservice architecture. THe part I really like is the “contract”: every services must have an SLA (and therefore SLIs), every service must monitor how other services use it, and every service must monitor how it uses other services. There’s some details in here, so check it out.
For those running AWS Aurora, you can send the MySQL logs (but not Postgres 😢) to Cloudwatch now . I’m actually rather surprised this wasn’t already a thing, but hey, better late than never, I guess.
h/t to SRE Weekly for this one, which I missed when it came out back in February. I’ve long argued that for the vast majority of use-cases, the observer effect is mostly theoretical and not a practical concern. For some use-cases, though, it’s kind of a big deal. The folks at Wallaroo Labs are one such use-case and put a lot of thought into instrumenting their data processing pipeline while both maintaining accuracy and limiting the impact of instrumentation on the performance of the pipeline.
If you’ve ever wondered how Prometheus does quantile calculation, here you go: an analysis at how the quantile algorithm is implemented. Postmates Engineering actually has some published code on this as well, and there’s an IEEE paper it all stems from, in case you’re interested.
Speaking of Prometheus, someone created a Chrome plugin to pretty-print the
/metrics endpoint output. Probably very useful for spot-checking and debugging.
The folks at SemaText just wrapped up their series on distributing tracers with a comparison of Jaeger and Zipkin. tldr you probably want Jaeger
The Zabbix Summit is coming up in October, held in Latvia. Both registration and the CFP are still open.
See you next week!
— Mike (@mike_julian)
Monitoring Weekly Editor