Content
@
0 reply
0 recast
0 reaction
Alex Loukissas 🍉
@futureartist
This week we found out during some load tests that our metrics library (cloudwatch) could not keep up with the requests, eventually OOMing the pods. This is an inherent issue with push-based metrics vs pull-based ones (e.g. Prom). With pull-based metrics you do zero outbound API calls, just have the scraper hit your metrics endpoint at a rate that you control. Prometheus ftw!
1 reply
0 recast
3 reactions
welp 🎩🌸
@we
Yeah, pull-based is vastly superior. For cloudflare workers, I don't have a choice though and have to push metrics at the end of each request to a service propped up in our AWS infra, that then gets scraped by datadog agents.
0 reply
0 recast
1 reaction
