Resource request and limit #640
Conversation
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #640 +/- ##
=======================================
Coverage 94.15% 94.15%
=======================================
Files 80 80
Lines 3252 3252
=======================================
Hits 3062 3062
Misses 190 190 ☔ View full report in Codecov by Sentry. |
There was a problem hiding this comment.
Unless you have a very good reason, such as an app is startup-hungry, we should be pinning requests and limits. It makes things more predictable, simpler to debug, and there's no reason not to if we know we have the resources to spare.
Each of these pins should be done one at a time.
(Addendum, this is all based on my own experience. If you have a Datadog graph that proves me wrong, I'd like to see it.)
| # requests: | ||
| # cpu: 100m | ||
| # memory: 128Mi | ||
| resources: |
There was a problem hiding this comment.
We should definitely pin Postgres to a core. It deserves it.
infra/helm/meshdb/values.yaml
Outdated
| resources: | ||
| limits: | ||
| cpu: 1 | ||
| memory: 512Mi | ||
| requests: | ||
| cpu: 200m | ||
| memory: 350Mi |
There was a problem hiding this comment.
Pin this guy to 512Mi. 200m is probably fine as a request, as I think most of the CPU is on startup with the tracer.
infra/helm/meshdb/values.yaml
Outdated
| resources: | ||
| limits: | ||
| cpu: 1 | ||
| memory: 100Mi | ||
| requests: | ||
| cpu: 250m | ||
| memory: 30Mi |
There was a problem hiding this comment.
This guy definitely doesn't need more than 250mcore and like, 128Mi of RAM. If our web proxy is taking up more than that, we've either got wayyyy more users than we thought we did, or we have a performance problem.
infra/helm/meshdb/values.yaml
Outdated
| resources: | ||
| limits: | ||
| cpu: 1 | ||
| memory: 512Mi | ||
| requests: | ||
| cpu: 50m | ||
| memory: 30Mi |
There was a problem hiding this comment.
Redis really doesn't like not being pinned, especially as an in-memory cache. I'd say pin it to 256Mi and like..... 250mcore. If that doesn't work, bump it.
infra/helm/meshdb/values.yaml
Outdated
| resources: | ||
| limits: | ||
| cpu: 1 | ||
| memory: 512Mi | ||
| requests: | ||
| cpu: 250m | ||
| memory: 250Mi |
There was a problem hiding this comment.
No fucking idea here. Can probably get away with pinning limit to the smaller request here tho.
Set resource requests + limits based on current production usage as seen in el doggo.
Docs: https://kubernetes.io/docs/concepts/configuration/manage-resources-containers/