Overview of grafana and prometheus related issues

[consideRatio]

There are several tickets relates to grafana and prometheus that seems relevant to overview, so this is a meta-issue overviewing those.

Grafana

• grafana, utoronto: deployer script unauthorized #2152
• pangeo-hubs, prometheus: server is crashing due to memory limits #2215
• grafana: ensure function of Node CPU and Memory utilization % #2191
• grafana: ensure functionality of OOMKiller dashboard #2213
• grafana: tighten grafana charts k8s RBAC permissions (ClusterRole -> Role) #2182
• grafana: learn if github auth supports either org or team authorization #2179
• Community facing information about grafana monitoring - access and monitoring guidance features#25
• Monitoring EKS/GKE spot instance pre-emption events #2369
• Monitor NFS servers - critical diagnostics to understand issues #2242
• Monitor node performance - network read/write speeds, ephemeral storage read/write speeds and capacity #2243
• Decide on a path for enabling GitHub auth to grafana for Community Reps of hubs on dedicated clusters #1850
• Consider exposing grafana via https://<hub domain>/services/grafana #535
• Create a few Grafana dashboards that we can use for reporting #1282
• Collect prometheus metrics from each notebook pod #53
• Tooling for engineers/admins/users to understand memory use #2107
• Example on requesting users authorize the github application for github auth in grafana: [Request deployment] New Hub: Smithsonian #2323 (comment), and another one from the jupyterhub github auth docs in https://infrastructure.2i2c.org/hub-deployment-guide/configure-auth/github-orgs/#granting-access-to-the-oauth-app

jupyterhub/grafana-dashboards

• Is it in scope for this project to document use for non-admin users? jupyterhub/grafana-dashboards#66
• Error when using last 7 days - RangeError: Invalid array length jupyterhub/grafana-dashboards#62
• Dashboard panel for node's available ephemeral storage space jupyterhub/grafana-dashboards#64
• Dashboard panel for pod evictions (out of memory, out of ephemeral space, manual node drains) jupyterhub/grafana-dashboards#65

Prometheus

• Ideas to upper bound prometheus-server's memory consumption #2222
• basehub: reduce prometheus-server's configured disk size (currently 100G) #2223
• Document how to programmatically access Grafana / Prometheus data #1785
• Tighten access to prometheus servers #1101
• 2i2c datasource showing up twice in central grafana #1510

[pnasrat]

Just adding a few useful references on monitoring

• https://www.brendangregg.com/usemethod.html
• https://sre.google/sre-book/monitoring-distributed-systems/
• https://www.weave.works/blog/the-red-method-key-metrics-for-microservices-architecture/

It might be useful to sketch out the system and think through key metrics for end to end debugging

[yuvipanda]

I tried to collect RED metrics from all user notebooks, and found that it is a bit too much - tracked in the links in berkeley-dsep-infra/datahub#1993. Prometheus crashes quickly.

However, I think we already do collect and store RED metrics from the hub, and nginx-ingress. We should expose these.

[colliand]

I learned about RED metrics today.

Rate, Error, Duration

[consideRatio]

I think this may no longer relevant enough to keep open, most issues are closed.