kubernetes
/
community
mirror of https://github.com/kubernetes/community.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity
community/sig-instrumentation/archive/meeting-notes-2018.md

18 KiB
Raw Blame History

Agenda (2018-12-13)

  • Metrics overhaul KEP discussion - in person in Seattle at KubeCon
    • Discussed what needs to be done, priority and what is already in-flight
    • Decided to keep any non-conformant metric labels for v1.14 but clearly state they are deprecated and will be removed in v1.15 (or v1.16 if we get any pushback)
    • Add histograms wherever there are summaries
    • Make summary metrics opt-in with a kubelet flag
      • Not a breaking change, can be done after v1.14 target
    • Update KEP status to implementable
      • Thanks @ehashman
    • Create plan to add dev, operator and user docs to metrics
      • I don’t remember all of the context on this, @directmanx12 this was something you brought up, can you fill it in a bit?
    • Discussed how to change a single global metrics registry to something that gets passed in and can be replaced with a no-op registry if desired
      • This pattern has been implemented in client-go as part of the controller runtime implementation with the logger object

Agenda (2018-11-29)

  • Demo on tracing Sam Naser

Agenda 2018-11-15

Agenda 2018-11-1

  • Elasticsearch logging addon - @coffeepac
    • Additional OWNER
    • New image repo
  • Metrics overhaul KEP opened and targeted for 1.14

Agenda 2018-10-18

Agenda 2018-10-04

  • Canceled due to having no agenda points to discuss.

Agenda 2018-09-06

  • Charter merged
  • We need to write a KEP (Kubernetes Enhancement Proposal) for metrics overhaul, because it affects lots of users
  • SIG Instrumentation has to use the Kubernetes organizations for now
  • Kube-state-metrics performance optimization

Agenda 2018-08-23:

Agenda 2018-07-26:

Agenda 2018-06-28:

2018-06-14:

2018-05-31:

2018-06-14

  • How to submit test results as a third party
    • Prefer to find sig-testing doc, will try and prepare a minimal sig-inst doc if needed

2018-05-31

  • Charter PR or doc should be coming tomorrow (6/1)
    • Charter defaults align with what we already do

2018-05-17

  • KubeCon recap
    • Medium well attended and lots of good questions
    • Very good audience
    • Lengthen one session to include a compressed intro and the entire deep dive and not one shorter topic on each
    • Energetic custom metric adapter interest from vendors (at least 3 new)
    • Public link for videos forthcoming
  • Heapster is now deprecated
    • Thanks @directxman12
    • This is official, feature requests closed
    • Make sure this makes it to the v1.11 release notes
    • What are the next steps to graduate kube-state-metrics out of alpha
    • Action item: @piosz to find current dashboard maintainers and determine what the current state of the dashboard is,
      • Historical API, does dashboard want to access data directly
  • Sig-instrumentation-kubernetes group
    • What is the policy for allowing projects
    • Need a charter
      • Includes official processes for a sig, structure of sig, etc.
      • @brancz to fill out template prior to next meeting @coffeepac to add template to this
        • README for what the process is, includes link to template
  • 3rd party/vendor test comments
    • What should be marked as ‘e2e’
      • @coffeepac to generate list of e2e tests we own, if a reasonable number share a spreadsheet to #sig-instrumentation slack
    • How to label 3rd party/vendor tests for viewing
      • @coffeepac to write up how to do this

2018-04-19

  • “Ignoring flakes: sig-instrumentation” https://groups.google.com/forum/#!topic/kubernetes-sig-instrumentation/cbbzkMXSMaw
    • If it is not kube code, then we should not have tests on them - Solly
    • Given we have one kind of e2e tests we are not fixing in time, we shouldn’t add more (Regarding last meetings discussion) - Frederic
    • What is the Kubernetes code being tested here (it looks like “can Stackdriver scrape Kube logs”)? If it’s “can thing X connect to Kubernetes”, then it probably shouldn’t be in Kubernetes e2e tests - Solly
      • Can we have a way for external projects to test integrations with Kube? Might want to reach out to SIG testing - Frederic
    • @coffeepac to ask sig-instrumentation about what is the desired way to handle 3rd party/vendor integrations for e2e testing
  • Prometheus cluster-monitoring addon https://github.com/kubernetes/kubernetes/pull/62195#issuecomment-382778622
    • Addons should not belong in the Kubernetes repository - Frederic/Solly
    • Cluster-monitoring seems like a lot larger scope than discussed e2e setup from last meeting - Frederic
    • Should have gone into a sig-instrumentation specific repo - @coffeepac
    • Contrib repo recommends Prometheus Operator - Frederic
  • Kubernetes Node Monitoring - Solly
  • Kube-pod-exporter POC demo

2018-04-05

  • [piosz] kube-up is in a bit of shaky position
    • Deprecate InfluxDB kube-up in 1.11, remove in 1.12
    • [sross] deprecate Influx e2e tests as well
    • [piosz] deploy Prometheus as well
      • [sross] it’s not needed for e2e tests, so I’d lean against
      • [piosz] want a “real” test for custom metrics, with an actual monitoring solution, Prometheus would be good for that, non-blocking
      • [sross] just need to be careful to avoid maintenance issues with Influx in the future
  • [brancz] have PoC for pod exporter, blocked on getting crio up with supports for stats endpoint, share it hopefully next meeting

2018-03-22

  • Aligning cAdvisor labels with official Kubernetes instrumentation guidelines (possibly related to https://github.com/kubernetes/kubernetes/issues/45043)
    • TODO(brancz): Share POC of pod-exporter once CRI implementation with stats endpoints is available
    • Further: brancz and directxman12 will take lead on stable metrics for pods in Kubernetes
      • Need to figure out pod-level cgroups, other data endpoints (device metrics, etc)
  • Road to heapster deprecation/phase out? Should we put a deprecation note at the top of the heapster readme?
    • Mark Heapster as being in maintenance mode
      • No new features
      • No new sinks
      • Only bugfixes
    • Come up with timeline for deprecation
      • No support
      • No new bugfixes
    • Need better docs on metrics-server setup
    • Docs missing?
  • Metrics Server Cleanup
    • Backport fixes from Heapster (IPV6, etc)
    • Remove unneeded code
    • Abstract out serving interface to serve resource metrics API from other sources (e.g. directly from monitoring pipeline), implement testing tools, etc
    • [directxman12] to publish a bunch of the refactor code
  • Proxying counter metrics in Prometheus client
    • Pain point of prometheus client library when writing exporters, where counter semantics cannot necessarily applied with available abstractions by the golang Prometheus library
      • Interim solution: Implement necessary semantics with “lower level” Prometheus “const” metrics
      • Long term: Learn from the interim solution in order to provide re-usable abstraction to Prometheus client-library

2018-02-22

2018-02-08

2018-01-25

  • Intro and Deep Dive Sessions in Copenhagen
  • The road to heapster deprecation?
  • State of metrics-server
  • Prometheus-k8s-adapter

Notes:

  • brancz@ is interested in making Intro for KubeCon (and DeepDive as well). Piotr can also prepare something for Intro.
  • Heapster deprecation:
    • kubectl top switched to metric-server in 1.10.
    • Google is need heapster for exporting metrics to Stackdriver. Their team is going to support it.
    • We can remove Metrics API from the Heapster. Dashboard may still rely on Model API of heapster.
  • Metric-server:
    • We don’t want to keep sinks in the codebase
    • Need well defined interface between metric-server and kubelet. Summary API is not ideal right now.
    • It’s not clear if PVC should be represented as separate entity or as a part of Pod stats.

2018-01-11

  • 2018 Vision
    • Move all sig-instrumentation projects to new home (cluster addons, contrib, standlone apps, etc) - @coffeepac to start planning
    • Make build/release of projects be publically viewable/triggerable
      • Find out where kubernetes/kubernetes is and start moving sig-inst work to mainline process - @coffeepac to find starting issue
    • Historical metrics API - @brancz follow up on VPA design doc to find out involvement needed from sig-instrumentation
    • Kubernetes Pod exporter - @brancz share prototype and figure out what the plan of CRI stats is going to be going forward
  • kube-state-metrics release