Test-infra: master blocking jobs should probably be on etcd3

Created on 6 Apr 2018 · 23Comments · Source: kubernetes/test-infra

$ cat jobs/env/ci-kubernetes-e2e-gci-gce.env
### job-env
# This list should match the list in kubernetes-pull-build-test-e2e-gce.
NUM_NODES=4
# For now explicitly test etcd v2 mode in this suite.
STORAGE_BACKEND=etcd2
TEST_ETCD_IMAGE=2.2.1
TEST_ETCD_VERSION=2.2.1

# Enable the PodSecurityPolicy in tests.
ENABLE_POD_SECURITY_POLICY=true

I'm not sure who decides this or exactly all which jobs should be running what, but other jobs appear to consume this env and my understanding is that kubernetes should be solidly on etcd3 now...

arejobs kincleanup prioritimportant-soon siapi-machinery sirelease sitesting

Source

BenTheElder

All 23 comments

/assign @krzyzacy
please punt to whoever is appropriate to triage which version we should use

BenTheElder on 6 Apr 2018

/sig storage
/assign @saad-ali @msau42

krzyzacy on 6 Apr 2018

https://github.com/kubernetes/kubernetes/issues/41422

cc @mml @wojtek-t @jpbetz

ixdy on 6 Apr 2018

👍1

given all other jobs are on default etcd I think we'd just remove this (and use default)

krzyzacy on 6 Apr 2018

👍1

discussed some offline, this should probably be
/sig api-machinery

BenTheElder on 6 Apr 2018

Personally, I'm fine with removing those jobs.

But at least in theory, we still support etcd v2 (though we are in deprecation period).
So from that point of view we may be forced to still have these tests - so let's wait for @lavalamp and @jpbetz thoughts.

wojtek-t on 9 Apr 2018

I've discussed this some more in other venues and it sounds like we need to have some etcd2 testing until 1.12, but I think our primary CI and presubmit jobs blocking on otherwise-defaults GCE e2e should actually be completely-defaults with maybe some other dedicated etcd2 testing jobs.

BenTheElder on 9 Apr 2018

xref: https://github.com/kubernetes/test-infra/issues/6114#issuecomment-385608767

BenTheElder on 1 May 2018

Issues go stale after 90d of inactivity.
Mark the issue as fresh with /remove-lifecycle stale.
Stale issues rot after an additional 30d of inactivity and eventually close.

If this issue is safe to close now please do so with /close.

Send feedback to sig-testing, kubernetes/test-infra and/or fejta.
/lifecycle stale

fejta-bot on 30 Jul 2018

/remove-lifecycle stale

BenTheElder on 30 Jul 2018

Per discussion today in #sig-release we will hold on this until q4
/assign
/assign @spiffxp

BenTheElder on 1 Sep 2018

Other issues related to etcd2's deprecation:

We announced in v1.9 that etcd2 is being deprecated "1.13 or 1.14"

I'm not planning to unpack which of those versions it is just yet. But I will gladly push for it to be 1.13 as we get closer to it.

spiffxp on 1 Sep 2018

👍1

/kind cleanup
/priority important-longterm

spiffxp on 1 Sep 2018

Is this important longterm only?

Or given https://groups.google.com/d/topic/kubernetes-dev/jMPA4JzKiY4/discussion is this something we should be doing NOW critical urgent?

tpepper on 6 Sep 2018

My understanding was that the most recent update from sig-arch stated we'd only be dropping etcd2 support fully by 1.13, at which point we'd want to drop it from testing.

BenTheElder on 6 Sep 2018

That sounds right to me as well.

jpbetz on 6 Sep 2018

sgtm then

tpepper on 6 Sep 2018

It's been at the top of the release notes as being deprecated and scheduled for removal in 1.13 in 1.10-1.12.

https://github.com/kubernetes/kubernetes/issues/69310 is open to remove etcd2 support and cluster-lifecycle (specifically kops) is working through switching defaults to etcd3 as well.

liggitt on 2 Oct 2018

yeah I've been trying to watch that, @spiffxp was talking about tackling this from the test side the other day I think :^)

BenTheElder on 2 Oct 2018

Yup, just opened a first pass, I'll see if I can give it some time tomorrow

spiffxp on 3 Oct 2018

/remove-priority important-longterm
/priority important-soon
/milestone v1.13
/sig release
/sig testing

spiffxp on 3 Oct 2018

/close
There are no more jobs that use etcd2

spiffxp on 8 Oct 2018

🎉1

@spiffxp: Closing this issue.

In response to this:

/close
There are no more jobs that use etcd2

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes/test-infra repository.