How to temporarily stop serving?

[josephlewis42]

In what area(s)?

/area autoscale
/area networking

Ask your question here:

What's the process to temporarily stop a serving.Service from scaling at all without deleting it and removing all of its history?

An example: let's say there's a mistake in your cluster where kservice A is handling user traffic and is causing an amplification issue to kservice B which is wrecking your databases. We'd like to be able to temporarily stop B (scale it to 0 and keep it there until we give it the go-ahead).

[markusthoemmes]

The way to do it would be via a traffic split. You'd assign 0% traffic to the revision that wrecks things.

[josephlewis42]

Awesome, thanks @markusthoemmes. One more follow-up if I may, will that also work if there's only one revision? Otherwise I guess we could push a dummy service as the first one that we can always fall back to in the event of an issue.

I'm hoping the percentages are true percentages rather than what I think Istio VirtualServices do which is proportion of the total values defined.

[vagababov]

I don't think that would work. The sum must be 100%.

…

On Wed, Jun 5, 2019 at 8:42 AM Joseph Lewis III ***@***.***> wrote: Awesome, thanks @markusthoemmes <https://github.com/markusthoemmes>. One more follow-up if I may, will that also work if there's only one revision? Otherwise I guess we could push a dummy service as the first one that we can always fall back to in the event of an issue. I'm hoping the percentages are true percentages rather than what I think Istio VirtualServices do which is proportion of the total values defined. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#4262?email_source=notifications&email_token=AAF2WX4E2F2L2YAQLHLKVV3PY7NGFA5CNFSM4HT2Y22KYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODXAEDTQ#issuecomment-499139022>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAF2WX2DCZO5UW5XPLBBQ6DPY7NGFANCNFSM4HT2Y22A> .

[mattmoor]

To stop autoscaling you can patch the PodAutoscaler resource with minScale/maxScale annotations with the same value, which can also be lower than the current scale of the K8s deployment.