Discussion support a restart/kill meta statement through psql

[xxchan]

<@U030W13SZKN> <@U034X02TB28> <@U030W140JEQ> are discussing if we can support a restart/kill meta statement through pgwire. We are finding the similar function in PG

Slack Message

[st1page]

It is for the users who have no control of deployment and their k8s cluster

[lmatz]

I love <@U030W13SZKN> <@U034X02TB28> <@U030W140JEQ> you all

but seriously,

It is for the users who have no control of deployment and their k8s cluster

I think this reasoning is quite important, as I always find:
the people who want to use RW are often not the people who have good knowledge of or have control of/access to the deployment and their k8s cluster, e.g. application developer versus infra operation engineer.

Once there is some problem, there is likely to be some back and forth which, as this quote suggested, better be avoided
especially at POC stage

Maybe we want to make sure as many necessary functionalities that needed to be done via risectl or kubectl should also be able to be done via psql.

If for safety concerns for messing things up, critical functionalities can be activated only by doing some extra, password or typing a confirmation message

[xxchan]

Since we already have #12999, this is less urgent for now.

[CAJan93]

I kinda like the idea. I think it is great for people who host their own RW instance.

In our cloud operation, I think we should be mindful with this. Giving customers a kubectl via psql sounds like a security risk to me, even if this is read-only. Unless we limit this somehow to their own tenant ns.

If the customer is able to kill meta via psql we provide them a tool to shoot themselves in the foot. If we provide this, I think we should make it clear in the docs to use this carefully.

CC @wjf3121

[lmatz]

That's a good point, one approach is to disable these "in-sql" functionalities in the RW image for RW Cloud.

I imagine my points above are only applicable to users who deploy RW in their private cluster as RW Cloud/BYOC both have nice GUI for users to do all kinds of admin stuff.

[wjf3121]

In our cloud operation, I think we should be mindful with this. Giving customers a kubectl via psql sounds like a security risk to me, even if this is read-only. Unless we limit this somehow to their own tenant ns.

As long as these features are flag protected and turned off on our managed cloud, we should be good.

[BugenZhao]

This may also help continuously integrate the tests for the error reporting for cluster recovery. 😄

• feat(meta): show recovery cause when request is rejected #13836

[github-actions]

This issue has been open for 60 days with no activity. Could you please update the status? Feel free to continue discussion or close as not planned.

[lmatz]

Is this sort of overlapped with #13142

[xxchan]

I think it's duplicated with #14700, and already done