Index billing team members

[akshaymankar]

Fixes https://github.com/zinfra/backend-issues/issues/1296

[akshaymankar]

This PR is now reviewable, things which are still pending before this can be merged:

• Make a tool to backfill data in production
• Add the new flag to helm charts
• Add the new flag to production config
• Integrate the new tool with helm charts

[akshaymankar]

When we call updateTeamMember we always have the previous permissions, right? Can we check that instead of always deleting?

I just asked the question about it. I will add it. I wasn't sure if there was any problem with that or not. For my education, can you please tell me if it is really bad in C* terms or is it just that we should reduce a query if we can?

[tiago-loureiro]

I just asked the question about it. I will add it. I wasn't sure if there was any problem with that or not. For my education, can you please tell me if it is really bad in C* terms or is it just that we should reduce a query if we can?

I was mostly concerned about many tombstones but @jschaul concern about race condition is also a good point. Here's a contrived example of what I am thinking about: https://gist.github.com/tiago-loureiro/65db0c0fe8db66579609159ea1b97f86

Here's what I tried:

INSERT ... ('bla',0);
flush tables to disk
DELETE ... ('bla',0);
flush tables to disk
DELETE ... ('bla',0);
flush tables to disk

In there you can see that there was a row in the first sstable and now we have 3 sstables. Whenever we want to read the value, we need to go through all these records which can add up over time. The particular use case may not be as problematic but, if we can avoid it, then all the better. And I think that, in all our use cases, we don't even need an extra DB lookup, do we?

[tiago-loureiro]

Just a little sync with @fisx on the on the one issue would be good