sorry

you can't use postgres replication

postgres doesn't scale quite as well because you can't use postgres replication

because then you could run into the exact problem you theorized

okay

my sense is that that problem is probably relatively unlikely, but that's useful information to guide our discussions

even if you could work around that problem, you're still left with the issue that all writes would have to go solely to the master

so while you can have failover, it is a single point of failure (however briefly) for writes

yeah. that one might not actually be a problem, because for the way our system currently works, all writes would be happening through a single region (the master's region) anyway

right

but that's another thing to raise

now, there is a potential solution for the zedtoken issue

but its currently just a vague thought experiment around having either an auto-retry to switch to the master for reading if the replica is not caught up yet

or somehow knowing where each replica is and using the master immediately if the replica isn't yet caught up

yeah, i think i'm gonna leave that one by the wayside. i'd rather steer them towards a solution that's more operationally complex but easier to reason about as a system.

yeah

CRDB or Spanner are definitely the more scalable solutions

Zanzibar was designed for spanner, after all 🙂

yep 🙂 i've been enjoying reading through your blog posts

it's also fun to watch the cycle of google creating a problem that nobody else has and then google solving that problem in a highly sophisticated and bespoke way

it's cool too to see how you guys have worked around those problems using cockroach

that's what we're here to do!

but @yetitwo, if you do decide you want to explore the Postgres option more, we can certainly work on a proposal with you on how to support replicas there

however that may be

thank you! i appreciate the offer

Actually defining schema is fine, but my worry is, in such policies, if there are 100K users , will end up creating equal number of relationship?

@Singha1 most likely, yes, but that's fine

depends on how your schema is defined

I see, any suggestion how to optimize such cases?,

let spicedb do what spicedb does