The token isolation. I think a lot of our problems are that every test assumes a clean slate, so we re apply the spicedb schema, seed data, and then try to tear it all down. and multiple tests maybe doing that at the same time

the problem is: to do that against a non-in-memory datastore would require either tearing down/resetting the database each time or using different logical databases internally, which isn't easy to do nor performant

that's why the testing mode is memdb based: its ensures fast test setup/teardown

creating a new database + tables is pretty fast

I'd take that

Hi Guys. We are currently trying out spiceDB and we hit an dilema. We believe cockroachDB is our best option for datasource, but we don't want to deal with it now as we want to first try it out. Is there a way to migrate data from postgres into cockroachDB?

@symion worth a proposal issue

The docs don't really describe what a token is in this context.

its whatever you want

ah

just sent in as the auth header token

you could call ReadRelationships from the old cluster and call WriteRelationships to the new one; another option would be to use a SQL dump, but that would have to be massaged a bit

this is different than the grpc shared key?

its in place of it; you just send whatever you want instead of the expected shared key

oh, ok. in testing mode, the server and client don't need to match?

in testing mode the server doesn't have a key

it accepts any token/key in the Auth header, and uses that as the isolation key

could the watch service be used in this scenario? or is that only for "present" events?

if you wanted to move over time, it could be

but if you're doing a single point in time transfer, Read is the preferred solution

for this scenario, over time would actually be better. (no down time? 🤔 ) Thank you for the reply

yeah, in that case, you could do an initial Read to fill in the existing data, then Watch to move it over as time goes on

or just start with Watch

to what extent?

moving to cockroach is actually a "future worry" as postgres will most likely to be enough for a while. Thank you, I will try it out.

how much massaging of what?

the SQL dump; Postgres datastore and CRDB datastore table schema is similar but not the same

a straight data dump should be pretty portable, no?

no; Postgres stores relationships with transaction IDs while CRBD uses its internal tracking

What is purpose of "nil" as described here, https://github.com/authzed/spicedb/blob/main/internal/services/testconfigs/nil.yaml