and since requests can hit different nodes, those nodes have different caches

right... okay

if it helps, the key point to understand about the cache is that its a tradeoff: most permissions questions can be answered with slightly lower consistency in favor of performance and scale

when the caller deems that not sufficient, they can use either a zedtoken or full consistency to ensure everything is always up to date, but at the cost of that performance

gotcha

but the vast, vast majority of applications do not need fully consistent at all times

that's why Zanzibar (and therefore, SpiceDB) is architected this way: both options are available so that applications can decide whether they need consistency vs performance

and, keep in mind, the cached consistency still generally catches up within a few 10s of seconds max (its generally much, much faster than that)

I'm just being conservative in my numbers

I hope i dont annyo to much with all the questions... So i try some stuff out... but i get quite some weird behaviour... the same permission check kind of oszilates in validation... i dont do any changes and that even happens after a reboot of the spicedb instance and after at least 1min since the last change...

(it should always be false)

did you write or delete a relationship before hand?

i deleted one

and how many SpiceDB nodes are you running?

just a single one (one single docker container)

okay, just making sure. anyway, with the single delete, the system is likely picking revisions before and after the delete and using that timestamp for lookup

i thouhgt it should sync after under 10s?

it should pick the latest revision after that period, yes (assuming your database's clock is correct)

now it seems to always return false... but that only happend after plenty of minutes....

could be that your clock is off

how... do i turn it on? xD you mean on spicedb, right?

this is my docker command to run it

docker run --rm -it -p 50051:50051 -p 8080:8080 authzed/spicedb:latest serve --grpc-preshared-key="somerandomkeyhere" --datastore-engine=postgres --datastore-conn-uri="postgres://fin_repo:test@host.docker.internal:5432/FicsItNetworksRepositoryAuthorization?sslmode=disable"

Docker is known to have clock issues over time when running containers

so if you're running Postgres in a container, its possible its clock is different from SpiceDB

i run posrgres as install on windows as service... i just run spicedb here in docker for my dev env

I'd have to look into it in more detail to know for sure

but it should converge faster than a few minutes, even with caching (could be up to a minute though)

its definetly more than 5min

could it be because i run it on docker for windows using WSL2 engine? (spicedb)

probably not