there is also the dispatch cache

beyond that, if you fire a SUPER big query (like an insanely large LR), it can just OOM just book keeping

the amount of memory is based on the caches + whatever is needed to process the requests

the memory recommended is based on your work load

ahhh that makes sense

we'll just have to monitor and tune

and also figure out how to get memory metrics into datadog ^.-

are those in the prometheus metrics?

should be

👍

and GC

default go metrics include memory

if you do encounter an OOM from a known query

I'd be curious to see the memory export

always places we can improve

for example, if you're on an older SpiceDB (pre v1.23 I believe), we used to load all rels into memory for a non-cursored ReadRels API call

which can become problematic

latest SpiceDB now batch loads and then streams out, then loads more, etc

i think we're on 1.24

so yeah, if you can repro and get a memory export

we can take a look

but it really depends on the query

i'm digging through our traces. thank you!

1.25 has some mem opts I believe, and the new namespace cache updated is definitely worth a Go if you are on cockroach, we observed massive gains

Hi, I'm trying out spicedb and playing with some toy schema on the playground. What I want to achieve is to allow runtime grant to any user as long as certain condition is met, on top of direct grant. I'm using caveat to achieve this and came up with: https://play.authzed.com/s/WPKlZtJ2f4Nz/schema. Running validation results in

did not find expected key

error. Does someone know what's wrong? Thanks in advance.

wrap the assertions with

with

statements in

'

ala

the spaces are confusing YAML

I've seen various uses of

...

in checks. Could someone point me to docs that explain what the ellipsis means? I'm struggling to find the docs for some reason