Here is a result with negating:

http.response_time:
  min: ......................................................................... 563
  max: ......................................................................... 9926
  mean: ........................................................................ 4299.6
  median: ...................................................................... 3984.7
  p95: ......................................................................... 8024.5
  p99: ......................................................................... 9230.4
http.responses: ................................................................ 396

and without by your request:

http.response_time:
  min: ......................................................................... 374
  max: ......................................................................... 9089
  mean: ........................................................................ 2564.5
  median: ...................................................................... 2018.7
  p95: ......................................................................... 6312.2
  p99: ......................................................................... 8024.5
http.responses: ................................................................ 390

A little better, but still far from what we need. We are running e2e load test, but just in case, I attach a screenshot with example of such trace to assure you that bottleneck is SpiceDB, but not something else https://cdn.discordapp.com/attachments/844600078948630559/1350114521381081190/image.png?ex=67d58fa1&is=67d43e21&hm=0345b5a499cbd02508cf356b2869bb555ed539ee945f9b7413c90dc38255203b&

which version are you using? there is a bug in check bulk that is fixed in main branch, so I suggest you check the nightly build and test it

HI. Thanks for reply. I tested several times on

latest

tag, but to be honest I didn't notice any performance impact. Here is a trace example that looks similar between v1.41.0 and latest tag. It's definitely database issue, because to implement our sophisticated requirements we have to check a lot of permissions and relations which cause many queries to database. So far, I don't see a problem in SpiceDB itself, rather in SpiceDB schema capabilities (using existing operators it's hard to express our requirements in easy way) or we lack of understanding how to model our requirements in effecient way. Thanks! https://cdn.discordapp.com/attachments/1350114521594986567/1350809224317239441/image.png?ex=67d8169f&is=67d6c51f&hm=a1c5143ed7dc3f9c718b50f948c05c5b425780d1008c4991a73011edcde83a36&

that graph shows a synchronous bulk check

I see. Any plans to release this fix to stable version v1.41.1?

it'll go out in 1.42.0 most likely

Could you also check quickly my question here? https://discord.com/channels/844600078504951838/1350114710690988082/1350810417529749545