what is the recommended number of read
# spicedbm
make it go faster
12/05/2023, 8:24 PMwhat is the recommended number of read connection pool that should be defined in spicedb for connecting to a cockroachdb backingstore runing a 5 node cluster with 8 cores per node
e
ecordell
12/05/2023, 9:45 PMthe general recommendation from the cockroach folks is 4 connections per core
so you're looking for 160 connections to divide between the read and write pools of however many spicedb nodes you're running
m
make it go faster
12/05/2023, 9:46 PMso 80 connections for read and write each per spicedb pod
make it go faster
12/05/2023, 9:47 PMor are we dividing 160 connections across N pods?
e
ecordell
12/05/2023, 9:47 PM160 should be the total, so if you have 4 spicedb nodes you'd have 40 connections per node to use
most folks don't have a 50/50 read write workload, so you might choose each node to get 30 read and 10 write, or 35 read and 5 write
ecordell
12/05/2023, 9:47 PMyeah
m
make it go faster
12/05/2023, 9:48 PMoh I see it's 160 overall
make it go faster
12/05/2023, 9:48 PMinteresting this brings an interesting question is that we have to tweak this number every time we scale up or down the spicedb pods
e
ecordell
12/05/2023, 9:50 PMyes, we've discussed methods for sizing the pool automatically but don't have it implemented yet
in practice though most people don't need to scale spicedb super dynamically; you'll size it for your peak load + a buffer and then track when your peak load is changing over time
ecordell
12/05/2023, 9:52 PMalso: it's not the end of the world to over or undershoot this recommendation
ecordell
12/05/2023, 9:53 PMcrdb performance may degrade somewhat, but a lot of that can be mitigated with the connection pool jitter flags