Open ctron opened 1 week ago
Alternately - in the future we might consider a (read only) graphql interface built on top of a graph representation- that will probably be more contained in terms of memory usage. Otherwise maintaining a graphql on top of relational models may mean we have to constrain interface to avoid problematic usage areas.
tagging @gildub to investigate
While working on the loadtests, I noticed excessive memory usage of the
trustd
API process. I had to increase the memory fortrustd
from 2 GiB to 16 GiB to have it success at least once.It turned out that the culprit was the grapgql endpoint. Disabling those tests made the loadtest success with 2 GiB.
While I do understand that just disabling the tests isn't the best idea, it's the only one that works for running those tests on the GitHub CI, as we only have 7 GiB total there.
On the other side, if a single graphql user can cause a memory consumption of more than 10 GiB, then I think we have a problem that needs to be addressed.
I would propose the following steps: