Benchmark the effect of decrypting secrets on Rails instead of Agent
MR: Pending
Description
In gitlab-org/cluster-integration/gitlab-agent!1011 (comment 1504356228) , we decided to keep the decryption of secrets to be injected into the workspace at Rails side. We need to benchmark if there are any effects of this on the response time of the API especially for full reconciliation of the agent at scale(10 agents with 500 workspaces each with 100 secrets per workspace).
Currently, there is no lower bound to the full reconciliation interval. What are the negative impacts of that in this scenario.
Acceptance Criteria
-
Benchmark results of the performance of the API at without decryption at various scales of workspaces. -
If possible isolate how much of the request time is taken by db access, and by invoking the devfile
CLI via thedevfile
gem. -
Benchmark again with decryption at various scales of workspaces X secrets. -
Based on the benchmark results, take a call if decryption can stay Rails or needs to be moved to Agent. Maybe reimagining(some sort of pagination) full reconciliation might help?
Edited by Chad Woolley