Provision compute pools for code indexing web servers and compactors
Problem to Solve
The new code indexing design adds components with different resource profiles: web servers that keep a warm cache in memory, and compactors that are bound by local SSD throughput. The current Orbit node pools are sized for the indexer and query workloads and do not fit these profiles. Deploying the new components on the existing pools either starves them or wastes capacity.
Proposed Solution
- Provision separate node pools for memory-bound and SSD-bound components in the Orbit clusters, starting with staging.
- Use taints and node selectors so that only the intended components schedule on the new pools.
- Size the pools from the load expected at GA and document the numbers so that Dedicated and self-managed operators can reproduce them.
Agent context — extended analysis, alternatives considered, related-code dumps
Source: item 7 of the "Scalable Code Indexing" section in Orbit GA Planning. Estimated size: 2 days. Related: item 6, object storage support, #1271.