What this choice changes
These actions answer different evidence. More CPU increases the compute budget; lower concurrency reduces simultaneous pressure on memory, connections and shared dependencies.
Compare the decision factors
| Factor | Add CPU allocation | Lower worker concurrency |
|---|---|---|
| Best evidence | Sustained runnable work and profiled CPU time on the useful path. | API latency or dependency pressure worsens as more jobs run together. |
| Immediate effect | More compute can execute CPU-bound work concurrently. | Fewer jobs compete at once; queue age may increase. |
| Memory | Does not add RAM unless selected separately. | Usually lowers simultaneous worker memory. |
| Database connections | Does not change pool demand by itself. | Can reduce active worker connections. |
| Does not repair | Duplicate work, slow external calls or inefficient queries. | A single CPU-heavy job or permanently inadequate throughput. |
| Measure after change | CPU time, useful completions, latency and errors. | Oldest-job age, completions, latency, memory and connection waits. |
When each option earns its place
If CPU stays low while queue age and database waits rise, lower concurrency and inspect the database first. If a profiled job keeps cores busy and useful completions rise with parallel work, compare a larger CPU allocation.