We have been seeing latency spikes in our internal method calls in JDC. We need to investigate the root cause of this. A few of the primary suspects are:
- We are using nested locking across many of our data structures. Migrating to simpler locking schemes could help reduce contention and latency.
- Moving compute-intensive tasks to a blocking thread pool, so we do not block the main thread pool with bulky tasks.
- We can start batching responses to each downstream when we have a large number of messages to send.
Attaching the hotpath latency values for a JDC run (this doesn't seem as bad as pool though).
We have been seeing latency spikes in our internal method calls in JDC. We need to investigate the root cause of this. A few of the primary suspects are:
Attaching the hotpath latency values for a JDC run (this doesn't seem as bad as pool though).