The source project of this merge request has been removed.
Reduce peak memory usage by shortening lifetime of response buffers
Due to high concurrency, we keep a lot of response buffers alive while the harvester wait for their I/O to reach the blocking thread pool. This change shortens their lifetimes where appropriate to reduce our peak memory usage and also removes some redundant parsing from the RLZ harvester.
(We are currently at 80% peak usage of the VM's 8GB RAM which isn't much but still seems more than we should need. Hopefully, this change allows us to stick to the smaller VM a little longer.)