Using buffered channel in the aggregation process #267
Description
Is your feature request related to a problem? Please describe.
We have multiple senders in the collection process with goroutines for each client, and multiple receivers with workers in the aggregation process. With the current implementation using the unbuffered channel, senders will wait for the worker to finish the processing of the message to move forward. In addition, other workers are not being run in parallel.
sched trace will show the usage of logical cores and this will let us show the extent of parallelism.
In performance tests, we should check the above. Because of this, with UDP clients sending to the collector process, we may see record loss. Moreover, with TCP clients, we won't be getting good performance in terms of bandwidth between the client and server in the collection process.
The data needs to be looked but it is mostly the case according to the golang concepts
Till now we looked at memory aspects of the aggregation process in the usage with Antrea. CPU time needs to be looked at.
Describe the solution you'd like
Move to the buffered channel.
Additional context
Anything else which is relevant and may help us understand the issue.