HTTP request flow:

1. Incoming connection is put in accept backlog or rejected if backlog is full
2. Connection waits in backlog
3. Connection is accepted
4. Request body is read
5. Response is generated
6. Response is transferred back to client
7. Connection is terminated

Problem:
When many connections are in the middle of stage 6 - it can consume a lot of RAM.

Proposed solution(s):
Count current number (or RAM size?) of requests in stage 6 - and if limit is reached:

1. Don't accept new requests on stage 3
   or
2. Block requests from moving from stage 4 to stage 5
   or
3. Drop some old connections that are on stage 6 for long time

Need to decide

How much is it - Many connections in the middle of stage 6?

In p.1 clients will receive the message that server is overloaded straightaway and for them it would be clearer the root cause of the issue. That's why p.1 is better than 2 and 3

Moved to Blocked due to lack of testing resources on DevNet, needs to recheck during Regression testing 2.3

For now timeout was decreased to 5 sec from 30 min (default timeout 2 min), still need to check if it not too short timeout and make sure that the amount of memory consumed is sufficient

RAM 1,5 GB

Moved to 2.4 - need more investigate of performance with decreasing limits

Could be postponed to 2.5 release