Skip to:

  1. Skip to Jira Navigation
  2. Skip to Side Navigation
  3. Skip to Main Content

mp listen queue backlog too small

Description

Because mp connection handshake requires sends and receives before another accept is called there is a good chance when many slaves are started at once that the connection backlog could exceed the current setting of 16. This increases it to 800 which should be sufficient for our largest clusters. NOTE: the kernel tuneable "somaxconn" should also be increased to handle this plus other thors on the same host.

Conclusion

None

Activity

Show:

Mark Kelly July 29, 2015 at 7:53 PM

Code checks for environment setting mpSoMaxConn at MP startup and if found compare to kernel somaxconn to warn user if needed.
mpSoMaxConn is defaulted to 128 in initial configuration but can be increased by user.

Mark Kelly July 28, 2015 at 8:52 PM
Edited

Warning added when number of connections exceeds kernel somaxconn value.
Could also maybe check other side for connect() failures and give warning, but that should be a separate task.
Ultimately should launch or pass to another thread to complete the MP handshake so that accept()s happen very quickly to reduce chance of backlog growing large.

Fixed
Pinned fields
Click on the next to a field label to start pinning.

Details

Components

Assignee

Reporter

Priority

Fix versions

Pull Request URL

https://github.com/hpcc-systems/HPCC-Platform/pull/7614
https://github.com/hpcc-systems/HPCC-Platform/pull/7614

Created July 27, 2015 at 1:32 PM
Updated July 31, 2015 at 1:11 PM
Resolved July 31, 2015 at 1:11 PM

Flag notifications