Improve memory usage with large number of unreachable nodes

Confluent memory usage over time was growing when trying to deal with unreachable nodes. This has been improved.

Various improvements to Ubuntu deployment support

The limited support for Ubuntu has been expanded

nodeconsole -t will now reuse existing tmux if started from within tmux

This will more intelligently use a detected in-use tmux session.

Add nodesensors for Cooltera CDUs

Cooltera CDUs will now provide some data to nodesensors.

Improve collective behaviors

  • When a collective leader goes down, provide better recovery and better ongoing monitoring.
  • non-voting members will no longer be promoted to the role of leader, aligning with likely treatment of non-voting members.
  • Slow dispatch operations will no longer hang up an entire confluent instance

New nodeapply command

The nodeapply command serves a similar role as updatenode did in xCAT: shorthand for a number of common operations.

Support to dump timestamped logs in nodeconsole

The nodeconsole command adds the -T option to allow writing a log with in-line timestamps.

EL9 is now supported for cloning

The command imgutil capture will now work on EL9 based nodes.

Custom roles support

Custom roles may now be specified in /etc/confluent/authorization.yaml. The roles are defined in a way similar to auth.py. This file must be consistent across a collective.

Various bug fixes

  • ProxyDHCP functionality is more robust in face of errored clients
  • SLP functionality persists in face of misbehaving SLP responders
  • Correct some misbehaviors on terminals in WebGUI (serialized operations, unrecognized functions)
  • Improve ansible behavior when ansible python version differs from confluent
  • Improve ansible behavior when a partial ansible inventory is provided
  • Fix SSDP behavior in EL9 management nodes