Plugins » Katello

Applied in changeset katello|55677460eea47e2d34a04364b485daaf59296cdc.

This actually causes foreman-tasks to die and stop processing. We need to improve this code, but removing it causes more issues than it solves

# hammer ping
foreman_tasks:  
     Status:          FAIL

putting the sleep back in and it goes back to normal status

Reasoning of the regression: removing the sleep, the thread was in a tight loop, suppressing potential activity of other dynflow threads that starved on CPU. Like hammer ping.

Code proposal:

          loop do
            begin
              message = fetch_message
              if message[:result]
                result = message[:result]
                @session.acknowledge(:message => result, :sync => true)
                suspended_action.notify_message_received(result.message_id, result.subject, result.content)
              elsif message[:error]
                suspended_action.notify_not_connected(message[:error])
                break
              else
                sleep 1
              end
              sleep 0.01 # sleep 1 was originally here
            rescue => e
              suspended_action.notify_fatal(e)
              raise e
            end
          end

I.e. call "sleep 1" only when no message was fetched from the queue, and have short sleep after fetching a real message - again to give other threads access to CPU.

I am not in favour of having only "sleep 0.01" without the "else sleep 1" option. That would mean katello periodically fetching qpid queue every 0.01 second, what is meaningless and redundantly takes resources.

Another alternative patch would be decreasing original "sleep 1" to "sleep 0.1" as a "compromise", but that would IMHO still redundantly fetch katello queue too often. Requirements to the two delays are contradicting each other, it doesnt make much sense to ave just one delay.

Applied in changeset katello|87ef553cfe840299389abbb4123a640c2227e0bb.