Fix the reconnection logic for parallel writes with HA Producer.

[hiimjako]

If there are parallel calls to Send or BatchSend in concurrent routines during reconnection, only one of these sends will unlock and send the message, while the others will remain in a locked state. This is because currently there isn't a broadcast notification with reconnectionSignal. To fix this issue the reconnectionSignal channel has been changed to a sync.Cond.

[Gsantomaggio]

Great find! Thanks @hiimjako

May I ask to change the example by adding a number for concurrent producers?, like:

   const concurrentProducers = xxx

Then you should test the example with these parameters by restarting the node.

I use:

declare -a arr=$(docker ps -q)

for image in $(docker ps -q); do
   echo Stopping ..$image
   docker restart $image
   echo Waiting 60s ..$image 
   sleep 40
   echo Waiting 20s ..$image
   sleep 20
   echo Waiting 10s ..$image
   sleep 10
   echo Waiting 10s ..$image
   sleep 10
   echo Waiting 10s ..$image
   sleep 10
done 

And I leave it running for some time. The result numbers should be coherent. thank you

[hiimjako]

May I ask to change the example by adding a number for concurrent producers?

Done! In the main branch the example returns more or less half of the messages, as expected, while in this branch all 20M messages.

[Gsantomaggio]

Thank you