Autoscaling killing pods

[daudn]

I managed to implement the autoscaler so that it scales based on the number of messages in the queue. However, initially, it scales appropriately and consumes the messages on the queue. Once these messages are consumed and no longer on the queue, the autoscaler scales the pods down and the messages are put back on the queue. This is more or less an endless cycle.

My process works:

1. n jobs are put on queue
2. n workers are spun up
3. n jobs consumed by n workers.

However if the queue becomes n-1, pods also scale to n-1. And so my jobs never really get processed.

Any suggestions on how I can work around this? Cause currently, it seems like this may not work for me.