I'm using the distributed approach, 1 queue, 10 consumers/containers. After some time, a consumer will "hang" in the sense that the streaming window closes and moves on, but the container from the previous window never finishes. I've noticed that this causes a big drag in performance.
In the picture below there should be 10 consumers/containers, but RabbitMQ shows 12. The prefetch count is at 100, and you see the messages ready and unacked messages at 200. This is the new floor as 2 of the consumers/containers are hung up. If I kill the job completely and restart, then these 2 hung consumers go away. Any idea what may be causing this? I've also pasted my configurations below.
Hi,
I'm using the distributed approach, 1 queue, 10 consumers/containers. After some time, a consumer will "hang" in the sense that the streaming window closes and moves on, but the container from the previous window never finishes. I've noticed that this causes a big drag in performance.
In the picture below there should be 10 consumers/containers, but RabbitMQ shows 12. The prefetch count is at 100, and you see the messages ready and unacked messages at 200. This is the new floor as 2 of the consumers/containers are hung up. If I kill the job completely and restart, then these 2 hung consumers go away. Any idea what may be causing this? I've also pasted my configurations below.