Throughput issue with SQS and long polling

[bjpirt]

I've opened an issue here - but thought I'd put something in the discussion too.

Conductor creates two queues for internal use for asyncronous processing which we don't use. We also create a queue to trigger workflows via SQS which gets used lots. All three queues were set with a 20 second long polling. The current implementation seems to check the queues in a serial manner and would therefore only pull one message off the active queue every 40 seconds. When we changed the long polling receive timeout to one second on the two inactive queues, the throughput went up accordingly.

Wondering if anyone else has run into this or if I'm missing some configuration. Obviously I can disable long polling, but I'd expect Conductor to be able to handle this as it's recommended config for SQS.

Also open to making a PR but this is stretching my Java skills so if anyone from the Conductor team can point me in the right direction I can take a stab at it