Sumeet,
Work message queue length is not considered a tunable. The queue length may be different from one configuration to another and one site to another. That difference is based on the number of nodes.
For configurations with fewer than 16 nodes, the message queue length for each work type is 20; otherwise it is the number of nodes + 5. The message queue length limit is for each work type individually, and few-AMP messages have their own flow control gates compared to all-AMP messages of the same work type.
For example, if you have 20 nodes, the message queue length limit for all-AMP messages of the Work00 worktype will be 25. The same limit of 25 would be in place for all-AMP messages of the Work01 work type.
I recommend sticking with the default queue length settings. If you believe they should be something different, you would need to call the support center and ask for their help and advice.
A longer message queue can result in an AMP going into flow control less quickly because there will be more room on the queue for more messages before messages start to be retried. On the other hand, longer message queues use more memory. And no matter what the message queue length, if you are out of AMP worker tasks, queries will be degraded anyway because they have to wait longer to get one and start work, either in the queue or doing retry logic.
The workload management focus is better placed on preventing the exhaustion of AMP worker tasks so you never fill up the message queue. Use the message queue as a safety valve should you ever need it, not as a regularly used staging area that often fills up.
Thanks, -Carrie
Sumeet,
Work message queue length is not considered a tunable. The queue length may be different from one configuration to another and one site to another. That difference is based on the number of nodes.
For configurations with fewer than 16 nodes, the message queue length for each work type is 20; otherwise it is the number of nodes + 5. The message queue length limit is for each work type individually, and few-AMP messages have their own flow control gates compared to all-AMP messages of the same work type.
For example, if you have 20 nodes, the message queue length limit for all-AMP messages of the Work00 worktype will be 25. The same limit of 25 would be in place for all-AMP messages of the Work01 work type.
I recommend sticking with the default queue length settings. If you believe they should be something different, you would need to call the support center and ask for their help and advice.
A longer message queue can result in an AMP going into flow control less quickly because there will be more room on the queue for more messages before messages start to be retried. On the other hand, longer message queues use more memory. And no matter what the message queue length, if you are out of AMP worker tasks, queries will be degraded anyway because they have to wait longer to get one and start work, either in the queue or doing retry logic.
The workload management focus is better placed on preventing the exhaustion of AMP worker tasks so you never fill up the message queue. Use the message queue as a safety valve should you ever need it, not as a regularly used staging area that often fills up.
Thanks, -Carrie