We want to understand why CPU usage is so low during a time of heavy extracts? And if there is a server configuration or hardware change we should make to speed up the Extracts?
We are running Tableau Server 10.1.1 with on Gateway and 3 worker nodes. Each Worker has 8-CPU and 32GB Memory. Each Worker has 4 backgrounders (for a total of 12).
All 12 backgrounders are busy, but CPU hovers around %10 on each Worker. The File Store on each Worker is Busy (Synchronizing), but every other Processes is in Active state.
We see there are 12 “In Process” Background Tasks, which makes sense given all those procs are Busy. There are also 4 Pending Background Tasks.
This occurs for at least 5-10 minutes (not sure how long system was in this state prior to observing it) and the Backgrounders eventually start to free up, pending tasks get execute, and CPU spikes occasionally (as we would expect during high load).
Everything we’ve read says backgrounders are CPU-intensive and recommendations are around adding Backgrounders threads and/or CPUs. Our observations are contradicting this logic/approach.
Has anyone noticed similar behavior? Thanks!
Everything you say checks out as far as I can tell except for one thing - the 4 backgrounders on an 8-core is a bit scary to be honest and not something I would recommend to the majority of customers.The fact that your CPU is so low is suspicious indeed and really just means that Tableau isn't doing much in the way of constant disk writes and data compression/pivot/optimization. This indicates a bottleneck upstream - specifically disk-write speed, I/O, network bandwidth, or the most common - a database which isn't prioritizing your requests in such a way that it's giving you data as fast as it can (or the database itself is just slammed and can't keep up with the requests). Look at the necessary places to monitor all of these and I'd imagine you find a smoking gun somewhere.
