What does CurrentNIOWriteQueueWait mean exactly?

[carenot]

LS versions: 6.0.1

When I'm load testing LS I'm seeing a steady increase in the metric, as seen here


What does the metric mean exactly? The doc says it's the averaged wait time over all socket write operations to be selected by selectors. How can it be several million ms then? The LS is still running fine without any delays.

Thanks for any help

[DarioCrivelli]

The CurrentNIOWriteQueueWait statistics has to be considered in bundle with CurrentNIOWriteQueue.
The average is restricted to the sockets counted in CurrentNIOWriteQueue (usually very few). They are the sockets for which an immediate write was not possible, hence the resort to a NIO selector has been needed. For such sockets, the client side is not consuming and if this state continues for long time, there must be a serious block at network level (for these clients only).

If the condition persists, a blocked client will accumulate wait time; note that a 1 million ms time is about 17 minutes. So, if all sockets currently blocked have been waiting for long time, the average can reach such levels. Note that, by default, the Server doesn't enforce a timeout on write operations; you can set one through the <write_timeout_millis> configuration setting.

Since we provide just average information, the chart is not regular: sudden changes in CurrentNIOWriteQueueWait depend on the start and termination of blocked write operations.