Details
Critical Description
Retry polices should not use ThreadUtil.sleepAtLeastIgnoreInterrupts. This prevents FsShell commands from being aborted during retries. It also causes orphaned webhdfs DN DFSClients to keep running after the webhdfs client closes the connection. Jetty goes into a loop constantly sending interrupts to the handler thread. Webhdfs retries cause multiple nodes to have these orphaned clients. The DN cannot shutdown until orphaned clients complete.