Curious. FWIW, I've been using Ubuntu 13.10 and a custom-compiled Linux 3.11 kernel without issues. I wonder if Ubuntu is doing something weird to the kernel, or it's just to do with your backend setup.
What kernel are you using now? The only thing that strkes out at me that could be related that has changed recently is tcp_early_retrans. Although maybe there's something I'm missing. Ben On Mon, Jan 13, 2014 at 05:42:09AM +0000, David Jericho wrote: > Following up with this, after much appreciated assistance (thanks Ben, > Leif and Yongming!), the system started behaving once I rebooted to > something other than Ubuntu 13.10's 3.11.0-12 kernel. Why I didn't try > that earlier, I don't know. > > > > Nearly twenty million successful requests later, I can confirm there are > no stuck cache reads, and no users reporting stalled sessions. I'm unable > to replicate the behaviour myself, even being able to trigger it on the > first query. > > > > David Jericho > > Senior Systems Administrator > > AARNet Pty Ltd
