Tony Devlin <tonydev...@gmail.com> wrote: > Thank you Eric, > > I will look into the other worker to see what is going on with it. I still > appreciate any hints you all can give me on where I can check. I'm also > looking into the OS TCP timeouts to see if what Daniel said may be a > problem.
General rule for me is to get the problem reproducible in the smallest possible way. That could mean removing features, cutting out large chunks of code, cutting out certain request types, reducing workers. More things: 1) Can you make sure nginx is not trying to maintain persistent connections? nginx should respect unicorn closing the connection but I haven't checked the latest versions of nginx. lsof can help here, too. unicorn currently does not do persistent connections, allowing an M:N relationship between nginx instances and unicorn instances[1] 2) Any other odd external dependencies such as NFS mounts, file locks, FIFOs, etc? [1] Perhaps persistent connections will be an option in the future if the support/documentation overhead is worth it, as nginx supports persistent connections to backends nowadays.