Willy / Lukus, sorry for slow response - got distracted by the end of year!
We do set the back log via the java API's but perhaps that is insufficient to really set it. I will try setting it by other means and get back to you. cheers On 24 November 2015 at 08:33, Willy Tarreau <[email protected]> wrote: > On Fri, Nov 20, 2015 at 08:37:58PM +0100, Lukas Tribus wrote: > > >> > > >> So anybody know what resource "free ports" relates to in the unix > > >> domain socket case? Are there any other debug options to find out > > >> more about what is happening. > > > > > > I suspect the connect() call returns EAGAIN > > > > Digging some more, it looks like the kernel returns EAGAIN when the > > backlog on the other side is full (the backend doesn't accept() fast > > enough). > > Yes this reminds me something as well. Greg, did you change the > backlog size on your server ? It's frequent to see values as low > as 128 or even 5 in some programs, maybe you're in a similar > situation ? It could be interesting to see if increasing the > net.core.somaxconn sysctl helps here. I have an old memory of it > being used by Unix sockets but I don't remember exactly how. > > Cheers, > Willy > > -- Greg Wilkins <[email protected]> CTO http://webtide.com
