On 2013/03/14 17:23, Rogier Krieger wrote: > On Thu, Mar 14, 2013 at 3:16 PM, Stuart Henderson <[email protected]> > wrote: > > On 2013/03/14 14:28, Rogier Krieger wrote: > >> I ran into a problem with the default value of authenticate_timeout > >> (1800s) that causes EDOM on setsockopt() with SO_RCVTIMEO. > > The actual max values depends on the resolution of the kernel timer > > (HZ). 60s seems reasonable I think, people can always fiddle with config > > if they want. > > On my test system, I could crank it up to 655s. Any higher gave me > EDOM. I had trouble finding the source for this, but the patch file > gave a good hint. I'll contact the author and will try to get this in > upstream.
Yes this (65535/100) would be normal for most arch, but iirc some use a different HZ value (e.g. I think loongson may use 128 instead of 100 i.e. max would be 511). > > My diffs had the library major version bumped. I don't recall the particular > > reason but I will have diffed the two source trees (old and new) to > > determine > > whether any function parameters where changed or whether any functions > > were added/removed. > > So that's an area where I'll look at the docs again. I had simply > upgraded, assuming the '0.nx had to do with upstream versioning > numbers being 0.0.n. http://www.openbsd.org/faq/ports/specialtopics.html#SharedLibs > > New diff below incorporating the above [...] > > Testing this as we speak and all seems to be fine (amd64, -current Mar > 1st 2013). I''ll test it with a bit more load and report back whether > I see anything odd. Thanks for looking at this; cool, thanks. let me know if you're happy with it under load and I'll commit it.
