On Wed, 2017-01-04 at 09:36 +0000, Adel Boutros wrote: > Thank you Andrew for this clear explanation. > > > Epoll is not supported on Solaris. So we will have to think of an > alternative for this OS (poll which you suggested should work). > > > As for libuv, we are currently compiling Proton 0.16.0 without it > because we don't have bandwidth to check if libuv is working on > Solaris. I will have to discuss internally to see if we can allocate > some bandwidth to dealing with libuv on Solaris. If you have any > input at this level, we will highly appreciate it. > > > Alan Conway stated in another discussion that libuv should fix the > slowness we detected in the event injection API. I don't know if > poll/epoll would also address this issue which is a bit critical for > us. >
To clarify - the event injection problem is not related to libuv specifically but is in code that is getting re-organized by the proactor IO rework so I was hoping to avoid fixing it twice. For solaris any of libuv, a native poll() proactor, or patching the injection problem directly in the old reactor-based code are possible resolutions. Ideally I'd prefer to bring solaris forward to proactor- based code that also solves the inject problem, but we may have to fix the inject problem in the old code first if that can't happen fast enough. Cheers, Alan. --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
