I haven't encountered the problem myself, but I see where it could happen, thanks to the bug report. This change should fix it, but I haven't got a test case, so no absolute guarantees are made.
Counted after 14 cpantesters.org failures of the form "Failed test 'first request honored asynchronously'". Added detailed error information to the failure. If it happens again, I'll be better prepared to fix it. Stopped relying on the order of first/second connection callbacks. Previously, if the second fired first, we'd shut down the TestServer too early. The first connection would then fail because. Meanwhile, always shutdown the connection manager (not just after the second response). Each session has its own, and they both need to go. Otherwise the program waits the obligatory connection pool timeout.
Sergei Kozunov came through with another great ticket. If two requests for the same host arrive, and the first one fails, the second one will time out in the queue. But only if no other requests are already pending. This change wakes up the queue when a connection fails, causing the subsequent request to be processed.
POE::Component::SSLify is non-blocking now. Add (and comment out) alpha-quality support for this feature. It needs work, and I hope someone who knows it better can help.
Sergei Kozunov pointed out that POE::Component::Resolver requests aren't canceled when POE::Component::Client::HTTP requests time out during DNS resolution. Cancel those requests!
Markus reports that this resolves rampant errors on solaris systems. I've verified that it doesn't break tests, which is bonus! Thanks!
Solaris (and probably other systems) don't have an /etc/services entry for http, so relying on the lookup fails here. Luckily the user also provides a port (which may or may not be symbolic). I'll use that for now and see where it gets me. Thank you, Michael Lackhoff, for reporting this bug.
Net::IP consumes a fair chunk of memory, Net::DNS stopped using it see https://rt.cpan.org/Public/Bug/Display.html?id=24525
I'm not sure why or which plugin is doing it, but making the NAME test one line is a serviceable workaround. Resolves rt.cpan.org ticket 69235, reported by Debian developer Gregor Hermann.
of POE::Component::Client::HTTP useragents from consuming all file descriptors. Reported by Karen_m and BinGOs on Freenode #perl, and fixed with the help of Karen's test case.
… Harris (CPAN ID RJH).
…an.org ticket 65069.