POE::Component::Client::HTTP, Keepalive, FIN_WAIT

nuclon has asked for the wisdom of the Perl Monks concerning the following question:

Replies are listed 'Best First'.
Re: POE::Component::Client::HTTP, Keepalive, FIN_WAIT_2 by rcaputo (Chaplain) on Nov 24, 2008 at 16:53 UTC
FIN_WAIT_2 happens when the local end of a socket is closed and the operating system is waiting for a signal that the remote end has closed their side. In your case, enough connections haven't that the operating system's socket table has filled, and new sockets can't be allocated. I have a possible solution, but I don't have anything to test it against. If you could provide a test case that reliably triggers the problem, I can debug it quickly and turn around a better answer. Meanwhile I have taken a guess at a solution, but you'll have to test it for me. If you can, please edit Client/Keepalive.pm (not Connection/Keepalive.pm). There should be a comment around line 582: `# Build a connection object around the socket.` I need you to add these three lines before that comment (wherever it may be): `use Socket qw(SOL_SOCKET SO_LINGER); setsockopt($socket, SOL_SOCKET, SO_LINGER, pack("sll",1,0,0)) or die +( "setsockopt: $!" + );` [download] When it comes time to close that socket, it should not linger in FIN_WAIT_2. If that works for you, keep it free of charge. Also, let me know so I can include it in a future release. Thank you. `-- Rocco Caputo - http://poe.perl.org/`	[reply] [d/l] [select]
Re^2: POE::Component::Client::HTTP, Keepalive, FIN_WAIT_2 by nuclon (Acolyte) on Nov 25, 2008 at 08:04 UTC
it seems it helped but now I have another behaviour :) Script works well some time. Then I start to get "connect error 60: Operation timed out", see a lot of "TIME_WAIT" connections in netstat (and no other connections at all) in this moment. after few minutes all TIME_WAIT connections dissapear and I see empty output of netstat, but script still log "Operation timed out". I've turned on DEBUG in Client::Keepalive and got a lot of following messages at the time when netstat output is empty: `ALLOCATE: enqueuing request for http:gdata.youtube.com:80 at /home/ser +vice/perl/lib/site_perl/5.8.8/POE/Component/Client/Keepalive.pm line +377. ... CON: request from session 5 for address 74.125.95.102 timed out at /ho +me/service/perl/lib/site_perl/5.8.8/POE/Component/Client/Keepalive.pm + line 507.` [download]	[reply] [d/l]
Re^3: POE::Component::Client::HTTP, Keepalive, FIN_WAIT_2 by rcaputo (Chaplain) on Nov 25, 2008 at 17:10 UTC
I'm not sure I can debug this one from a distance. Can you provide a minimal test case to reproduce the problem? Ideally it would be something I can run without installing proprietary code or data. By the way, I released new versions of POE::Component::Client::Keepalive and POE::Component::Client::HTTP with a modified version of the SO_LINGER code and some other bug fixes. `-- Rocco Caputo - http://poe.perl.org/`	[reply]
Re: POE::Component::Client::HTTP, Keepalive, FIN_WAIT_2 by EvanCarroll (Chaplain) on Nov 26, 2008 at 05:09 UTC
You can apparently change the time that a socket can remain in FIN_WAIT2 with a simple echo to /proc/sys/net/ipv4/tcp_fin_timeout. I'd be interested to know how POE handles a kernel reap of the socket. Evan Carroll I hack for the ladies. www.EvanCarroll.com	[reply]
Re^2: POE::Component::Client::HTTP, Keepalive, FIN_WAIT_2 by rcaputo (Chaplain) on Nov 26, 2008 at 05:40 UTC
I'm not sure I understand your question. Sockets in FIN_WAIT_2 have been closed at the application level. This often happens as a consequence of an IO::Handle object (or an instance of a subclass) being DESTROYed. Please clarify: What do you expect an application to do with a closed socket? How do you expect an application to call methods on a DESTROYed IO::Handle object? What kernel syscalls do you expect to work with closed file descriptors? Thank you. `-- Rocco Caputo - http://poe.perl.org/`	[reply]