On Tue, 2008-03-25 at 14:44 -0700, Ankur Shah wrote: > Thanks oleg, > > Do you see any issues in the approach that i was thinking about (and have > implemented already) other than the fact that its not very clean?
Ankur, The problem is the connection manager can only catch connect timeout exceptions, but you may also want to retry requests against the secondary host in case of other I/O failures as well (for instance, when the primary host is able to accept connections but is unable to process requests) > Could you > elaborate a bit more on what you mean by creating service on top of > httpclient, (i understand what it'll do but just didn't know what/how should > i register that with httpclient)? > There's no need to register anything. Just implement whatever fail-over logic is appropriate for you application on top of HttpClient Oleg > On Tue, Mar 25, 2008 at 1:10 PM, Oleg Kalnichevski <[EMAIL PROTECTED]> wrote: > > > > > On Mon, 2008-03-24 at 15:46 -0700, Ankur Shah wrote: > > > Hi Oleg, > > > > > > Thanks for the response. The server that i am running (acting as a > > client in > > > this particular case) is working with multiple servers (hosting > > different > > > webservices) with each having a primary and a failover pair. From this > > > perspective the thought was that rather than having different classes > > (all > > > having their own business logic around invoking the service) handle the > > > connection failed attempt, somehow have some central code that attempts > > to > > > connect to the primary and fails over to the backup. I also wanted to > > make > > > sure that in case of a connection failure with primary, a new pool is > > > established against the backup server. > > > > > > From this perspective, an after thought that I had was this. I was going > > to > > > extend the MultiThreadedHttpConnectionManager and have a failover > > capable > > > connection manager. In there, i was thinking about exposing a method (or > > a > > > new constructor) that allows backup host and ports to be added. This > > manager > > > will cache a map of primary and backup server. In the overriden > > > getConnectionWithTimeout method, i'd have super return me a connection > > that > > > i'll check and make sure that is valid and if not, i'll go to my cache, > > get > > > the backup host config and open up a connection to that. > > > > > > Let me know if you think this would be a good idea or not. > > > > > > Ankur > > > > > > > Ankur > > > > My recommendation would be to implement the fail-over support as a > > service on top of HttpClient > > > > (1) execute a request against the primary host > > (2) if an exception is thrown, see whether it can be treated as > > recoverable > > (3) if so, mark the primary host as unavailable for a certain period > > time and retry the request against the secondary host > > (4) when the quarantine time elapses, mark the primary host as > > potentially available > > (5) repeat > > > > Hope this helps > > > > Oleg > > > > > > > On Mon, Mar 24, 2008 at 12:47 PM, Oleg Kalnichevski <[EMAIL PROTECTED]> > > > wrote: > > > > > > > > > > > On Thu, 2008-03-20 at 22:05 -0700, Ankur Shah wrote: > > > > > I am using httpclient to connect to a remote server. Now for each > > > > primary > > > > > server there's also a backup failover server that i must talk to if > > > > > connection attempts with the primary server fails. To achieve > > > > > this what i'd ideally like to do is have HttpClient use a custom > > factory > > > > > that tries to create a socket connection with the primary server and > > if > > > > that > > > > > fails (or times out), it'll go about talking to backup server (and > > > > retry a > > > > > few times before giving up altogether). > > > > > > > > > > Now, searching on the web, i did find references to a > > > > > MultiHomeProtocolSocketFactory (that extends DefaultSocketFactory) > > that > > > > is > > > > > presumably in apache contrib package. But what that does is that in > > the > > > > > overriden > > > > > createSocket method, it takes host as a param (among other thing) > > and > > > > then > > > > > simply does a InetAddress.getAllByName which is not what i want. > > From > > > > what i > > > > > understand this caters to a DNS based failover which is not the case > > > > here. > > > > > > > > > > What i'd ideally like to do is create a factory and have a > > createSocket > > > > > method that takes primary and secondary host and port as args and > > then > > > > tries > > > > > both. I am not sure though how can i make HttpConnection class > > > > > invoke my method. > > > > > > > > Ankur, > > > > > > > > What's the point of trying to implement the failover support on the > > > > ProtocolSocketFactory level? What's wrong with just retrying failed > > > > requests against a secondary server in case the primary one is > > > > unavailable? > > > > > > > > Oleg > > > > > > > > > > > > --------------------------------------------------------------------- > > > > To unsubscribe, e-mail: [EMAIL PROTECTED] > > > > For additional commands, e-mail: [EMAIL PROTECTED] > > > > > > > > > > > > > > --------------------------------------------------------------------- > > To unsubscribe, e-mail: [EMAIL PROTECTED] > > For additional commands, e-mail: [EMAIL PROTECTED] > > > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
