I use two levels of controls, which seems working smoothly. 1. when exception is thrown, check if it's the type that results from connection dropping, i.e. IOException or HTTP code=4xx or error message; reconnect only if this is true. there may be many other types of exceptions, but don't reconnect in those cases. Normally, I only notice a couple of disconnection a day. 2. set a max number of reconnection; when reaching the max, don't auto-reconnect, but requires a manual reconnect instead. This way, if the api server goes wrong, you won't bombard the server.
-aj On Tue, Jun 23, 2009 at 3:49 PM, danielo <ohboyot...@gmail.com> wrote: > > I had a similar question. I think you've mostly answered it, but I > want to be clear so as to avoid harassing the API. > > I'm developing a client to connect to the streaming API (nothing fancy > at the moment; just spritzer), and of course, I'm bungling it up > regularly. I'll hack a bunch, try it, watch it break, shut it down, > and hack some more. Is there a practical limit at which point I should > apply the human throttle-back? Or is there no realistic human limit at > which I risk a ban from the streaming service? I imagine that if a 15- > second wait period is sufficient to avoid bad things, the more likely > 1-to-2-minute wait between my attempts will be fine. I ask, > nonetheless, as my repeated requests will persist for the duration of > my work, whereas a running client would (hopefully) snag a valid > connection after some time and stop "spamming" at that point. > > Thanks! > > On Jun 14, 8:14 pm, John Kalucki <jkalu...@gmail.com> wrote: > > AJ, > > > > If you had a validconnectionand theconnectiondrops, reconnect > > immediately. This is encouraged! > > > > If you attempt aconnectionand get a TCP or IP level error, back off > > linearly, but cap the backoff to something fairly short. Perhaps start > > at 20 milliseconds, double, and cap at 15 seconds. There's probably a > > transitory network problem and it will probably clear up quickly. > > > > If you get a HTTP error (4XX), backoff linearly, but cap the backoff > > at something longer, perhaps start at 250 milliseconds, double, and > > cap at 120 seconds. Whatever has caused the issue isn't going away > > anytime soon. There's not much point in polling any faster and you are > > just more likely to run afoul of some rate limit. > > > > The service is fairly lenient. You aren't going to get banned for a > > few dozen bungled connections here and there. But, if you do anything > > in a while loop that also doesn't have a sleep, you'll eventually get > > the hatchet for some small number of minutes. If you get the hatchet > > repeatedly, you'll be cut off for an indeterminate period of time. > > > > There are four main reasons to have yourconnectionclosed: > > * Duplicate clients logins (earlier connections terminated) > > * Hosebird server restarts (code deploys) > > * Laggingconnectiongetting thrown off (client too slow, or > > insufficient bandwidth) > > * General Twitter network maintenance (Load balancer restarts, network > > reconfigurations, other very very rare events) > > > > We plan to have enough spare capacity on the surviving servers to > > absorb the load from server restarts. You must ensure that your client > > is fast enough and that you have sufficient bandwidth and a stable > > enoughconnectionto consume your stream. I usually see connections > > that survive for a few days before mysteriously being dropped. Just > > reconnect in these cases. > > > > -John Kalucki > > Services, Twitter Inc. > > > > On Jun 14, 3:31 pm, AJ <cano...@gmail.com> wrote: > > > > > Thestreamingapiis great, but it sometimes closes theconnectionfor > > > whatever reason. my realtime system must figure out when to reconnect > > > automatically. the auto-reconnection can't blindly request a > > >connectionwhenever it is not connected, otherwise it will floor the > > >apiand may cause theapito ban or refuse the user's request. it's > > > bad to bombard theapiserver with repeatedconnectionrequests. > > > Could theapiteam recommend some best practice for dealing with auto- > > > reconnection? > > > > > maybe certain error code or error message can indicate the cause of > > > droppingconnectionand wait time for nextconnectionrequest. I just > > > a long list of exceptions fromstreamingapias a result of repeated > > >connection, and the different messages are: > > > > > twitter4j.TwitterException: Address already in use: connect > > > twitter4j.TwitterException: Authentication credentials were missing or > > > incorrect. > > > twitter4j.TwitterException:Connectionrefused: connect > > > twitter4j.TwitterException: No route to host: connect > > > twitter4j.TwitterException: Stream closed. > > > twitter4j.TwitterException: The request is understood, but it has been > > > refused. An accompanying error message will explain why. > > > twitter4j.TwitterException: connect timed out > > > > > How to prevent such situation of repeated connections requests? > > > > > thanks, > > > aj > -- AJ Chen, PhD Co-Chair, Semantic Web SIG, sdforum.org http://web2express.org Palo Alto, CA