On Monday 26 February 2001 18:45, Oskar wrote: > Both these problems have been thoroughly discussed previously. > But Oskar, the world has changed...
> On Sun, Feb 25, 2001 at 08:42:13PM -0500, Kirk Reiser wrote: > > I have noticed a couple of other potential problems with cvs > > currently. I believe this one has been discussed here recently but > > still exists after the past few days of bug fixes. If the server gets > > swamp with insert requests. It will affectly quit accepting inserts. > > That is they all start failing. I am not sure how many you need to > > send the server at once to cause the condition, but gj InsertMp3s will > > do it nicely with a -simRequest of 5 and -retries of 3 on about a > > dozen files. I duplicated this behaviour twice in a row. The only > > way out of the situation is to kill off the server and restart it. > > I do inserting on a regular basis for my webpage. I have also seen these symptoms, though the problem seems much less acute in the last couple of days. Maybe the network is getting better to the point where it is harder to blow out all routes or saturate nodes upstream? Another related problem is that even localhost connections are refused once the inbound connection limit is reached. It seems like there should be an exception for inbound connections from localhost, just like localhost connections are not throttled. My node often hits the default limit of 50 inbound connection threads. Increasing the number of threads isn't the right thing to do, since my outgoing bandwidth is already saturated most of the time. I know that I could set up a separate transient node just for local use, but that seems like a ridiculous burden to impose on joe clueless end user. Keep in mind that this problem looks really ugly to the end user. Their node hits the connection limit and then fproxy starts to fail intermittenly with no explanation of why. I have already put a quick hack in my local tree that accepts inbound localhost connections on unpooled threads when the thread pool is empty. Should it commit this to the head branch? > > Another thing I've noticed and it may have to do with the network > > being munged with the 0.3.7 servers is that requests, through fproxy > > anyway are extremely unreliable. I usually insert and request with > > htl of 30 which I would suspect should give a fairly wide distribution > > yet I cannot receive files that are larger than a few 'k' usually, > > inserted on one server and requested on another. It is possible that > > htl 30 doesn't bring these two servers into close aproximation. > > However, they are about ten hops from each other on the actual > > network, That seems rather odd though. I just figured I should > > mention it to see if anyone else has noticed this behaviour. > > I guess that there are a lot of things that may be going wrong. One problem I see quite often when I look over my log files is DNV exceptions. Is it understood why there are so many? Or is this a bug. --gj -- Web page inside Freenet: freenet:MSK at SSK@enI8YFo3gj8UVh-Au0HpKMftf6QQAgE/homepage// _______________________________________________ Devl mailing list Devl at freenetproject.org http://www.uprizer.com/mailman/listinfo/devl
