On Mon, 2005-07-25 at 23:55, mark burdett wrote:> For 100% reliability I ended up removing dir.xiph.org from my > list of YP directories. My guess is it was taking too long to > respond to requests, which could be a problem if an icecast > server has a large number of mounts. But it could be a > problem unique to me (I'd be curious to know if anyone else > has had trouble).The directory handler itself serialises the updates so the number of mounts shouldn't matter directly however if long timeouts trigger with many stream details then the touch updates may not get around quickly enough.> the dir.xiph.org admins might want to make sure the db is well > tuned (e.g. query caching, etc. in my.cnf) so it can perhaps > respond to touches more quickly..It will probably be looked at over the next few days, we reached over 900 on it today, so it has been steadily increasing. karl.
> The directory handler itself serialises the updates so the number of > mounts shouldn't matter directly however if long timeouts trigger with > many stream details then the touch updates may not get around quickly > enough.if they are serialized and the posts happen one after the other for each mount, and each one takes 8 seconds, then with 30 mounts that's 4 minutes. that might be too long; the YP might expect updates every 2 minutes. could updates for multiple YP directories happen in parallel, so if one YP directory is slow, that won't prevent other YPs from being touched? --mark
On Tue, 2005-07-26 at 00:59, mark burdett wrote:> > The directory handler itself serialises the updates so the number of > > mounts shouldn't matter directly however if long timeouts trigger with > > many stream details then the touch updates may not get around quickly > > enough. > > if they are serialized and the posts happen one after the > other for each mount, and each one takes 8 seconds, then with > 30 mounts that's 4 minutes. that might be too long; the YP > might expect updates every 2 minutes.I believe the removal is 5 mins currently, but yes the current access is becoming 8 or more seconds, as I said that needs looking into, I don't think the setup is optimal.> could updates for multiple YP directories happen in parallel, > so if one YP directory is slow, that won't prevent other YPs > from being touched?we could serialise on just each directory server, that would isolate a problem server. something to look into later, so can you report this on trac.xiph.org so that it doesn't get forgotten? karl.