Re: [Server-devel] Collaboration problem again 0.5.1
On Tue, Feb 17, 2009 at 5:46 PM, Martin Langhoff martin.langh...@gmail.com wrote: On Tue, Feb 17, 2009 at 1:08 PM, David Leeming leem...@pipolfastaem.gov.sb wrote: 9-0.5.1 ejabberd-xs-2.0.1-12.fc9.olpc.i386 Hmmm, the versions are correct -- I can't test this right now in a clean environment, as I have bits and pieces of different versions everywhere because I am working on something else ATM. Grrr. Double gr... I made some tests, and they seemed to fail. Not watertight tests as I don't have a clean env, so please please please corroborate. I do have some notes, and something to play with... - When you restart everything, and generally everytime you check whether the laptops see eachother, check olpc-netstatus on the laptops involved, and 'ejabberdctl connected-users' on the server. The laptops are now (8.2.x) very good at switching quickly to salut if they don't see the ejabberd server. And then they switch quickly back to gabble. Magic and quick -- good for users, but very misleading for testing. (Just between us, I suspect that the above issue may have made me misreport success with the patch I applied a few weeks ago. Right after restarting everything, the XOs are sometimes ostensibly on the 'School Server Mesh' and yet using salut. Happens if the server booted up late, or if they didn't see it immediately after dhcp.) - Please make independent tests setting the 'Online' group membership to '@all@' instead of '@online@'. If this works, it is a valid and usable fallback unless your schools are truly huge. - I built a new RPM that has ejabberd 2.0.3 (instructions below) - please test it, trying @online@ and @all@ with it. As of now, if you say yum --enable-repo=olpcxs-testing install ejabberd-xs it should bring version 2.0.3-1-olpc . In my limited testing it works with @all@ and fails with @online@ -- and I think @online@ is thoroughly broken on it. You can also grab it from http://xs-dev.laptop.org/xsrepos/testing/olpc/9/i386/ejabberd-xs-2.0.3-1.fc9.olpc.i386.rpm and install it with rpm -Uvh filename May a thousand test reports come... cheers, m -- martin.langh...@gmail.com mar...@laptop.org -- School Server Architect - ask interesting questions - don't get distracted with shiny stuff - working code first - http://wiki.laptop.org/go/User:Martinlanghoff ___ Server-devel mailing list server-de...@lists.laptop.org http://lists.laptop.org/listinfo/server-devel
Re: [Server-devel] Collaboration problem again 0.5.1
I'm re-CC'ing server-devel. On Tue, Feb 17, 2009 at 11:04 PM, David Leeming leem...@pipolfastaem.gov.sb wrote: Initial results using @all@ ... starting from scratch, deleting the ejabberd data, recreating the account with ejabberdctl etc and deregistering and re-regsitering XOs etc With which rpm? Seems to have solved the problem. I have rebooted the system many times and it is still immediately picking up all the XOs collaborating. Will try the new rpm For now, what is the limit on XOs simultaneously connected on a single server, using @all@ or otherwise? Depends on other aspects of the infra, mainly the APs. 40 was tested ok, and some later tests handled 60. Using @all@ instead of @online@ should not change things too much, unless you have lots more than that, in which case you have other problems anyway. In other words: @all@ is a valid workaround in all the scenarios I can think of. cheers, m -- martin.langh...@gmail.com mar...@laptop.org -- School Server Architect - ask interesting questions - don't get distracted with shiny stuff - working code first - http://wiki.laptop.org/go/User:Martinlanghoff ___ Server-devel mailing list server-de...@lists.laptop.org http://lists.laptop.org/listinfo/server-devel
Re: [Server-devel] Collaboration problem again 0.5.1
On Tue, Feb 17, 2009 at 13:44, Martin Langhoff martin.langh...@gmail.com wrote: In other words: @all@ is a valid workaround in all the scenarios I can think of. @all@ should have the same effect in the UI as @online@, but with a potential difference in the performance: @all@ is the conventional shared roster and would send you presence for everyone registered on the server, whereas @online@ is a bit more magic and only sends you presence for people who are actually on line at the time. Perhaps some of its magic smoke escaped. @online@ is more useful for servers where only a (small) percentage of the registered users are on line at a time. Regards Morgan ___ Server-devel mailing list server-de...@lists.laptop.org http://lists.laptop.org/listinfo/server-devel
Re: [Server-devel] Collaboration problem again 0.5.1
On Wed, Feb 18, 2009 at 12:53 AM, Morgan Collett morgan.coll...@gmail.com wrote: On Tue, Feb 17, 2009 at 13:44, Martin Langhoff martin.langh...@gmail.com wrote: In other words: @all@ is a valid workaround in all the scenarios I can think of. @all@ should have the same effect in the UI as @online@, but with a potential difference in the performance: @all@ is the conventional shared roster and would send you presence for everyone registered on the server, whereas @online@ is a bit more magic and only sends you presence for people who are actually on line at the time. Perhaps some of its magic smoke escaped. @online@ is more useful for servers where only a (small) percentage of the registered users are on line at a time. I'm 100% in agreement. My simple (simplistic?) analysis of the situation for a deployment right now is that the payload of a roster fetch is still not that large, and rosters are seldom fetched (this last bit comes learned from looking at ejabberd verbose logs till my eyes bleed). So yes, using @all@ isn't best in network efficiency but I suspect it's negligible for small deployments. And large deployments have many other probs... which I'm working on :-) cheers, m -- martin.langh...@gmail.com mar...@laptop.org -- School Server Architect - ask interesting questions - don't get distracted with shiny stuff - working code first - http://wiki.laptop.org/go/User:Martinlanghoff ___ Server-devel mailing list server-de...@lists.laptop.org http://lists.laptop.org/listinfo/server-devel
Re: [Server-devel] Collaboration problem again 0.5.1
Good Martin, how many with 0.6 cater for? When is the anticipated release date/month/quarter? David EVEN WITH MULTIPLE APs, THE LIMIT IS 60? WE HAVE THREE SCHOOLS WITH 100 PUPILS AND TEACHERS EACH. THEY WOULD BE UNLIKLEY TO HAVE ALL 100 CONNECTED 100% OF THE TIME. DO YOU THINK WITH MULTIPLE APs WE COULD GET AWAY WITH ONLY ONE XS? in that case, the network will hold but the network view stops working well. Working on this for 0.6. ___ Server-devel mailing list server-de...@lists.laptop.org http://lists.laptop.org/listinfo/server-devel
Re: [Server-devel] Collaboration problem again 0.5.1
On Wed, Feb 18, 2009 at 10:20 AM, David Leeming leem...@pipolfastaem.gov.sb wrote: Good Martin, how many with 0.6 cater for? When is the anticipated release date/month/quarter? With 0.6 you can separate the groups that see each other, so the network view on the XO doesn't get overwhelmed. Reuben wants it to work for 3K users on high-end HW. Target: end of march. m -- martin.langh...@gmail.com mar...@laptop.org -- School Server Architect - ask interesting questions - don't get distracted with shiny stuff - working code first - http://wiki.laptop.org/go/User:Martinlanghoff ___ Server-devel mailing list server-de...@lists.laptop.org http://lists.laptop.org/listinfo/server-devel
Re: [Server-devel] Collaboration problem again 0.5.1
On Tue, Feb 17, 2009 at 1:08 PM, David Leeming leem...@pipolfastaem.gov.sb wrote: 9-0.5.1 ejabberd-xs-2.0.1-12.fc9.olpc.i386 Hmmm, the versions are correct -- I can't test this right now in a clean environment, as I have bits and pieces of different versions everywhere because I am working on something else ATM. Posting the logs from /var/log/ejabberd/ would be great. Enable verbose logging (see the /etc/ejabberd/ejabberd-xs.conf file), delete the user registration, the logs, restart ejabberd and do the whole dance of first time success, second time failure. At that point the log should have good info. On the run up to the release, I could not get this problem to happen in my testing in NZ, and neither could Reuben at 1CC. Nobody else seems to have tried any RC. We discussed various times setting up a phonecall with David but it didn't happen. Sigh. If anyone can help a bit, much appreciated. Trying to repro (2 XOs and 1 XS are enough) would be fantastic. I will try to get into diagnosing this but it will take me a while before I am free. m -- martin.langh...@gmail.com mar...@laptop.org -- School Server Architect - ask interesting questions - don't get distracted with shiny stuff - working code first - http://wiki.laptop.org/go/User:Martinlanghoff ___ Server-devel mailing list server-de...@lists.laptop.org http://lists.laptop.org/listinfo/server-devel
Re: [Server-devel] Collaboration problem again 0.5.1
2009/2/17 David Leeming leem...@pipolfastaem.gov.sb: I thought this was fixed in 0.5.1 What do the following commands say? cat /etc/fedora-release rpm -qa ejabberd-xs cheers, m -- martin.langh...@gmail.com mar...@laptop.org -- School Server Architect - ask interesting questions - don't get distracted with shiny stuff - working code first - http://wiki.laptop.org/go/User:Martinlanghoff ___ Server-devel mailing list Server-devel@lists.laptop.org http://lists.laptop.org/listinfo/server-devel