Dan Pritts wrote: >>> ...which are still happenning, although the timeouts are shorter. >>> More detail to follow. sigh. >> Depending on the symptoms the problems may have been fixed in 1.4.2. > > symptom is that the first time a user opened a file with a windows > 1.5.10 client, there was a noticeable delay, on the order of 10-30 > seconds. > > a few minutes later, I experienced a similar delay when i attempted > to ls the directory containing this same file. This was on a mac, > ppc, os 10.4.8, openafs 1.4.2fc3 (it appears from the /Library/Receipts > entry; is there a way to get this out of "fs" or one of the other > command line tools?). > > Windows firewalls were open on UDP port 7001. My mac firewall was > turned off (but it looks like it doesn't filter UDP anyway). > > The server is redhat 4, 32-bit x86, still openafs 1.4.1. > > The server FileLog contains this entry: > Mon Oct 23 14:15:24 2006 CB: RCallBackConnectBack failed for x.x.x.x:7001 > > This happened maybe 20-30 minutes *before* the delays i mentioned. The IP > is a third client, windows/1.5.10, which might have been looking at > something in the directory, or its parent directory (still in the same > volume), but he didn't think so. This client also has UDP 7001 open. > > I saw mention in your blog about something to do with mobile clients; > all the machines in question are laptops and regularly roam from a wired > to a wireless subnet. > > The log entry above is from right at the time he got up from his desk > at to go to the meeting where this all happened. the IP address logged > was his address on wireless. He's 99% sure he was *not* accessing this > directory from his desk. > > Also in the log from around then are several "WhoAreYou failed" messages > but the client IP address was that of one of my colleagues who wouldn't > have been accessing this part of AFS. > > danno
This is the bug fix that was demonstrated live at the Best Practice Workshop. Its was part of 1.4.2-beta1 and is in 1.4.2 final. The bug is that the file server does not destroy the old RX connection when the client migration is detected. Therefore, it must timeout first. Jeffrey Altman
smime.p7s
Description: S/MIME Cryptographic Signature
