[00:47:14] --- dev-zero@jabber.org has become available [00:47:24] --- dev-zero@jabber.org has left: offline [01:10:38] --- sxw has become available [01:15:46] --- sxw has left [01:20:15] --- dev-zero@jabber.org has become available [01:20:37] --- dev-zero@jabber.org has left: offline [01:21:06] --- Claudio Bisegni has become available [01:23:13] --- Claudio Bisegni has left [01:23:41] --- dragos.tatulea has become available [03:14:50] --- cclausen has left [05:29:40] --- Jeffrey Altman has left: Replaced by new connection [05:50:56] --- abo has become available [06:13:45] --- cclausen has become available [06:57:50] --- deason has become available [07:31:54] --- dragos.tatulea has left [07:33:28] --- dragos.tatulea has become available [07:37:19] --- dragos.tatulea has left [07:37:43] --- Derrick Brashear has left [07:40:24] --- dragos.tatulea has become available [07:51:14] --- dragos.tatulea has left [07:54:26] --- mmeffie has become available [08:20:39] --- reuteras has left [08:50:40] --- dragos.tatulea has become available [08:50:44] --- dragos.tatulea has left [08:59:26] --- Derrick Brashear has become available [09:10:28] --- dev-zero@jabber.org has become available [09:16:15] --- rra has become available [09:34:18] --- haba has become available [09:35:57] Simon: The newest patch you have listed is for OpenSSH 5.0p1 (http://www.sxw.org.uk/computing/patches/openssh.html) [09:36:40] Simon: Have you made something for 5.2p1 or does it "just work"? [09:43:06] derrick: am I correct in saying that the distinction between server lists is made while parsing the list, and setting up a filter of sorts? src/afs/afs_pioctl.c ~line 3617 (moved here from IRC) [09:43:10] that is, there aren't two separate lists, the only difference is in which servers are displayed, based on whether or not the vlonly flag is set [09:49:42] Simon: Ok, patch for 5.0 has at least 3 conflicts which are more than fuzz when applied to 5.2. [10:18:45] --- Jeffrey Altman has become available [10:20:31] jake: I'm not sure I understand your question. [10:20:31] Simon: Something like this compiles: /afs/stacken.kth.se/home/haba/public_html/openssh-5.2p1-gsskex-20090610.patch [10:21:30] jeff, it looks like all of the servers are stored in the same data structure, and if vlonly is set, then only the servers that match that mask are printed [10:23:10] DECL_PIOCTL(PGetSPrefs) does not print anything. It generates a list of servers to transmit to the caller of the pioctl. That is why I am confused by your question. [10:23:38] The list of servers that is returned is dependent upon the parameters that the caller of the pioctl specified [10:24:36] sorry, I was confusing PGetSPrefs and the wrapper around it in fs.c [10:25:05] fs.c does not wrap PGetSPrefs [10:25:42] not wrap. It makes the pioctl call, and prints the output [10:25:54] fs is a program that issues a pioctl call which is communicated to the openafs kernel module which in turn processes the request, send a response which in turn is received by fs and processed [10:26:03] yes [10:27:36] so I was wrong about printing in PGetSPrefs, my apologies. In PGetSPrefs though, am I correct that afs_srvAddrs[] is a list of all the servers, both file servers and vldb servers, and the ones given back to fs.c are dependent on the flag? [10:29:46] afs_srvAddrs[] is a hash table that includes all servers of all types. The conditional "if ((vlonly && isfs) || (!vlonly && !isfs))" is used to filter out either the file servers or the vldb servers. [10:29:50] --- dev-zero@jabber.org has left [10:30:26] right, that's what I thought. So then, in sorting, the hash table could be sorted and ranked. [10:30:53] > fs is a program that issues a pioctl call... So yes, "fs getserverprefs" can be said to wrap PGetSPrefs [10:32:57] afs_srvAddrs[] contains all of the servers that will need to be ranked. [10:34:42] and there's no reason the ranking should happen after the lists have been filtered [10:34:49] ? [10:35:58] please take a step backwards and explain what it is that you are attempting to accomplish at this point in time [10:36:53] what is your goal? [10:36:58] my working theory has been to do the ranking within the current infrastructure [10:37:28] so as to reduce changes in the way the code works. [10:37:48] all of your ranking must be performed within the cache manager without the use of outside triggers such as a pioctl. [10:38:10] there's an event loop you can use to trigger it, in afs_daemons.c [10:38:23] you need to collect data over time and periodically use that data to re-rank the servers [10:38:41] yes, I understand that [10:38:50] --- abo has left [10:39:26] --- abo has become available [10:39:30] --- deason has left [10:39:38] you have to rank all of the servers whether they are file or vldb servers [10:40:16] --- deason has become available [10:40:29] am I looking in the wrong spot for code? is there anything limiting the cache manager from interacting with afs_srvAddrs[] ? [10:41:06] afs_srvAddrs[] is a hash table that contains all of the servers regardless of type [10:41:33] the contents of afs_srvAddrs[] are constructed by the cache manager and managed by the cache manager [10:41:37] --- stevenjenkins has left [10:41:48] Jeff, you're not giving useful answers. [10:42:31] Jake is asking questions about the architecture of the cache manager, because he is modifying the cache manager. In such situations, "that belongs to the cache manager" is not a useful answer. [10:42:53] I think it would be useful for you to write up a description of how the cache manager currently discovers new servers, determines ranks, and uses those ranks to select which server is contacted. [10:43:56] the sense I'm getting is that I've been looking in entirely the wrong spot for this [10:45:05] --- stevenjenkins has become available [10:45:09] where is the code for the cache manager located? [10:45:23] you are reading it. /src/afs/ [10:46:12] alright, thanks [10:46:39] --- deason has left [10:46:50] --- deason has become available [10:47:14] --- haba has left [10:47:21] I gave you the assignment of writing a variation of the set server prefs pioctl as a warmup because it would point you in the right direction and build upon the work you were doing last Summer with kafs [10:48:02] --- haba has become available [10:48:33] That feeling that you are looking totally at the wrong place (when actually passing over the right spot serveral times) is not completely unknown :-) [10:48:53] I'm either confusing myself or missing something [10:49:11] let me go poke around, and I'll ask if I have questions [10:49:42] try to answer the question: "how does the cache manager select a server for a given connection?" [10:50:08] ok [10:51:40] --- stevenjenkins has left [10:55:13] --- stevenjenkins has become available [11:07:05] * haba forwarded "that iTunes" email from Jeff to local support staff at KTH. [11:07:42] --- dev-zero@jabber.org has become available [11:09:33] --- haba has left [11:20:47] --- asedeno has become available [11:29:17] --- dev-zero@jabber.org has left [11:38:48] Yeah, I sent it along to Stanford folks as well. [11:51:37] --- dev-zero@jabber.org has become available [11:55:15] --- sxw has become available [12:02:14] --- sxw has left [12:12:32] --- dev-zero@jabber.org has left [12:16:10] --- sxw has become available [12:28:33] --- haba has become available [12:48:05] --- sxw has left [12:50:11] --- dev-zero@jabber.org has become available [13:58:32] --- Jeffrey Altman has left [14:12:29] --- Jeffrey Altman has become available [14:13:55] Microsoft published WindowsXP-KB971421-x86. Must be requested direct from PSS. It will be a couple of weeks before the KB article is published to the web site. [14:17:03] what's this? [14:17:12] fix for smb redirector deadlock [14:18:22] So, it turns out if you add pages to the page cache, but don't stick them on the lruq, you lose. [14:21:30] Jeff: that is the hot fix I mentioned on Friday during my status report [14:21:53] and identical to the one that I wrote to openafs-info about a month ago for win2003 [14:21:59] and xp-64 [14:22:19] essentially, if you are using openafs on Windows, you want it [14:24:31] does anyone have a summary of the remaining known issues with DAFS? If not, I will pull them out of RT [14:24:47] I think RT is the currently know list. [14:25:20] cclausen/rini were suggesting that there may still be problems with volumes being created offline in 1.5 with non-DAFS, but I haven't heard any more about that. [14:25:40] I don't read -info; it's too voluminous. And I forget what I was doing, but I'm sad to say I wasn't paying very close attention during your report. Is this related to the problem you encountered earlier in the week? [14:26:44] --- stevenjenkins has left [14:26:47] the Windows hotfix corrects a deadlock in the XP mrxsmb.sys driver and adds new functionality that removes the 45 second timeout on smb requests sent to the AFS cache manager [14:27:10] Oh, that second thing sounds incredibly useful! [14:27:17] the problem experienced at stanford was entirely related to their WINS configuration [14:27:35] No, you had a problem with a machine at home deadlocking. [14:28:16] the same deadlock is also present on Vista/2008 and the new functionality is not present there. No one with an appropriate level of support contract has filed a request to make sure it is fixed there. [14:28:32] oh yes. the deadlock problem at home was this issue. [14:28:53] I've known about it for several months. Just had to wait until the fix got published. It is now installed on that machien. [14:29:13] --- cclausen has left [14:29:23] --- cclausen has become available [14:30:31] What level of support contract does one need? [14:30:36] --- stevenjenkins has become available [14:30:49] "big" [14:30:52] I assume :) [14:31:38] one that can escalate a bug to the Escalation Services Group [14:32:00] I have access to three levels of support contract: - random user - university - "hey larry" not a strict heirarchy [14:32:08] larry won't work [14:32:13] random user won't work [14:32:20] university might [14:32:58] can we just refer to the hotfix and say "fix this same thing on Vista and 2008" ? [14:32:59] Well, that's the one that's most complicated for me to use. I have to wait until I actually see my officemate. [14:33:06] Do you not have a Campus agreement? [14:33:10] we do not [14:33:14] its sad really [14:33:33] We have one of those, which pretty much seems to me to mean "We don't charge you market rates, you don't cause us any trouble" [14:36:48] --- Jeffrey Altman has left [14:37:00] there's a patch in RT for the volume creating offline with-DAFS issue, btw [15:10:15] screen -d -r main [15:10:19] feh [15:10:39] what's the problem? [15:10:41] * summatusmentis claps [16:06:48] Bah. When comparing the speed of a thing running in a VM, the page cache in the host OS is not your friend. [16:08:31] * haba ->sleep [17:36:27] --- edgester has become available [17:39:20] --- deason has left [17:39:52] --- deason has become available [17:52:18] what is the KB # or hotfix name for the SMB timeout fixes from microsoft? [18:00:05] KB971421-x86 [18:00:13] (From backlog from Jeff.) [18:03:06] thanks. [18:03:21] I thought that might be good to put in the newsletter. [18:03:38] the numbers weren't included in the 1.5.60 release notes. [18:05:40] Microsoft is sometimes weird about stuff like that, so I'd double-check with Jeff just to be positive that it's okay to put into something news-y as opposed to just public but not advertised widely like here. [18:05:54] It's probably fine, but I've said stuff that Microsoft has been unhappy with Stanford for us saying before. [18:06:42] well, I can't find the KB article anyways [18:06:53] It's not published yet -- you have to have a support contract to request it. [18:07:13] Should be public in a couple of weeks per Jeff. [18:08:47] can you get it with an MSDN subscription? [18:09:01] is that KB for XP or 2003? [18:10:38] usually anyone can call MSFT product support and mention specific KB article numbers to obtain a hostfix [18:10:52] there should not be a charge for actual bug fixes [18:11:34] but it should be available publicly in a few weeks? [18:11:45] Jeff said that you need to get it from PSS, but I'm basically just reiterating stuff that's probably better-explained in the logs of this conference from earlier today. [18:11:52] I have no personal knowledge. [18:15:19] ok. thanks. I rad the backlog and sent email to our windows guy. [18:17:08] any objection to adding a short blurb about the admin guide in the newsletter? [18:17:16] Definitely not here. [18:17:23] k [18:19:06] I forget who reviewed the chp1 patch. Jeff A did. Did you review it too, russ? [18:20:17] Yeah, I looked it over and did the commit. [18:20:31] ok, thanks [18:20:51] is there anything that you would like to see different in the newsletter this time? [18:21:01] do the CVS statistics add anything? [18:21:16] would lists of tickets make a difference? [18:21:20] I'm not sure. I like statistics in general, though. [18:21:30] Lists of resolved tickets would be awesome, although I don't know how easy that information is to get. [18:22:12] resolved since when? [18:22:18] The last newsletter. [18:22:27] hold on [18:22:55] --- abo has left [18:23:14] --- abo has become available [18:24:03] how about something like this: [18:24:04] 124579: resolved Apr 02, 2009 XDR array overflows in Unix cache manager 124580: resolved Apr 02, 2009 Linux panic in __follow_dir when ls'ing directory with active writer 124581: resolved Apr 02, 2009 stack overflow afs_GetDownD 124582: resolved Apr 02, 2009 patch: DAFS volume offline error [18:24:20] Yeah, that looks great. [18:24:21] I a script that generates these [18:24:24] cool [18:30:26] --- matt has become available [18:33:58] --- matt has left [18:54:30] --- edgester has left [18:54:34] --- mmeffie has left [20:40:28] --- Jeffrey Altman has become available [20:46:05] the KB number for the XP-x86 hot fix was not listed in the release notes because it wasn't assigned until last Thursday [20:48:42] I'm always torn about what to do regarding major OS vendors when they have bugs. If I advertise the fact that there is a bug someone always ends up getting annoyed and it becomes that much harder to get help the next time. [20:49:11] on the other hand it is important for the users to know about them [22:12:40] i like the "hey larry" level, except i barely know larry [22:14:37] --- deason has left [23:01:53] --- reuteras has become available [23:03:29] I see no open ticket in RT for a DAFS issue regarding volumes being created offline. There was an older ticket that described that symptom but it was resolved. [23:03:37] i see it. [23:03:56] 124492 [23:04:00] not resolved. [23:05:55] the description of the fix, which i haven't read yet, matches the fix i did for non-DAFS [23:06:03] its too late. brain is not functioning well. read it, but it didn't connect [23:06:10] the patch is trivial [23:06:51] i assume it's a one line patch. so was mine [23:07:21] i just had no hardware laying around at the time to try DAFS on. i suppose now i could make such a machine. [23:15:21] --- Jeffrey Altman has left [23:15:33] --- Jeffrey Altman has become available [23:32:31] --- kula has left