[00:32:42] --- Simon Wilkinson has left [02:49:54] --- geekosaur has left [02:54:47] --- geekosaur has become available [05:13:42] --- Born Fool has become available [05:21:38] --- Simon Wilkinson has become available [05:25:53] --- jaltman has left: Disconnected [05:36:00] --- mdionne has become available [05:40:02] --- Born Fool has left [06:09:01] --- Simon Wilkinson has left [07:11:17] --- deason has become available [07:30:56] --- abo has left [07:31:27] --- abo has become available [07:31:58] --- reuteras has left [07:32:11] --- reuteras has become available [07:51:36] --- Philip Jessel has become available [07:51:45] --- Philip Jessel has left [07:55:20] --- Bob Harold has become available [08:04:20] --- jaltman has become available [08:19:20] --- reuteras has left [09:04:52] --- Bob Harold has left [09:05:17] --- meffie has left [09:06:35] --- meffie has become available [09:12:05] --- Bob Harold has become available [09:25:04] http://github.com/blog/671-do-you-use-git-on-windows [09:36:04] --- matt has become available [09:38:44] is 127554 a security issue, or does it have private data in the ticket? [09:39:24] the former. [09:56:58] --- Russ has become available [10:06:44] ... that was fast. I didn't even get a chance to note that I didn't mention in the commit message that reclaim fails, but it's not clear that such mention would be needed. [10:08:04] i don't think it is. we need to look into that. but not right this second :) [10:19:05] --- Simon Wilkinson has become available [10:22:16] > right this second What, are you cutting 1.6 as we speak or something :-P [10:23:32] were cache bypass changes now out for some reason? [10:24:14] As far as I'm aware, then intention is still to enable at build time by default, but have it disabled through the pioctls. [10:24:21] (Same as disconnceted) [10:24:33] Gah. My typing is terrible today. [10:24:47] ok, no change then [10:29:58] actuallty, what i was doing right that second was folding the laptop and changing locations [10:33:30] --- kaj has become available [10:44:52] jaltman: Would it be necessary to have building and testing of strsafe.h to be automated on Windows (similar to make/make check on Linux)? As it is now, I've managed to build the tests individually in the command line on Windows, but no automation of any kind is done to build and run all tests. [10:45:11] This means that to test all functionality, each test file has to be built and run individually and the TAP output from each test has to be interpreted by the user. [10:45:18] It should be enough to test the test cases themselves against the Microsoft strsafe.h implementation though, since that only has to be done when the test cases change and not when the strsafe.h implementation does. [10:46:06] I think we (== OpenAFS) need to determine how automation on Windows is going to work before I can tell you that you are responsible for automating it. [10:47:26] and that would only be the case if we decided to use your strsafe.h on Windows. As you say, when using the Microsoft strsafe.h all that is important is that the tests be run once to determine a baseline behavior for use in determining the correct answers for the automated tests on Unix/Linux [10:48:05] And I guess it's a bit too much work to automate the testing when it's only required when additional test cases are added. [10:48:26] If you want to write a C wrapper that runs the tests on Windows against Microsoft strsafe.h and generate the answer key that would be committed to the repository, that would make a lot of sense. [10:51:07] For what it's worth, the only bogosity in reclaim that is obviously visible to me is that the refcount is -1 by the time of the freshly-added VOP_PRINT. [10:52:28] So you can now recycle arbitrarily? [10:52:54] I haven't pushed it very far because the VOP_PRINT becomes spew pretty fast. [10:53:34] That might be worth doing. I might look into that later. For now, I'll have to come up with what I should test on the extended functions. I've done a draft implementation of them, but no tests are written for them so far. [10:59:20] --- kaj has left [10:59:35] jaltman: By the way, what is the stance on documentation for the library? Can it be assumed that the user reads the Microsoft documentation or should I document the functionality myself so we don't have to depend on Microsoft keeping their documentation published? I've documented the basic functionality of Cat/Copy/Length so far and I could do the rest of the functions as well. [11:01:05] the Microsoft documentation is the reference. I believe you should document the build system, the test framework and any discrepancies from the reference implementation. [11:02:23] I would use doxygen comments in the source code to document your implementation. If necessary, documentation can be generated from those comments. [11:03:43] Doxygen is configured already and I've tried building documentation for the documented functions, so that should be no problem. [11:04:09] --- kaj has become available [11:07:59] Rather, since I have the log mail in front of me: +afs_vop_reclaim: afs_FlushVCache failed code 16 +tag afs, fid: 177.536994664.8.398, opens -1, writers 0 + states statd [11:45:40] --- mdionne has left [12:23:08] I'm on old bug patrol again. Just fixed one from 2003 [12:23:18] Wheee. [12:58:49] --- mdionne has become available [14:06:14] --- Bob Harold has left [14:13:34] Simon and Marc: am I misreading the diff? does the patch not in fact move the cognate code already taken from (forget what preceeded afs_fetchstore.c) from within the enclosing brace, to outside it? [14:15:53] Well, the problem is that the code, as stands doesn't deal with the case where cache bypass is the first to discover that a server doesn't have 64 bit support. [14:16:15] That is, if we get RXGEN_OPCODE, then we'll just return that to the user, rather than falling back to FetchData. [14:16:43] Right, it's incorrect, and the patch moves the check, already present, outside the enclosing brace? [14:16:52] Yes. [14:17:17] So I think Marc's commit message is correct to describe the effect of the patch. It perhaps doesn't describe exactly what the patch does, but it explains why it does it fine. [14:17:19] It's not like the original author of the check isn't Derrick, I just pasted it wrong. [14:17:43] Actually it doesn't. It implies that he took the cognate check from afs_fetchstore.c. [14:18:03] Well, I think it implies that he verified correctness against the code in afs_fetchstore.c [14:18:10] I won't belabor it further, but I think my request is reasonable. [14:18:17] (Which is what I did, when I wanted to check that it does the right thing too) [15:05:40] Hm, I kind of feel like changing 87 int 88 osi_VM_FlushVCache(struct vcache *avc, int *slept) 89 { ... 95 if (avc->opens) 96 return EBUSY; to check avc->opens > 0, at least for some interim. [15:23:48] --- matt has left [15:47:25] --- JSund has left [15:47:27] --- JSund has become available [15:47:34] --- ktdreyer has become available [15:56:25] Hi all, I was curious as to the magnitude of work required to finish the 2008 GSoC project for RW replication. I'm just looking for a rough estimate to see if we can work it into our budget. [15:57:21] ktdreyer - Derrick would be the best person to ask, and he's not around at the moment. [15:58:12] my guess would be "a lot of work", though [15:59:32] that is my impression as well, after reading over Noora's masters' thesis [15:59:34] I think Vishal got pretty far through it. [15:59:51] The major missing piece, I think, was a mechanism for telling the fileserver about read/write replicas [16:00:10] Bear in mind that OpenAFS's approach was not to do the method that Noora advocated, but instead one of the alternatives. [16:00:25] I'm trying to find the RT ticket with the code in it. If memory serves, there's a pretty good summary there. [16:00:38] --- meffie has left [16:01:17] cool, I would like to look it over [16:01:26] http://grand.central.org/rt/Ticket/Display.html?id=114116 [16:03:13] --- meffie has become available [16:07:23] --- deason has left [16:07:31] And I've found my notes from the GSoC debrief, too. Those indicate that the implementation needs "only" a recovery mechanism to get the replicas back in sync after an outage, or when you add a new slave, and a way of selecting, and tracking, the master copy of a volume at any given time. [16:07:55] Of course, the code in that ticket was written 2 years ago. A lot has changed since then - it almost certainly won't apply cleanly. [16:15:28] Also, I really ought to figure out why unmounting (-f) /afs, unload/reloading the kernel module, then starting afsd results in a kernel panic (mutex AFS global lock owned at rx_kcommon.c:998) It's kind of annoying. [17:53:43] --- kaj has left [17:56:29] --- kaj has become available [18:10:10] --- jhutz@jis.mit.edu/owl has become available [18:10:45] So, how long should I give on the afs3-stds charter consensus call? 2 weeks seems kind of short, but if we want to stick to the election timeline, we have to start no later than August 20, and I'd really prefer to start as close to August 1 as possible. [18:36:37] I have an annoucement ready to go, except for dropping in the expiration. [18:39:14] --- jaltman has left: Disconnected [18:45:47] --- jaltman has become available [18:48:14] Hm, printing three lines to the (serial) console per vnode while copying /usr/src into AFS may not have been the greatest idea ... [18:56:34] How fast is your serial console? [18:58:30] I ... think it's at 112500, but that's still pretty slow. [19:06:33] --- mdionne has left [20:21:44] Well, the copy finally finished. Time to turn off the print statement and buildworld! [20:21:48] --- ktdreyer has left: Lost connection [20:22:31] --- ktdreyer has become available [20:30:24] --- abo has left [20:31:17] --- abo has become available [20:38:19] --- abo has left [20:38:54] --- mho has left [20:39:07] --- kaj has left [20:40:36] --- abo has become available [20:40:37] --- kaj has become available [20:59:25] --- Born Fool has become available [21:10:27] Sorry, Derrick, you want me to panic in osi_VM_FlushVCache in that error case and get a backtrace? [21:11:09] or just try what i suggested and see if it fixes it, then backtrace if not? [21:12:21] Okay. [21:13:26] note that darwin does the same thing for similar reason (iirc anyway) [21:21:06] darwin has a flushpages as well ... [21:22:14] --- kaj has left [21:22:25] well, vop_pageout looks like the fbsd putpages, but has the FakeOpen/FakeClose (sorry, trying to do other htings too) [21:24:00] Er, darwin has a osi_FlushPages(tvc, vop_cred); above the FakeOpen [21:24:27] i was worried about the afs_write inside it [21:25:29] Right. [21:26:22] in fact i don't think the FlushPages there will matter. i hope [21:28:10] Okay. (rebooting for new kernel module) [21:34:01] FakeOpen is not enough; dumping (and going home) ... [21:44:58] ok [21:45:06] FakeOpen iincrements opens [22:24:58] --- reuteras has become available [22:34:03] Derrick: /afs/sipb.mit.edu/user/kaduk/freebsd/openafs/opens-2010-07-02/trace.txt for your pleasure [23:48:08] --- reuteras has left [23:58:05] --- Simon Wilkinson has left