[01:02:37] --- Russ has left: Disconnected [02:59:09] --- dev-zero@jabber.org has become available [03:39:51] --- SecureEndpoints has left: Replaced by new connection [03:49:26] --- dev-zero@jabber.org has left [03:51:01] --- dev-zero@jabber.org has become available [04:29:54] --- SecureEndpoints has become available [05:01:57] --- dev-zero@jabber.org has left [08:22:54] --- dev-zero@jabber.org has become available [08:45:25] --- dev-zero@jabber.org has left [08:46:32] --- dev-zero@jabber.org has become available [09:27:53] --- dwbotsch has left [09:37:38] --- Russ has become available [09:40:01] --- matt has become available [09:42:40] --- dwbotsch has become available [10:13:26] --- dev-zero@jabber.org has left [10:32:35] --- SecureEndpoints has left: Disconnected [10:43:17] --- edgester has become available [10:49:28] --- dev-zero@jabber.org has become available [10:52:55] --- SecureEndpoints has become available [10:57:14] hi jeff [11:15:24] hi edgester [11:25:53] SecureEndpoints: I verified bug 124069 under 1.5.5505, the private build from 12/27/2007. [11:26:18] you verified that the bug still exists or that it is fixed? [11:26:31] still exists [11:26:44] the 520 limit is in 1.5.5505 as well. [11:31:38] --- dev-zero@jabber.org has left [11:34:24] I haven't attempted to fix 124069 [11:34:44] k [11:41:59] --- dev-zero@jabber.org has become available [11:43:49] 1.5.5506 uploaded. fixes the panic in rxi_PrepareSendPacket() by removing a race condition in rxi_WriteProc() that would permit a freed or perhaps in use packet to be passed to rxi_PrepareSendPacket(). also changes the reported max component name length to 255 (even though the docs disagree.) [11:51:09] I show it as 255 now. Thanks! [11:52:55] hmmm, bug 124069 now fails with a 2 char filename, weird [11:53:57] what is the file name? [11:54:07] aa [11:57:31] feel free to smack me down on some of the bugs I bring up. Half the time, I don't know what is the correct behavior. [12:01:40] I gotta bail in a few minutes, but I'll be available intermittently after that. [12:02:14] I haven't looked at your file name test. [12:02:36] that's ok [12:02:39] there is no difference between "aa" in ascii and "aa" in utf-8 [12:02:50] I know [12:03:15] how are you writing in ascii and unicode? [12:03:32] The filename test writes a file using the normal non-unicode open, then uses createfilew to open the same file to reading [12:04:09] that doesn't matter to afs. it doesn't have a non-unicode interface [12:04:36] and the win32 ansi function is implemented as a shim over the unicode version [12:04:51] what is the error you are getting? [12:05:22] "error: print on closed filehandle" [12:06:19] what does ProcMon report? [12:06:43] I'll have to install it [12:07:08] ugh, sorry, I gotta go [12:07:10] http://technet.microsoft.com/en-us/sysinternals/bb896645.aspx [12:07:36] you can run it off the web site or out of \\afs\athena.mit.edu#user.jaltman\sysinternals\procmon.exe [12:41:06] --- matt has left [12:42:06] * edgester beats his head against the wall at his own stupidity! [12:43:49] SecureEndpoints: false alarm. The script couldn't create the file "aa" because there was already a directory named "aa" in the working directory. False alarm. After removing the folder, I get the usual 203/204 limit that I reported before the 255 length change [12:48:25] --- matt has become available [13:26:32] SecureEndpoints: would you please close bug 124069? I don't think it's an AFS bug. I think it's a limit in perl or somewhere because the limit seems to pertain to the max full pathname length, not the filename length. [13:47:37] --- matt has left [14:05:49] --- dev-zero@jabber.org has left [14:25:04] --- dev-zero@jabber.org has become available [14:33:00] --- matt has become available [14:36:27] edgester: done [14:37:10] SecureEndpoints: thanks. Sorry about the false alarm. [14:37:39] np [14:37:44] thanks for trying [14:37:52] you're welcome. [14:39:03] SecureEndpoints: any recent bugs that are screaming for test code? [14:45:16] not that I can think of. there have not been many bugs recently [14:46:04] I'd like to get some better test coverage of the disconnected code. I'm pretty sure there are reference counting bugs lurking there, I just can't provoke any at the moment. [14:46:06] ok, I guess that's good. no new bugs is good, but it might just be because of the holidays. [14:46:43] is there windows build with disconnected mode enabled? [14:46:52] there is no disconnected mode for windows [14:47:03] ah [14:47:08] Sorry. I'm thinking of Windows. [14:47:12] s/Windows/Unix/ [14:47:33] wait, huh? [14:47:35] Wasn't sure if you were only looking at Windows test code, or for Unix stuff too. [14:48:00] both [14:48:35] I try to write tests that work on linux + windows, but I have some windows-specific tests. [14:49:12] Well, if you fancied writing some stuff to put the disconnected ops stuff in the Unix CM through its paces, that would be great. [14:50:32] (Cache some files, disconnect, do some operations, reconnect, flush the local cache, then check that the state is what you expect) [14:50:38] Do we have tests for general file operations (ie, on Unix, and not counting Windows stress test suite)? [14:51:38] src/tests seems to contain some. I don't know when they were last run, though :) [14:53:57] asanka wrote a program which uses windows crypto operations to produce randomly generated and independently verifiable files. we use this to ensure that all blocks that are written by the test program are actually written to the file server. [14:54:53] we plan to modify the stress test suite to incorporate the algorithms in the generation of its test files. [14:55:11] What does the stress test use at the moment for its files? [14:55:27] That seems very useful. It might be interesting if someone tried making the stress test portable? [14:55:29] randomly generated 1mb files [14:55:58] why don't you just use one of the many pre-existing file system stress tests for *nix [14:56:02] My understanding was that the stress test was wedded to the SMB layer. [14:56:40] the stress test is designed to exercise operations implemented by the smb server. it is not smb specific [14:56:53] it is afterall a client app [14:58:49] Ah. Ok. For some reason I thought it spoke directly to the SMB server, rather than using normal file operations. Another misconception cleared up! [15:01:06] I think what you want on linux is something like samba's vfstest [15:01:41] I forget the name of the test app that apple recommended for macos x [15:02:13] --- dev-zero@jabber.org has left [15:19:45] --- dev-zero@jabber.org has become available [15:33:22] I'm sure that samba's vfstest is useful. I talks specifically about loading the vfs modules. There doesn't seem to be a working folder or anything like that. [15:34:44] Simon Wilkinson: are there linux packages available for 1.5 with disconnected support? [15:34:56] Yes. [15:34:59] But not officially. [15:35:24] ok, I'm fine with that. would you point me to the packages? [15:35:44] RPMs are at /afs/inf.ed.ac.uk/group/afsbuild/1.5.55/ [15:35:59] Let me know if you'd like a platform other than the 2 there (fedora 10 and rhel5) [15:36:09] I think Russ was also doing 1.5.x builds for Debian. [15:36:26] thanks! [15:36:29] (Those builds are --enable-disconnected and --enable-demand-attach-fs, so be careful what you do with them!) [15:36:42] understood [15:37:10] is fedora 10 the latest? [15:37:18] --- tkeiser@sinenomine.net/owl has become available [15:37:36] Yes. Latest Fedora, and there should be RPMs for the latest Fedora 10 kernel there, too. [15:37:50] Those builds are just spat out from the same system that does all of the RPM builds. [15:38:06] cool [15:38:23] In _theory_ it means that RPM support for 1.5 is ready to go. [15:38:27] hmm, debating how to proceed [15:38:54] install on my main test server, or set up another VM just for disconnected support [15:39:17] I'm running disconnected on my laptop without major issues currently. [15:39:19] my main test server has a hand-compiled openafs install, so a new VM might be best [15:39:31] Yes - it might, if you want to preserve that install. [15:39:38] ya [15:39:55] Just so you know - the following are possible issues with disconnected, which I'm not convinced we've fully resolved yet [15:40:15] are there ubuntu packages for 1.5? [15:40:28] Unsure. Russ would be the one to ask about that. [15:40:40] For full disclosure, possible disconnected issues of which I'm aware are: [15:41:15] *) I suspect that there are some issues with the vnode reference counting, such that we leak vcaches. This may cause an ooops on module shutdown [15:41:57] *) There may be an issue with us not locking properly when toggling between 'connected' and 'disconnected' state. Derrick found this the first time he tried it, but both Dragos and myself have been unable to reproduce it. [15:41:57] --- stevenjenkins has become available [15:42:23] *) If you reboot whilst disconnected any data you've written will be lost [15:42:37] vfstest indeed looks specific to library testing of samba3 vfs modules [15:42:38] *) There's no clean way of specifying which files should be cached. [15:43:57] --- dev-zero@jabber.org has left [15:44:07] --- dev-zero@jabber.org has become available [15:45:22] are there any docs that specify the behavior of disconnected mode? Things like "what happens to cached-read-only data when the written data would exceed the cache size?" [15:49:40] hi steven [15:52:11] --- matt has left [16:21:39] --- tkeiser@sinenomine.net/owl has left [17:16:47] --- edgester has left [18:23:41] --- dev-zero@jabber.org has left [18:42:19] --- tkeiser@sinenomine.net/owl has become available [19:47:17] --- edgester has become available [20:06:12] --- edgester has left [23:29:40] --- Russ has left: Disconnected [23:38:14] --- reuteras has become available