[00:39:12] --- dev-zero@jabber.org has become available [00:39:16] --- dev-zero@jabber.org has left: offline [06:36:04] --- Simon Wilkinson has become available [06:36:31] --- Simon Wilkinson has left [06:59:16] --- Russ has become available [07:15:09] --- deason has become available [07:47:15] --- Derrick Brashear has left [08:09:43] --- Russ has left: Disconnected [08:22:04] --- Russ has become available [08:27:35] --- Derrick Brashear has become available [08:33:17] --- reuteras has left [09:40:14] --- Simon Wilkinson has become available [11:03:20] --- mmeffie has become available [11:54:34] --- dev-zero@jabber.org has become available [11:54:42] --- dev-zero@jabber.org has left: offline [11:56:38] AFs does seem to be a bit broken on this network. [11:56:50] It doesn't not work for me, but I keep losing fileservers. [11:56:57] no problems here [11:57:10] Have you checked under the sofa? [11:57:35] works for me too [11:59:15] RT is being glacial today. [11:59:55] What did you do to it? It was working fine for me a moment ago. [12:00:22] Well I asked it to tell me about a bug. Is that not recommended? [12:00:59] It's disrecommended if the bug is particularly traumatic. [12:01:04] :-) [12:03:16] AFS is working for me. [12:03:43] Maybe the cache manager is coffee-intolerant :) [12:04:30] Oh, Simon, we mentioned briefly in the middle of some other discussion, but I never followed up. Do you know of a bug tracking system that uses Git to track bug state similar to Bugs Everywhere but that's actively developed? [12:04:59] No. Not currently. [12:05:27] Shawn has a plan to make Gerrit use git for its backend storage, but that itsn't really a fully flown bug tracking system. [12:05:35] And isn't started yet, let alone done. [12:05:50] hm. i wonder why bozo-fs-create-unused-ptr-20090319 seems to have broken hpux's cc. [12:06:15] * Russ would really like to find a bug tracking system that I can use for my stuff. The best candidate so far seems to be redmine. [12:06:52] But it's not packaged for Debian, and I have low tolerance for Ruby package insanity. [12:07:10] Derrick: Are we mismatching on a prototype somewhere? [12:07:16] What's the error message? [12:07:44] the patch added a prototype change and a function args change at the same time [12:08:06] Yeh. I see that. There isn't an HPUX specific prototype lurking somewhere is there? Or some other HPUX specific usage? [12:08:24] shouldn't be. [12:08:56] to be fair, before that change, fsbnodeops didn't build on hpux, so something else may have broken it. unless "unused" is a keyword to hpux cc [12:09:01] --- deason has left [12:09:02] --- stevenjenkins has left [12:09:19] --- deason has become available [12:09:24] I guess it could be. But that would be evil. [12:09:40] ... [12:12:44] --- stevenjenkins has become available [13:00:02] --- Russ has left: Disconnected [13:07:28] --- mmeffie has left [13:34:41] --- Russ has become available [13:52:07] --- cclausen has become available [13:58:17] --- mmeffie has become available [14:02:24] --- dev-zero@jabber.org has become available [14:03:07] --- dev-zero@jabber.org has left: offline [14:24:57] --- matt has become available [14:25:18] kula: are you interested in volser also? [14:29:06] matt: do you mean getting that info directly from the volser and not having do a dump first? yeah, i am, actually, that idea has passed through my mind. [15:02:28] I was thinking that knowing a lot about dumps would help doing with the (short version of) the back end of per-file acls. [15:02:56] I'll be very interested to see what Marc Dionne has done there. [15:03:20] Marc Dionne seems to have the client experimentally ready to deal with it. Since Marc hasn't remarked, I'm unsure if he has...? [15:04:14] Marc said "a working prototype" [15:04:19] I guess we'll see what that means ... [15:04:57] Personally, I think the first thing to do with volume dumps is to design a generic up and downgrade mechanism. [15:05:12] Once we've done that, we can start to add new stuff. There's a fair bit of new stuff I'd like to see. [15:05:25] But we'll need to involve people like Teradactyl in that discussion. [15:05:52] They know a lot about it. Happy to see them jump in... [15:06:24] Yeh. Anything we do has to not break their backups :) That is, there needs to be a way of saying to a new fileserver "Give me an old format volume dump" [15:06:54] ahh. i'm always interested in volser, but that primarily comes from my all to frequent "what the hell is vos doing? start trolling through src/vol and src/volser to untangle the chain...." [15:06:55] Yes, that's true, if possible, or certainly not blow it if impossible. [15:07:26] I think it has to be possible. It seems like the obvious way to handle up and downgrades. [15:07:38] Upgrade: vos move all your volumes to a new fileserver [15:07:48] Downgrade: vos move all your volumes back to an old one ... [15:07:50] You can't put per-file ACLs into an old-format dump? [15:07:55] So, they go. [15:08:01] But you should keep the rest of your data. [15:08:05] sure [15:08:15] --- stevenjenkins has left [15:08:20] --- deason has left [15:08:35] --- deason has become available [15:09:03] I understand it's not technically easy, that seems like a lot of ACLs lost, potentially [15:12:07] --- stevenjenkins has become available [15:13:06] I suppose you could pack them into a file, or something. [15:43:35] --- mdionne has become available [15:46:59] hi - yes I have per-file ACLs working. Ironed out a few tricky bugs, but it now survives some stress testing [15:48:10] but I haven't looked at dumps, salvager, etc. I picked what seemed to be the simplest back-end to implement for a first pass [15:48:12] can you mention how you did the volume side? (e.g., large vnodes store files?) [15:48:56] I use a separate file ACL file in the "special" directory, which leaves the original files untouched [15:49:32] vos copy is vos move -nodelete, isn't it. sigh. [15:49:34] So the volume index version currently unchanged? [15:50:04] other than the extra file, the only change is an extra entry in the volume header for the handle [15:50:06] --- deason has left [15:50:38] --- stevenjenkins has left [15:50:48] --- deason has become available [15:52:37] so an old fileserver can run with the same files in the /vice partition and not die [15:53:40] --- stevenjenkins has become available [15:54:34] No, it's not. vos copy allocates a new volume ID for the target [15:54:55] ok. good. [15:55:03] vos copy was first. i am glad i didn't screw up [15:56:18] mdionne: that's very orthogonal. [15:57:28] orthogonal? [15:58:14] like, the nice way to do it from the viewpoint of getting it out there sooner than later [16:00:26] if you can make it work with CForeign we may be able to "just ship" an extension [16:00:57] and it will simply work for everyone with a vaguely modern (openafs only) client [16:01:42] I'll try the "foreign" bit tonight - had to work out those other bugs first [16:01:49] great work [16:03:00] the one thing that's not very optimal is that I store an ACL for every file. Which can get big if you have a lot of files on a volume [16:04:22] but doing a more sparse format would be more complex, and would probably require changes to the existing Vnode structure to keep a pointer [16:05:53] The large vnode approach presumably is doable, not future proof. Dare I ask, sqlite, file per volume? [16:06:21] start with hash tables hashed by vnode number, with each record being the address of a block with an acl for a given vnode? [16:09:10] hash: need indirect records, packing, and the like? [16:10:49] I was thinking a hash table on the ACL itself, to quickly see if your ACL already exists. A hash on the vnode could mean you don't have to modify the vnode structure [16:14:03] If you're going to write a new format, or borrow it, should ask whether it could represent xattr data? [16:22:34] well sure, the extra data could be made more general, where an ACL would be one type [17:00:30] --- mmeffie has left [17:01:15] --- matt has left [17:01:47] --- Simon Wilkinson has left [17:04:29] --- Derrick Brashear has left [17:17:40] --- Russ has left: Disconnected [18:38:18] --- deason has left [19:30:05] for CForeign we would need to implement AFS_Lookup in the fileserver. it's currently a stub that just returns EINVAL [19:37:29] --- mdionne has left [21:45:06] --- deason has become available [21:50:00] --- deason has left [22:57:04] --- Russ has become available [23:11:58] --- reuteras has become available [23:21:45] --- dev-zero@jabber.org has become available [23:21:49] --- dev-zero@jabber.org has left: offline [23:29:57] --- Derrick Brashear has become available [23:34:43] --- Simon Wilkinson has become available [23:47:38] --- Russ has left: Disconnected