 |
|
|---|
Yay! This is the second issue of the OpenAFS newsletter. The newsletter summarizes what is happening in the OpenAFS community. This issue has coverage of the workshop.
As always, volunteers, patches, bug reports, or any other type of help is greatly appreciated.
Feedback on this newsletter is welcome. The goal is to summarize the various development efforts and news of OpenAFS for the community. Please let Jason Edgecombe <jason@rampaginggeek.com> know what you would like to see out of this newsletter.
The current and past issues of this newsletter are available at http://www.openafs.org/newsletter/
OpenAFS 1.4.11rc1 was released on June, 11, 2009. It's expected to be the last release from CVS. A transition to git will be made and 1.5.x releases will come from that, with the current CVS HEAD becoming the starting point for the next 1.5.x release. Tools will be needed and are being developed to work with git. We obviously seek people to help us with git (and gerrit) tools, the latter of which requires Java expertise. Many build system changes are in place in the new 1.5.x tree, most for the better.
Testers are wanted. Reports of success or failure of the release candidate should be sent to openafs.info@openafs.org.
OpenAFS 1.5.60 was released on May 31, 2009. Source and binaries are available from http://www.openafs.org/release/openafs-1.5.60.html
The 1.5 branch is now dead. In preparation for the conversion to git, the CVS HEAD was purged of anything that is not on 1.5, and the 1.5 and HEAD branches were then synchronized. 1.5.61 will be cut from git once the head is officially converted.
Known issues in 1.4.10:
A infrequent panic in the Linux VFS layer, which is difficult to reproduce.
Data can be lost when a mapped file is much larger than the cache and the file is written to after a close call.
fileserver host list can be corrupted.
Volumes can go offline during release
Memory corruption when CellServDB contains more than 8 DB servers:
In Windows OpenAFS clients, iTunes can cause issues on 32bit Windows with cache sizes larger then 500MB
Ref: Bug# 124917 http://rt.central.org/rt/Ticket/Display.html
The Sixth Annual International AFS & Kerberos Best Practices Workshop was held at Stanford University on June 1-5, 2009.
Ref: http://workshop.openafs.org/afsbpw09/index.html
Despite a slightly smaller than usual attendance, OpenAFS administrators and developers from a half dozen countries gathered at Stanford University the first week in June for the sixth annual AFS & Kerberos Best Practices Workshop. As always, there were full-day classes in the use and best practices of OpenAFS and Kerberos. Social activities included a fantastic evening at the Computer History Museum and a BBQ dinner. (Organizer Derrick Brashear commented, "No blows were exchanged." However, there was much beer consumed.)
Great talks over two and a half days ranged from status reports and updates on existing software to new toys, plus a repeat of last-year's favorite, the troubleshooting panel. Keynotes were given by Roland Dowdeswell, talking about how a hypothetical multi-national financial firm could use Kerberos in their world-wide infrastructure, and Jim Morris, talking about the origins of what's now called AFS. Slides from most talks are available from the website (http://workshop.openafs.org/).
Plans are already underway for the seventh Workshop, to be held May 24-28, 2010, at the University of Illinois at Urbana-Champaign. We hope to see you there.
Project Contacts:
Simon Wilkinson <sxw@inf.ed.ac.uk>
Dragos Tatulea <dragos.tatulea@gmail.com>
There have been no code changes since the last newsletter, although a number of people continue to test the code. Please consider trying it out, and share any experiences, good or bad. --Simon
There was no progress on my side. -- Dragos
Security Officer:
Simon Wilkinson <sxw@inf.ed.ac.uk>
No security issues since 1.4.9, 1.4.10, and 1.5.59 were released in April 2009.
Project Contact:
Simon Wilkinson <sxw@inf.ed.ac.uk>
Fedora 11 contains updates to the kernel and glibc which means that OpenAFS 1.4.10 will neither build, nor run correctly. Prebuilt binary and source RPMs containing the necessary patches are available from the OpenAFS website, or simply install http://www.openafs.org/dl/openafs/1.4.10/openafs-repository-1.4.10-1.noarch.rpm and run 'yum install kmod-openafs openafs-client openafs-krb5' to install an OpenAFS client.
If you wish to use an existing source tarball on Fedora 11, then you will need to grab the deltas STABLE14-dprintf-rename-20090427 and STABLE14-linux26-defer-cred-changing-20090511 from http://www.openafs.org/cgi-bin/wdelta . The forthcoming 1.4.11 release will contain both of these fixes.
Project Contact:
Simon Wilkinson <sxw@inf.ed.ac.uk>
As has been heavily trailed, our change over from CVS to git for revision control is almost upon us. Along with the new revision control system, we also intend to make use of Google's gerrit code review tool. Gerrit will provide a mechanism by which anyone in the community may submit, review and comment upon forthcoming code changes. More details, and a sneak preview of what is to come, are available in http://workshop.openafs.org/afsbpw09/talks/fri_2/FromTinyAcorns.pdf
Project Contact:
Simon Wilkinson <sxw@inf.ed.ac.uk>
With the arrival of Fedora 11, we say goodbye to Fedora 7. No new binary builds for OpenAFS on Fedora 7 will be produced. This is a significantly longer life than Fedora themselves provide for releases, but please do let us know if this will cause problems. We currently automatically build RPMs for all kernels on both ix86 and x86_64 architectures for Fedora 8, Fedora 9, Fedora 10, Fedora 11, RHEL4 and RHEL5. All these packages are available from the OpenAFS site.
Project Contact:
Jason Edgecombe <jason@rampaginggeek.com>
Progress is good. We have a second issue of the newsletter. I received positive feedback from multiple people. Even one of our very active contributors was surprised by how much was going on. This means that the newsletter is serving a valuable purpose.
FYI, the newsletters are writen in POD format and stored alongside the html files on the OpenAFS web site.
OpenAFS received four slots for the 2009 Google Summer of Code.
Go to http://socghop.appspot.com/org/home/google/gsoc2009/openafs for more information about the GSoC projects.
Three of the projects are still active. Unfortunately, one student had to back out.
The active student projects are:
Student Developer: Jake Thebault-Spieker <summatusmentis@gmail.com>
Mentor: Derrick Brashear <shadow@gmail.com>
Self-intro: https://lists.openafs.org/pipermail/openafs-devel/2009-April/016590.html
Current work in re-ordering the server preferences list takes into account four things. Timeout time from the server, the round-trip time smoothed to allow for short-term network hiccups, and the ratio between packets send and the packets that needed to be resent. Work is still being done to decide how the rank should be scaled for each of these variables, but this may get pushed back until later, as the goal at this point is to get something working. Things in the pipeline: How this affects initial rank after boot-time, having the ranking be adjusted at some time interval, data available to be inferred from the application layers, and the windows client.
Student Developer: Brant Gurganus <brant@gurganus.name>
Mentor: Jeffrey Altman <jaltman@secure-endpoints.com>
Self-intro: https://lists.openafs.org/pipermail/openafs-devel/2009-April/016591.html
I'm intending to make something available this week to make sure it's installing fine and so forth. Last week I didn't really do much since I was at the workshop, though it was nice to get a few questions about the project.
I've probably done more work prior to Google Summer of Code's official coding start time. In that time, I've researched how to tie the MMC framework to Windows Forms, how to install and register an MMC snap-in assembly, and simply looked into how the MMC framework is architected. I've also looked at authoring an MSC console.
(Ed: This was received on June 7, 2009)
Student Developer: Wang Lei <wang840925@gmail.com>
Mentor: David Howells <dhowells@redhat.com>
Self-intro: https://lists.openafs.org/pipermail/openafs-devel/2009-April/016595.html
In brief, we have finished three kAFS pioctl operations and have several pioctls partially implemented. I have read some of the OpenAFS documents and I am becoming more familiar with kAFS and OpenAFS as the project progresses.
With the help of my mentor, I have built an environment with one OpenAFS server, one OpenAFS client and one kAFS client for development and testing, and I have become familiar with the tools and development environments, e.g. git and the formatting of kernel patches. I have learnt a lot about OpenAFS and dealing with open source projects, since the Linux kAFS GSoC project started. Next, I will work to accelerate the completion of the rest of the pioctl operations.
Workshop presentation on kAFS: http://workshop.openafs.org/afsbpw09/thu_2_2.html
Project Contacts:
Matt Benjamin <matt@linuxbox.com>
Marcus Watts <mwd@umich.edu>
Rxk5 patches against 1.5.60 and 1.4.10 are available for folks who would like to try rxk5 in a test cell. (A number of sites have offered to do this.) We will provide assistance as needed. Marcus is working on a draft protocol description to be published in afs3-standardization. David Howells is beginning an implementation of rxk5 in kAFS.
Project Contacts:
Matt Benjamin <matt@linuxbox.com>
Support for OpenBSD 4.5 was -really- added. Build support for Freebsd 7.2 should appear shortly. Fixes for at least one outstanding locking issue on FreeBSD post 7.0 are being worked on. A NetBSD cache manager port is in progress.
Project Contacts:
Matt Benjamin <matt@linuxbox.com>
I've created preliminary patches against openafs-devel-1.5.x for extended callbacks and dependencies. There's still some testing work for the latest xcb changes, which added features over the previous milestone of Aug. 2008. I'll be publishing paths to patches and requests for review over the next few weeks.
Project Contacts:
Hartmut Reuter <reuter@rzg.mpg.de>
Felix Frank <Felix.Frank@Desy.de>
Christof Hanke <hanke@rzg.mpg.de>
RXOSD is an extension for OpenAFS to allow file data to be put into dedicated "object storage", which is arbitrarly spread across servers. Among other things, this allows read/write replication of data. An enhancement to the basic object storage paradigm is the ability to transparently archive and restore data to and from HSM systems, similar to the way MR-AFS used to facilitate. Hartmut Reuter and Christof Hanke (both RZG) and Felix Frank (DESY) work on the integration of this extension of OpenAFS. Patches submitted to the OpenAFS-Gatekeepers are aimed to be small and amount to no more than one change each. Everyone interested in RXOSD and/or its integration into OpenAFS is cordially invited to send an email to the integrators or the openafs-info mailing list.
Ref: https://lists.openafs.org/pipermail/openafs-devel/2009-May/016625.html
Concerning vicep-access I can tell you that this will come into OpenAFS along with RXOSD because it's just another protocol supported by the restructured client. However, it's ifdefed and configured separately so that it could be used independently from RXOSD. But, of course, it makes more sense to use it along with RXOSD because filesystems such as Lustre and GPFS are good for large files, but slow for file creation and deletion. So you would prefer to keep the small files in the fileserver's local partition.
Workshop Presentation on vicep direct access: http://workshop.openafs.org/afsbpw09/thu_2_3.html
Project Contacts:
Russ Allbery <rra@stanford.edu>
Jason Edgecombe <jason@rampaginggeek.com>
Thanks to contributions from Steven Jenkins and Mike Robinson of End Point, our man page coverage is now essentially complete. Every OpenAFS command has a corresponding man page.
Thanks to Jeff Altman of Your File System, Inc., the old IBM architecture and protocol documentation has been converted to header files with Doxygen comments, allowing us to generate that documentation from editable source and eventually better integrate it with the OpenAFS source. This work was funded under a SBIR grant.
The build system for the OpenAFS manuals has been overhauled to use more current tools. We now generate Windows Help files from the same source as well as much better HTML and PDF output and get the version number for the manual from the OpenAFS build system.
Thanks to edits by Jason Edgecombe and reviews by Jeffrey Altman and Russ Allbery, chapter 1 of the Admin Guide has been updated to give warnings about kaserver and include Kerberos 5.
Here is a list of tickets that have been resolved since May 1, 2009:
ticket # state created title
20947: resolved Aug 18, 2005 Better cache partition checks in afsd
23750: resolved Nov 29, 2005 1.4.0 client crash
60275: resolved Apr 25, 2007 'make dest' fails to create man dir + manpages
124130: resolved Jan 14, 2009 Convert MR-AFS Residency Command to Rx OSD
124359: resolved Feb 13, 2009 Read-write volume corruption on openafs 1.5.x
124514: resolved Mar 19, 2009 OpenAFS 1.4.9 milestone
124522: resolved Mar 20, 2009 krb5 platform build defaults
124551: resolved Mar 28, 2009 [PATCH] encryption for PTS queries
124552: resolved Mar 28, 2009 [PATCH] Use $AFS_POST_INIT in Red Hat initscript if defined
124563: resolved Mar 31, 2009 Print unsigned values as such in rxdebug
124569: resolved Mar 31, 2009 fbsd 1.5.x sig_t old typo
124581: resolved Apr 02, 2009 stack overflow afs_GetDownD
124627: resolved Apr 15, 2009 Re: [OpenAFS-devel] Cache inconsistency in client 1.4.8 and above
124671: resolved Apr 22, 2009 Handling last store of dirty pages with memcache
124681: resolved Apr 23, 2009 patch for building on freebsd5.5
124709: resolved Apr 28, 2009 linux26 issue with pagsh.krb and Pag location in group list
124715: resolved Apr 30, 2009 minor path remods for src/tests; refactor of src/tests/afs-rmcell.sh to use new style
124716: resolved Apr 30, 2009 src/tests/OpenAFS/Dirpath.pm.in should be cvs remove'd
124719: resolved May 01, 2009 obsd 44 and 45
124725: resolved May 04, 2009 Please pull up XML documentation
124737: resolved May 05, 2009 periodic kernel panic - linux 2.6.29.1 / OpenAFS 1.4.10
124742: resolved May 06, 2009 AFS Explorer Shell IsPathInAFS broken for dead-end symlinks
124770: resolved May 12, 2009 1.5 build fix: mmap patch fallout
124776: resolved May 15, 2009 "vos partinfo" gives wrong output on windows.
124787: resolved May 18, 2009 AFS client 1.5.59 crashing with "Unhandled exception" on Windows 2003
124788: resolved May 18, 2009 compile_et man page
124789: resolved May 18, 2009 add copyauth man page
124790: resolved May 18, 2009 fs monitor man page
124791: resolved May 18, 2009 man page for restorevol
124792: resolved May 18, 2009 man page for rmtsysd
124793: resolved May 18, 2009 man page for vsys
124794: resolved May 18, 2009 fix CellServDB man page coveraged of -dynroot
124799: resolved May 18, 2009 fix .krb commands
124800: resolved May 18, 2009 New man pages for Windows installer
124810: resolved May 26, 2009 errno compile error in JAVA source
124878: resolved May 27, 2009 crash dump from Windows XP (32-bit) and 1.5.59 client
124880: resolved May 27, 2009 Add -rxmaxmtu parameter to afsd to help avoid VPN fragmentation of UDP packets
124891: resolved Jun 01, 2009 CellServDB parsing subject to buffer overflows
124893: resolved Jun 02, 2009 ptserver debug flag correction
124894: resolved Jun 02, 2009 ptprocs vicelog newlines and loglevels
124898: resolved Jun 03, 2009 src/aklog/klog.c uninitialized variable tofree
124910: resolved Jun 08, 2009 Windows: cm_cell_t lists not properly protected by cm_cellLockThe CVS statistics have been removed for now. The HEAD branch has been synchronized with the 1.5 branch in preparation fo the git transition. The 1.5 branch is now dead.