Whamcloud - gitweb
yujian [Mon, 26 Jun 2006 10:07:42 +0000 (10:07 +0000)]
b=9863
Modify the condition to check whether a target was made to be high-available.
yujian [Mon, 26 Jun 2006 10:04:42 +0000 (10:04 +0000)]
b=9853
1) Improve help infomation, add descriptions of each field in the csv format
2) Remove "heartbeat channels", "service address" and "heartbeat options" fields from the csv format. All the generated HA config files are suffixed with ".lustre".
yujian [Mon, 26 Jun 2006 09:54:48 +0000 (09:54 +0000)]
b=9853
1) Add more checks and prompt messages before generating configuration files
2) Write Lustre HA information into *.lustre files instead of changing the existing HA configuration files
nathan [Fri, 23 Jun 2006 19:41:48 +0000 (19:41 +0000)]
Branch b1_5
b=9858
more test fixes for b1_5
nathan [Fri, 23 Jun 2006 19:02:32 +0000 (19:02 +0000)]
Branch b1_5
b=8007
ia64 doesn't have BLKGETSIZE64
nathan [Fri, 23 Jun 2006 18:22:00 +0000 (18:22 +0000)]
Branch b1_5
b=8007
some better comments
scjody [Fri, 23 Jun 2006 01:28:12 +0000 (01:28 +0000)]
Branch b1_5
b=9387
r=adilger
Select next connection according to last successful connection time.
nathan [Fri, 23 Jun 2006 00:12:12 +0000 (00:12 +0000)]
Branch b1_5
b=9858
fix lots of cancel_lru_locks calls
nathan [Thu, 22 Jun 2006 23:34:53 +0000 (23:34 +0000)]
Branch b1_5
b=9858
fix cancel_lru_locks()
nathan [Thu, 22 Jun 2006 22:44:46 +0000 (22:44 +0000)]
Branch b1_5
b=6221
6221 didn't apply cleanly to b1_5
nathan [Thu, 22 Jun 2006 22:44:09 +0000 (22:44 +0000)]
Branch b1_5
b=9858
remove all special mountconf test exceptions
vitaly [Thu, 22 Jun 2006 22:09:31 +0000 (22:09 +0000)]
Branch b1_5
do not allow client_common_fill_super to LBUG with osfs.os_bsize == 0,
mfs_statfs returned 'cached' osfs value, when there was no a such.
green [Thu, 22 Jun 2006 16:36:37 +0000 (16:36 +0000)]
AC_CHECK_HEADERS([linux/xattr_acl.h]) check is incorrect and unused. Removing.
lsy [Thu, 22 Jun 2006 15:17:37 +0000 (15:17 +0000)]
minor fix of the offset for liblustre.
green [Thu, 22 Jun 2006 14:37:01 +0000 (14:37 +0000)]
linux/bit_spinlock.h is not present in some kernels. Introduce configure check
vitaly [Thu, 22 Jun 2006 11:52:50 +0000 (11:52 +0000)]
Branch b1_5
b=10318
make new lfind behaviour the default one, only if some old (obsolete)
lfind option is specified, lfind works in the 'old' mode.
green [Wed, 21 Jun 2006 21:48:09 +0000 (21:48 +0000)]
Make patchless lustre client to compile and load on vanilla 2.6.17
scjody [Wed, 21 Jun 2006 21:22:47 +0000 (21:22 +0000)]
Merge b1_5 from b1_4 (20060621_1641)
scjody [Wed, 21 Jun 2006 20:16:58 +0000 (20:16 +0000)]
Merge b1_5 from b1_4 (20060616_1515)
nathan [Wed, 21 Jun 2006 16:59:51 +0000 (16:59 +0000)]
Branch b1_5
b=9860
a little safer and more verbose upgrade script
yujian [Wed, 21 Jun 2006 10:48:16 +0000 (10:48 +0000)]
b=9863
1) Get "--stripe-count-hint" option
2) Calculate the inode size of mdt according to stripe count
nathan [Tue, 20 Jun 2006 23:25:45 +0000 (23:25 +0000)]
Branch b1_5
b=8007
better leave this verbose until we get the kinks out
nathan [Tue, 20 Jun 2006 21:42:51 +0000 (21:42 +0000)]
Branch b1_5
more info on err
nathan [Tue, 20 Jun 2006 20:00:30 +0000 (20:00 +0000)]
Branch b1_5
b=8007
remove extra comma
nathan [Tue, 20 Jun 2006 18:26:06 +0000 (18:26 +0000)]
Branch b1_5
b=8007
-tell users they may need our e2fsprogs
-tunefs.lustre must figure out if mgsnode is set
alex [Tue, 20 Jun 2006 17:35:58 +0000 (17:35 +0000)]
- few more defines for liblustre to get updated ldlm buildable at least
alex [Tue, 20 Jun 2006 16:31:09 +0000 (16:31 +0000)]
- missed to fix ns_lock in liblustre
scjody [Tue, 20 Jun 2006 15:53:46 +0000 (15:53 +0000)]
file ext3-check-jbd-errors-2.6.9.patch was added on branch b1_5 on 2006-06-21 21:22:45 +0000
scjody [Tue, 20 Jun 2006 15:53:39 +0000 (15:53 +0000)]
file ext3-check-jbd-errors-2.6.5.patch was added on branch b1_5 on 2006-06-21 21:22:45 +0000
alex [Tue, 20 Jun 2006 11:30:18 +0000 (11:30 +0000)]
b=9332
- instead of taking a lock covering whole requested region ll_file_read()
and ll_file_write() grab locks that cover a single stripe at most (but
not in O_APPEND case). this way we improve overall stability because
we don't hold locks awaiting some failed OST to recover.
yujian [Tue, 20 Jun 2006 09:01:05 +0000 (09:01 +0000)]
b=9863
Change "failnode=" to "failover.node=" according to lustre_param.h
brian [Tue, 20 Jun 2006 00:53:46 +0000 (00:53 +0000)]
Nice catch Nathan. Forat difference between the three lines threw me for a
loop, so fix that too. :-)
brian [Mon, 19 Jun 2006 23:59:11 +0000 (23:59 +0000)]
Stay in parity with lustre_param.h, revision 1.1.4.5.
nathan [Mon, 19 Jun 2006 23:30:53 +0000 (23:30 +0000)]
b=9846
Change parameters to hierarchical format
nathan [Mon, 19 Jun 2006 20:14:17 +0000 (20:14 +0000)]
Branch b1_5
b=8007
r=adilger
remove cruft in utils
nathan [Mon, 19 Jun 2006 20:09:03 +0000 (20:09 +0000)]
Branch b1_5
b=8007
minor error message improvement
nathan [Mon, 19 Jun 2006 19:20:51 +0000 (19:20 +0000)]
Branch b1_5
sector_t-overflow doesn't apply to my 2.6.10 fc3 uml
alex [Mon, 19 Jun 2006 10:29:26 +0000 (10:29 +0000)]
b=10088
- fine-grained locking for LDLM landed
brian [Sat, 17 Jun 2006 22:05:32 +0000 (22:05 +0000)]
Need to remember that these tests are not just done "in-tree", they are also
done on properly installed (make install, rpm, etc.) lustre so...
When trying to install a modules, test if the module is in-tree and if not,
try to modprobe it.
brian [Sat, 17 Jun 2006 16:50:25 +0000 (16:50 +0000)]
/mnt might not be on a writable filesystem, must $MOUNT should be, so just
use its parent directory.
brian [Sat, 17 Jun 2006 12:20:07 +0000 (12:20 +0000)]
Create the mountpoints for the servers.
yujian [Sat, 17 Jun 2006 07:14:08 +0000 (07:14 +0000)]
b=9853
Figure out the scripts path instead of hardcode it.
brian [Fri, 16 Jun 2006 21:32:25 +0000 (21:32 +0000)]
As per lustre_config.sh's new functionality, default the mount options to
"noauto".
brian [Fri, 16 Jun 2006 20:27:51 +0000 (20:27 +0000)]
The mkfs_options are I guess more properly mkfs parameters and need to be
included, wrapped with --param="" in the format options.
brian [Fri, 16 Jun 2006 19:31:34 +0000 (19:31 +0000)]
mkfs options needs to be passed to mkfs.lustre inside of --param="". I
wonder if lustre_config.sh shouldn't do this wrapping for us before passing
to mkfs.lustre.
brian [Fri, 16 Jun 2006 18:57:41 +0000 (18:57 +0000)]
VS: Branch b1_5
Off-by-one in the OST entries.
brian [Fri, 16 Jun 2006 17:29:16 +0000 (17:29 +0000)]
Quiet perl down a bit.
brian [Fri, 16 Jun 2006 16:53:33 +0000 (16:53 +0000)]
Allow the caller to specify where the mounts on the nodes should be done.
Remove that silly "main()" function.
yujian [Fri, 16 Jun 2006 16:06:27 +0000 (16:06 +0000)]
b=9853
1) Remove "-d opts" option, as mentioned in comment #96 of bug9853, using "mount options" field in the csv file to specify the options to be added into /etc/fstab.
2) Also, change the meaning of "-m" option to "don't modify /etc/fstab". If using this option, then the value of "mount options" field will be passed to mkfs.lustre instead of being added into /etc/fstab.
yujian [Fri, 16 Jun 2006 13:53:40 +0000 (13:53 +0000)]
b=9863
"extents,mballoc" are default mount options of OST with ldiskfs.
yujian [Fri, 16 Jun 2006 10:38:56 +0000 (10:38 +0000)]
After adding the vim modeline into this file, the indentations of it became a bit mess due to my mix-using spaces and tabs originally. This version is neatly arranged.
brian [Fri, 16 Jun 2006 00:29:07 +0000 (00:29 +0000)]
Have to include the contents of man_MANS in EXTRA_DIST, not just a reference
to it. This is because man_MANS is conditional on building the utils but
even if the tree is configure without utils, it should not prevent the utils
files from being put in the "dist" tarball.
nathan [Thu, 15 Jun 2006 22:20:49 +0000 (22:20 +0000)]
Branch b1_5
b=9861
- turn on extents, mballoc on OSTs by default
- don't try to pre-check lustre/llite/ldiskfs options in mount.lustre.
Instead, just let the mount call fail.
nathan [Thu, 15 Jun 2006 20:55:21 +0000 (20:55 +0000)]
Branch b1_5
b=8216
put mount.lustre.8 back
nathan [Thu, 15 Jun 2006 20:51:37 +0000 (20:51 +0000)]
Branch b1_5
b=8216
add mount.lustre man page
brian [Thu, 15 Jun 2006 20:27:23 +0000 (20:27 +0000)]
Remove mount.lustre.8 until the file shows up. Probably just a forgotten
cvs new on it, but this is breaking builds.
brian [Thu, 15 Jun 2006 19:40:43 +0000 (19:40 +0000)]
Quick hack to be able to define the options in the /etc/fstab for the lustre
filesystem added.
Probably some logic could be added in get_mntopts to detect the failover
option and only add noauto if it was not added by the command line argument.
nathan [Thu, 15 Jun 2006 18:52:20 +0000 (18:52 +0000)]
Branch b1_5
b=8216
man pages for mkfs.lustre, tunefs.lustre, mount.lustre, and Lustre
brian [Thu, 15 Jun 2006 03:01:14 +0000 (03:01 +0000)]
Don't format devices. This hack will be replaced by a command line switch
in lustre_config.sh to ask for a "no format, but configure" operation.
nathan [Thu, 15 Jun 2006 00:11:52 +0000 (00:11 +0000)]
Branch b1_5
b=8007
remove some non-applicable options from tunefs
nathan [Thu, 15 Jun 2006 00:05:44 +0000 (00:05 +0000)]
Branch b1_5
b=8007
add a tunefs.lustre manpage
nathan [Wed, 14 Jun 2006 23:28:53 +0000 (23:28 +0000)]
Branch b1_5
b=8007
oooh, a man page.
brian [Wed, 14 Jun 2006 22:03:42 +0000 (22:03 +0000)]
The start of a script to convert lmc "batch" specifications to the CSV
format used by the lustre_config.sh script to configure lustre.
The number of lmc options the script understands currently seems to cover
ltest's lmc batch generation.
This is the "cheap way" to get ltest to do mountconf and be able to do full
testing on b1_5 and has the added benefit of testing the lustre_config.sh
scripts while it's at it.
Note, that arguably we should be parsing the XML, not lmc batch commands.
Fortunately the way the script works, changing out parsing of lmc commands
for parsing XML is easy to do and will still work with the backend CSV
generation.
green [Wed, 14 Jun 2006 20:57:09 +0000 (20:57 +0000)]
b=6062
r=adilger(original author)
SPEC SFS NFS v2 validation failure fix - make blocksize constant at 2*RPC size.
green [Wed, 14 Jun 2006 20:26:10 +0000 (20:26 +0000)]
patchless support.
nathan [Wed, 14 Jun 2006 19:55:38 +0000 (19:55 +0000)]
Branch b1_5
b=8007
mtab must be checked before writing anything.
we no longer need the loopback setup for reading.
brian [Tue, 13 Jun 2006 19:05:14 +0000 (19:05 +0000)]
Some safety so that DEVICE_NAME[i] being "/dev/loop1" doesn't delete an
fstab entry with a device of "/dev/loop11".
brian [Tue, 13 Jun 2006 15:49:29 +0000 (15:49 +0000)]
Whitespace consistency. I believe our standard is 4 spaces for indentation,
no hard tabs.
yujian [Tue, 13 Jun 2006 13:57:53 +0000 (13:57 +0000)]
b=9863
Use debugfs instead of mounting the target device to read the Lustre data files.
vitaly [Sun, 11 Jun 2006 16:54:36 +0000 (16:54 +0000)]
Merge b1_5 from b1_4 (20060607_2142)
yujian [Sun, 11 Jun 2006 10:45:02 +0000 (10:45 +0000)]
b=9863
Add figure_journal_size() to get the default journal size of lustre target.
nathan [Fri, 9 Jun 2006 23:11:05 +0000 (23:11 +0000)]
Branch b1_5
b=9860
better errors
nathan [Fri, 9 Jun 2006 23:10:00 +0000 (23:10 +0000)]
Branch b1_5
b=9860
Add upgrade helper script per Andreas' suggestion
nathan [Fri, 9 Jun 2006 19:29:43 +0000 (19:29 +0000)]
Branch b1_5
b=8007
Better error message for wrong fsname
nathan [Fri, 9 Jun 2006 18:05:04 +0000 (18:05 +0000)]
Branch b1_5
b=8007
remove lmc and lconf man pages
allow setting fsname before target type
yujian [Fri, 9 Jun 2006 11:26:08 +0000 (11:26 +0000)]
b=9853
Repeat "--mgsnode=" or "--failnode=" keyword to delimit multiple mgs nodes or failover nodes.
lsy [Fri, 9 Jun 2006 02:04:46 +0000 (02:04 +0000)]
fix for the reply offset, found in sanity.sh 75.
nathan [Thu, 8 Jun 2006 22:15:23 +0000 (22:15 +0000)]
Branch b1_5
b=9858
more test fixes
nathan [Thu, 8 Jun 2006 21:52:59 +0000 (21:52 +0000)]
Branch b1_5
b=9858
more test fixes
nathan [Thu, 8 Jun 2006 17:02:25 +0000 (17:02 +0000)]
b=10425
lnet_nid2str() should never do hostname lookups
b=10515
resolve hostname for all nettypes in mkfs.lustre and mount.lustre using
the above functionality
nathan [Thu, 8 Jun 2006 16:17:45 +0000 (16:17 +0000)]
Branch b1_5
b=9858
full ost tests don't work on uml
yujian [Thu, 8 Jun 2006 09:46:49 +0000 (09:46 +0000)]
b=9863
If the target service is a failover partner service, then add "--noformat" option.
nathan [Wed, 7 Jun 2006 23:29:15 +0000 (23:29 +0000)]
Branch b1_5
b=9858
- add ogdb debug file
- load modules from setup if needed
nathan [Tue, 6 Jun 2006 22:35:11 +0000 (22:35 +0000)]
b=8007
- change module name from "llite.ko" to "lustre.ko" (r=adilger)
- drop ldap and python from rpms
- remove initd scripts from rpms
- remove old cluster_scripts dir
nathan [Mon, 5 Jun 2006 20:55:32 +0000 (20:55 +0000)]
Branch b1_5
b=9858
make "-v" verbose option work
nathan [Mon, 5 Jun 2006 19:51:38 +0000 (19:51 +0000)]
Branch b1_5
b=8007
No more lustre init scripts. Mount it in fstab.
nathan [Mon, 5 Jun 2006 19:30:18 +0000 (19:30 +0000)]
Branch b1_5
nathan [Mon, 5 Jun 2006 18:32:01 +0000 (18:32 +0000)]
Branch b1_5
b=9858
setup for other acceptance-small tests
qiyong [Mon, 5 Jun 2006 09:44:24 +0000 (09:44 +0000)]
warning fix: lload_SOURCES multiply defined
yujian [Mon, 5 Jun 2006 09:03:59 +0000 (09:03 +0000)]
b=9863
1) Support getting --param="<key>=<value>" format options.
2) Convert IP addresses in MGS nids into hostnames.
3) Get CluManager configurations.
yujian [Mon, 5 Jun 2006 09:00:11 +0000 (09:00 +0000)]
b=9853
Support --param="<key>=<value>" format options.
yujian [Sun, 4 Jun 2006 11:40:41 +0000 (11:40 +0000)]
b=9863
Get lnet module options without start and end markers.
yujian [Sun, 4 Jun 2006 07:31:39 +0000 (07:31 +0000)]
b=9853
Fix a bug - "Device does not exist for Service".
nathan [Fri, 2 Jun 2006 23:01:09 +0000 (23:01 +0000)]
Branch b1_5
b=10510
allow client remount
fix up broken mtab options code
nathan [Fri, 2 Jun 2006 19:24:51 +0000 (19:24 +0000)]
Branch b1_5
b=10586
Removed the few bits of code assuming a single MGC. Now, name MGCs with the
first NID of the MGS they point to; start a new MGC if the name doesn't exist.
This allows maximal re-use of the MGC (all servers/clients pointing to the same
MGS), but also allows connects to new MGSs as needed.
yujian [Fri, 2 Jun 2006 10:34:51 +0000 (10:34 +0000)]
Add Lustre cluster scripts.
yujian [Fri, 2 Jun 2006 10:22:20 +0000 (10:22 +0000)]
b=9863
Script to generate a csv file from a running lustre cluster.
yujian [Fri, 2 Jun 2006 10:21:21 +0000 (10:21 +0000)]
b=9853
Script to verify the service IP and the real interface IP in a remote host are in the same subnet.
yujian [Fri, 2 Jun 2006 10:19:54 +0000 (10:19 +0000)]
b=9853
Script to generate the Red Hat Cluster Manager HA software's configuration files.
yujian [Fri, 2 Jun 2006 10:18:46 +0000 (10:18 +0000)]
b=9853
Script to create mon.cf file for Heartbeat.