| Commit message (Collapse) | Author | Age | Files | Lines |
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
* Previous contents of base/ are now in base/all.
* create_node_COMMON() takes an extra argument (inserted as $3) that
specifies the node type. All of the create_node_* functions pass
"$type" for this argument.
* setup_base() now has an optional argument, which is only passed by
create_node_COMMON. Rather than simply copying and substituting
templates for the entire $BASE_TEMPLATES/ subdirectory it now does
this for $BASE_TEMPLATES/all and then $BASE_TEMPLATES/$type.
Therefore, the contents of the type-specific directory override the
contents of "all" directory.
* releases/SoFS.common uses the "all" version of SoFS-postinstall.sh
when setting POSTINSTALL_TEMPLATE. This variable also now respects
$BASE_TEMPLATES instead of hard-coding the base/ subdirectory.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
| |
Having this list of packages hardcoded in the postinstall script is a
maintenance nightmare. This does the same thing using a variable,
with the default value in releases/SoFS.common.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
| |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
| |
Thanks to Klaus Smolin and Ronnie Sahlberge.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
|
| |
GPFS 3.3 requires that adminMode be set to allToAll for mm commands to
be run from any node. The redirect from /dev/null ensures that older
versions of GPFS won't get stuck when they prompt you to warn about an
unknown option - hopefully the failure will be graceful.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
| |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
| |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
| |
I must have copied from an example that I tested and inadvertently
changed @@CLUSTER@@ to "m1" in the GPFS cluster name.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
| |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
| |
The previous commit introduced an error because the template
substitution seems to eat backslashes in certain circumstances.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
|
|
|
| |
SOFS_GUI_PACKAGES can now be set to empty to stop these packages being
installed by the postinstall.
Also, if SOFS_GUI_PACKAGES is non-null, WITH_SOFS_GUI is null and
there is at least 1 sofs_gui node in NODES, then still attempt to
install the GUI pacakges.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
| |
The previous version made the script confusing to read once the
template substitution had been done. Assign the value to a variable
for clarity.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
| |
Indicates how many NSDs are used by the default GPFS filesystem that
is setup by setup_gpfs.sh. The default empty value indicates that all
available NSDs should be used.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
| |
Use only the storage nodes and n1. Might refine this later.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
| |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
|
| |
This allows IDE to be used for system disks and some other type,
perhaps virtio, to be used for the shared disks. I seem to be seeing
some corruption issues with the combination of qcow2 and virtio, so
this will allow me to experiment with the raw shared disks and virtio.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
|
| |
Recent versions of CTDB no longer comment out the relevant line in the
config file, so weaken the substitution. While we're here, clean
things up a bit: use "sed -i" to do in-place modification and use
cnpushfile for clarity.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
| |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
| |
The version of kvm under RHEL5.4 does not support the SCSI emulation.
So allow the use of ide or virtio instead.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
| |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
| |
There looks to be nothing harmful in conf.base-SoFS-1.5.3, so merge it
all into conf.base-SoFS and remove the 1.5.3-specific version.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |\ |
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
ctdb 1.0.78 changes the format of the output of "ctdb -Y status",
causing the cluster_is_healthy() function in
base/root/scripts/functions to always fail. This relaxes the matching
to look for a 1 bit in the output for a node rather than a particular
number of 0s.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |/
|
|
| |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
| |
No longer mysteriously fail if there is only one entry in NSD_SERVER -
use the first node in the CTDB cluster as the secondary GPFS server.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
|
|
|
|
| |
Both scripts now use nsd_server_common.bash to re-exec on the first
node defined in $NSD_SERVERS if that variable is set (to the list of
sofs_storage nodes by 50sofs.defconf). They also also include the NSD
servers in the file that is passed to mmcrnsd and mmcrfs.
setup_gpfs.sh includes any NSD servers in the list of nodes for the
GPFS cluster, using them as the primary and secondary GPFS servers.
setup_gpfs.sh just sources mknsd.sh to avoid duplication of logic.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Using "yum clean all" before doing a "yum update" seems to avoid
certain weird problems. We were doing a clean before an update in
both the basic postinstall (which is always present) and the
SoFS-postinstall. However, this wasn't avoiding problems when a base
image was being used for some other purpose. The current thinking is
to put a clean after the update in the basic postinstall - so this is
the last thing that happens when a base is created without the
SoFS-postinstall. This should mean that, when a node is created using
the base, any updates will need to fetch fresh repo data... so the
update should be clean.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
|
| |
There is occasionally weirdness where something to do with yum gets
left in a strange state that causes dependency problems. Adding a
gratuitous "yum clean all" before doing a "yum update" seems to avoid
some of these problems... and it isn't very expensive...
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
New function make_public_addresses() generates a template fragment for
public addresses. New configuration variable PUBLIC_ADDRESSES is set
to "|make_public_addresses", which builds a sensible set of public
addresses with addresses in a range relative to nodes' management
addresses. conf.base-SOFS* files include @@@PUBLIC_ADDRESSES@@@
rather than "static" public address definitions.
This means that the default public addresses for multiple virtual
clusters on the same virtualisation host will no longer clash.
The old behaviour, with public addresses whose suffixes start at 101,
can be re-enabled by setting PUBLIC_ADDRESSES to
"|make_public_addresses 101 1 3". A more sensible approach, which
generates public addresses for the right number of nodes and excludes
public addresses from node 1, involves setting PUBLIC_ADDRESSES to
just "|make_public_addresses 101".
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
| |
Forgot to "git add" it last time and it got missed.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Rename conf.base to conf.base-SoFS, since it is SoFS-specific. Add
conf.base-SoFS-1.5.3 and modify setup_samba.sh to look for a
version-specific file if it exists.
Revert default value of MEM to 256MB and add 128MB to it in any
release files (SoFS-1.5.2, SoFS-1.5.3) that tell the kernel to reserve
128MB for kdump. This introduces more order dependencies when using
release files... but there are obvious advantages. Updated GUIMEM to
be a nice round 704MB and set it to 1GB in recent release files
(SoFS-1.5.2, SoFS-1.5.3) since it now seems much easier to run out of
memory.
Update note in README about configuration order being important when
using release files. Add new file TODO.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
| |
invocation of "sofs config" in setup_samba.sh to "cnconfig".
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
manage 2 level clusters, where each node's disk image has its own base
file. Such as cluster can be reset by simply recreating each node's
disk image. The main issue is that nbd can be unreliable and we try
to avoid using it from cron. Details:
* Rename force-net.sh to basic-postinstall.sh, and move repository
setup and initial yum update here from postinstall.sh.
* Rename postinstall.sh to SoFS-postinstall.sh, since it is now
SoFS-specific.
* The default value for POSTINSTALL_TEMPLATE is now "" - the SOFS
release files set it to SoFS-postinstall.sh. Set it back to "" in a
configuration file to build bare images that can be used as bases.
Other changes:
* Hack substitute_vars() so it uses /dev/null if passed a zero length
input file and so it echoes to stdout if no output file is given -
the latter allows inline use in command substitutions.
* The name of the install yum repository configuration file is now
taken from the basename of the YUM_TEMPLATE option.
* with_release() now prints a list of available releases if an
incorrect release is specified.
* Fix the -c option if a non-local file is given.
* Fix typo in SoFS.repo for SoFS-daily. It should now work.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |
|
|
|
|
|
| |
postinstall.sh configurable via the new configuration variable
GPFS_BASE_VERSION. Set it to 3.2.1-10 for SoFS-1.5.3.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |\ |
|
| | |
| |
| |
| |
| |
| | |
to build 32-bit winbind libraries.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |\| |
|
| | |
| |
| |
| | |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| | |
| |
| |
| | |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |\| |
|
| | |
| |
| |
| | |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| | |
| |
| |
| |
| |
| |
| |
| | |
/etc/httpd/conf.d/shares.config to fix HTTP. Miscellanous additions
to match what the GUI does: idmap:cache = no, wide links = no,
CTDB_MANAGES_GPFS = yes, DAEMON_CORE_LIMIT = unlimited.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| | |
| |
| |
| | |
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| | |
| |
| |
| |
| |
| | |
cluster_is_healthy.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| | |
| |
| |
| |
| |
| |
| |
| | |
waits for a completely healthy cluster. chroot jail generation goes
after winbind start. We also no longer try to fix the jail - assume a
working cnjailgen is hacked in via some other process.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| | |
| |
| |
| |
| |
| | |
disappeared.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| | |
| |
| |
| |
| |
| |
| |
| | |
GUI packages. New option EXTRA_KERNEL_OPTIONS can be used to pass
extra options to the kernel when booting nodes. Removed unneeded
comments from config.default.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
separate template that only gets included in the kickstart templates.
Take openldap-devel and -scsi-target-utils out of EXTRA_PACKAGES - you
can't change a previous selection. setup_base needs to call
setup_indirects, since postinstall.sh needs YUM_REPOS. Add a 2nd
network interface to install.xml (and bootbase.xml), so we can keep
the 2nd network interface specified in the kickstart template. Fix
some stuff in the kickstart templates, comment out some things that
don't work, add AUTOCLUSTER_POSTINSTALL.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| | |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| | |
/root/postinstall.sh. This has also been removed from the 1.5
kickstart template and gets substituted in via new config option
KICKSTART_TEMPLATE, which should be set to /dev/null if none of the
fancy kickstart post-install magic is required. chkconfig and start
of ntpd moved to setup_samba.sh. Also added chkconfig for ctdb there.
Added to default value of EXTRA_PACKAGES: openldap-devel,
-scsi-target-utils, -xorg-x11-drivers, -xorg-x11-server-Xorg. New yum
repo templates: SoFS-1.5.1.repo, SoFS-1.6.repo, SoFS-daily.repo.
Updated SoFS-1.5.repo to use SoFS-1.5-Stable and removed SoFS-daily
from it. made kickstart-1.5.cfg as consistent as possible with the
official SoFS version... and duplicated it to kickstart-1.5.1.cfg.
Signed-off-by: Martin Schwenke <martin@meltin.net>
|
| |/ |
|