The getInstReasonFilename is built to resemble the python corresponding
function.

Signed-off-by: Michele Tartara <mtartara@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

The tests are already using this library, so it's not really a new build
dependency, but it was not specified esplicitly.

Furthermore, it's going to be used by the instance status collector, so it's
added to the requirements for the monitoring subsystem.

Signed-off-by: Michele Tartara <mtartara@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

While this use case was described in the design document, and
mentioned several times as motivation for changes in commit messages,
it has never been added to a user-facing documentation. This commit
adds at least an example to the man page.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

While the multiple-tags test was added to verify that coloring is done
only after node selection (otherwise it wouldn't be possible to get in
both cases a single reboot group), it can easily be extended to also
verify that the correct nodes are selected by --node-tags.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

In the example configuration, the graph constructed by just connecting
primary and secondary instances is two-colorable. However, when taking
conflicting locations of secondary nodes into account, three reboot
groups are needed. Moreover, these reboot groups are not subordinated
to any two-coloring of the first-mentioned graph.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Thomas Thrainer <thomasth@google.com>

Make hroller take into account the nodes (redundant) instances
will be migrated to. This be behavior can be overridden by the
--offline-maintenance option which will make hroller plan under
the assumption that all instances will be shutdown before starting
with the rolling reboots.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Thomas Thrainer <thomasth@google.com>

For online rolling reboots, there are two kind of restrictions. First,
we cannot reboot the primary and secondary nodes of an instance
together. Secondly, two nodes cannot be rebooted simultaneously, if
they are the primary nodes of two instances with the same secondary
node. The second condition requires knowledge of all nodes, not only
those the graph is to be constructed on.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Thomas Thrainer <thomasth@google.com>

Add a new option to hroller to only output information about the first
reboot group. Together with the option --node-tags this allows for the
following work flow. First tag all nodes; then repeatedly compute the
first node group, handle these nodes and remove the tags. In between
these steps, other operations can be carried out on the cluster.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Thomas Thrainer <thomasth@google.com>

Make hroller output the node groups not containing the master node
sorted by size, largest group first. The master node still remains
the last node of the last reboot group. In this way, most progress
is made when switching back to normal cluster operations after the
first reboot group.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Thomas Thrainer <thomasth@google.com>

Regular expressions are not shell globs. So "any symbol" is expressed
by a dot, not a question mark. In this case, the confusion lead to a
too liberal expectation, hence the test passed. Fix it nevertheless.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

The order of evaluation of the conditions is changed, so it's easier to add
more (foreseen) checks for exclusive_storage.

Signed-off-by: Bernardo Dal Seno <bdalseno@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Prereq checks relative disks are grouped together and moved in a separate
method. This reduces the clutter in CheckPrereq().

Signed-off-by: Bernardo Dal Seno <bdalseno@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

The example is a cluster of 6 nodes, paired into 3 group by three
instances. So the whole cluster would need two reboot groups. The two
tags select, in two different ways, one node of each group. So, when
restricting to one tag, a single reboot group suffices, but no
coloring of the whole cluster would achieve this.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Add option --node-tags to tell hroller to consider only nodes
with these tags. A use case would be a tag tracking on which
nodes the maintenance has not yet been carried out, e.g., if
rolling reboots are interleaved with other cluster operations.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Since the htools representation of a node now allows adding
the node tags, populate this field correctly in the Rapi
backend.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Since the htools representation of a node now allows adding
the node tags, populate this field correctly in the LUXI
backend.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

In order to allow htools to make use of node tags, add them to the
text format. This is done by adding a new column at the end of the
node lines. If this column is missing, the default value (which
is the empty list) is left unchanged, thus yielding the current
behavior.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Since hroller (and probably other tools in the future) will support
node selection based on node tags, extend the node data structure to
allow adding this information.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Hroller used to first compute a coloring of the node graph and then
filter out the nodes that it had to work on. While the only filtering
was according to node groups this did not make a difference, as there
shouldn't be any instance with primary and secondary node on different
node groups. With more elaborate filtering, however, reducing the graph
first can lead to better reboot groups.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Change the behavior of mkNodeGraph to tacitly ignore all instances
where one of the nodes is not in the list of nodes. In this way, we
can construct sub-graphs by filtering the nodes and ignoring any
possibly added isolated nodes for the missing indexes.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

In hroller, the option -O can be used to mark certain nodes as offline.
These nodes should then not be part of any reboot group. Add tests
to verify this behavior.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

In commit 7dbe4c72 the new option --force was introduced to
hroller. Change the man page to reflect this change.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Mention the main features of DRBD 8.4 support in the NEWS file.

Signed-off-by: Thomas Thrainer <thomasth@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

In DRBD 8.4, Ganeti does no longer use indexed external meta data but
flexible external metadata. Therefore, there is no longer a meta data
index in the output of `drbdsetup show`.

Signed-off-by: Thomas Thrainer <thomasth@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

This corrects the storage types design doc regarding the specification
of the RPC call.

Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Bernardo dal Seno <bdalseno@google.com>

Instead of computing a coloring for one condition first and then refining
it for the other condition, we can construct a graph with edges for all
conditions that prevent simultaneous reboots. This will not only result
in simpler code, but might also lead to better colorings.

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Starting from Debian Wheezy, /var/run is a symlink to /run. If rsync
does not treat receiving symlinks to directories as directories, it
would remove the original symlink and create a new, empty directory
where only the files to sync are put in. In the case of /var/run, this
behavior severly impacts the running system (as, for example, several
PID can't be found anymore).

Signed-off-by: Thomas Thrainer <thomasth@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

This check is performed during gnt-cluster verify and outputs warnings
if at least two different DRBD versions are found within a node group.
In such a case, all nodes with their installed DRBD version are
displayed.

Signed-off-by: Thomas Thrainer <thomasth@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

All functionality specific to a single DRBD8 devide is now in DRBD8Dev,
whereas functionality which is valid for the whole DRBD "installation"
on a device is collected in DRBD8.

This makes it possible to remove a couple of FIXME's and clarifies the
code in some areas.

Signed-off-by: Thomas Thrainer <thomasth@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Right now the DRBD8 class has multiple responsibilities: a) it
reprensents a device which can be set up, grown, etc. and b) it
represents the whole DRBD system on a node which has a usermode helper,
which knows how to shut down _all_ devices, etc. Therefore, the DRBD8Dev
class will only be responsible for a device, whereas the soon to be
extraced DRBD8 class is responsible for the DRBD system.

Signed-off-by: Thomas Thrainer <thomasth@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

gnt-cluster verify should issue a warning if there are multiple DRBD
versions present in a node group. In order to do so, the DRBD version
has to be queryable from noded.

Signed-off-by: Thomas Thrainer <thomasth@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Conflicts:
	lib/bdev.py
	lib/cmdlib.py

Signed-off-by: Michele Tartara <mtartara@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

The way disk templates were displayed in instance policies became confusing
since the introduction of the enabled_disk_templates cluster parameter.

Fix issue 440.

Signed-off-by: Bernardo Dal Seno <bdalseno@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

This makes sure that issue 401 doesn't come back.

Signed-off-by: Bernardo Dal Seno <bdalseno@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Renaming the current parameters dealing with spindles is nice, as their
meaning would be clearer, but that would mean changing a lot of code,
change external interfaces, and renaming internal variables for little
benefit. Also, instance policies would end up containing two related
parameters, one used when exclusive_storage is enabled, and the other when
disabled. Reusing the existing parameters by changing slightly their
semantics makes more sense.

Signed-off-by: Bernardo Dal Seno <bdalseno@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Bullet lists start after a blank line.

Signed-off-by: Bernardo Dal Seno <bdalseno@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Conflicts:
	INSTALL
	NEWS
	configure.ac
	src/Ganeti/Query/Group.hs
	test/hs/Test/Ganeti/Objects.hs
	test/hs/Test/Ganeti/Query/Query.hs

Signed-off-by: Michele Tartara <mtartara@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Curl is not optional anymore. Remove reference to it in the optional libraries
section.

Signed-off-by: Michele Tartara <mtartara@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

DRBD 8.4.3 does not support specifying volume ids directly while
creating the meta device. It requires a slightly different syntax (using
flex-external as volume id) which is fortunately also supported by (at
least) DRBD 8.4.2.

Signed-off-by: Thomas Thrainer <thomasth@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>