Reviewed-by: amishchenko

Quoting Michael: "why is this even working?"

Reviewed-by: imsnah,amishchenko

The default value of 'False' wasn't initialized properly. It doesn't
require initialization, but it's cleaner this way.

Reviewed-by: amishchenko

Reviewed-by: amishchenko

Two problems: the iallocator.hypervisor wasn't initialized to None in
the constructor, so pylint doesn't realize it's initialized later with
setattr.

Second, 'hypervisor' is a module, so we shouldn't use it as a variable.

Reviewed-by: amishchenko

And a small whitespace fix.

Reviewed-by: amishchenko

hypervisor is a module, so we shouldn't use it as an argument.

Reviewed-by: amishchenko

A few cases of wrong variable name.

Reviewed-by: amishchenko

This is a leftover from the abstraction of AdjustCandidatePool, and it
also requires the config lock, so it's better to remove it.

Reviewed-by: amishchenko

Reviewed-by: amishchenko

For safety, we use the same algorithm as in disk count.

Reviewed-by: amishchenko

_RecursiveFindBD takes a parameter that isn't used; moreover, nowhere in
the SVN history can I find a case that it has been used.

As such, remove this parameter and fix its callers.

Reviewed-by: amishchenko

Reviewed-by: amishchenko

Instead of relying that the usage of the parameter is ok with mutable
default parameters, let's just make it safer..

Reviewed-by: amishchenko

Reviewed-by: amishchenko

Remove unused vars

Reviewed-by: amishchenko

Reviewed-by: imsnah

In the iteration we don't care about the node names, so we change the
for loop to be over the values (and not itervalues).

Reviewed-by: amishchenko

It'll be queried from other classes.

Reviewed-by: iustinp

It's more appropriate.

Reviewed-by: iustinp

This patch should fix all outstanding epydoc parsing errors; as such, we
switch epydoc into verbose mode so that any new errors will be visible.

Reviewed-by: imsnah

epydoc seems to be mightily confused by decorators and how they change
functions (it starts mixing the parameters of the decorated function
into the decorator itself); so we want it to parse only and not look at
the objects themselves.

Reviewed-by: ultrotter

Reviewed-by: iustinp

Reviewed-by: iustinp

This is a leftover from old code.

Reviewed-by: iustinp

This solves a problem with many concurrent requests. By default, 1024
is the maximum backlog on Linux kernels. We limit the number of clients
through MAX_CHILDREN, too. The idea of just increasing the backlog is
taken from lighttpd.

Reviewed-by: amishchenko

Adding compression to larger amounts of data is more efficient than
transferring it (len(nodes) - 1) times over the network without
compression. We were able to compress a 800KB config file to about
30 KB, which is about 40 KB with Base64 encoding (required due to
the way SimpleJson handles strings).

Reviewed-by: ultrotter

The patch also changes the result to error for non-reachable secondary nodes
(as for primary nodes).

Reviewed-by: ultrotter

Currently the ConfigWriter.MaintainCandidatePool returns node names, and
_AdjustCandidatePool uses them as such, but then it passes these to
context.ReaddNode which in turn passes them to jqueue.JobQueue.AddNode which
uses them as objects.Node instances.

Since this is currently the only usage, we change return type from
ConfigWriter.MaintainCandidatePool to be objects and adjust the logging of
their names, so that the auto-adjusement works.

Reviewed-by: ultrotter

This patch changes gnt-node modify and the associated opcode/lu to allow
modification of the node offline attribute.

Setting a node into offline mode automatically demotes it from the
master role.

Reviewed-by: ultrotter

This patch changes the _MultNodeCall and _SingleNodeCall helpers to not
actually make calls to offline nodes, but instead generate fake
responses which have a parameter caller 'offline' set so that callers
can check for this value if they want (otherwise, it's just a failed RPC
call).

Reviewed-by: ultrotter

When an upload is done to a node which doesn't have any version of
ganeti installed, this prevents a non-executable-initd error later in
the upload.

Reviewed-by: imsnah

This patch changes cluster verify to not alert on offline nodes, but
instead just show a note at the end with the number of such nodes.

It also removes warnings in verify-disks and hooks about failures to
make rpc calls to such nodes.

Reviewed-by: ultrotter

This patch adds checks for offline nodes in most instance LUs so that we
can work with offline secondaries, but not with offline primaries. Some
cases (like grow disk, which needs both sides up) are not allowing
offline nodes at all.

Reviewed-by: ultrotter

These will be used for parameter checking and node status checking.

Reviewed-by: ultrotter

Since some nodes can be offline, we can't just take the length of the
node list as the maximum possible number of master candidates.

The patch adds an utility function to correctly compute this value and
replaces hardcoded computations with the use of this function. It then
adds utility functions to automate the maintenance of the node lists.

Reviewed-by: ultrotter

The combination of the current buffer splitting method and (4KB) buffer
size is very inefficient when writing big amounts of data. Just walking
over a 16 megabyte string using a 4K buffer takes (on a random computer)
1m06s, whereas using slices will decrease this to 0.080s, and slicing
with 32 KB size decreases this to 0.073s.

This means that uploading a big config file (it nears 1MB for big
clusters) will take more and more time per the number of nodes, since it
needs lots of slicing.

I happened upon this by accidentally setting all nodes as master
candidates, at which point just uploading the config file to all nodes
took 40s. Applying the patch decreases this to 15s (this probably can
still be optimized).

The patch also removes a duplicate constant (the one actually used is in
http/client.py), and changes the receive buffer size to use the same
constant.

Reviewed-by: imsnah

The patch also changes the various node list generation to be more
consistent.

Reviewed-by: imsnah

This patch adds a simple rpc which makes a backup of the config file and
then removes it. This is done so that cluster verify doesn't complain
immediately after demoting a node.

Reviewed-by: imsnah

This patch changes the LUQueryInstances to show a different state for
offline nodes and also modifies the watcher to understand the offline
state in its checks.

Reviewed-by: ultrotter