This patch changes the iallocator framework to work with and properly
export to plugins offline nodes. It does this by only exporting the
static configuration data for those nodes, and not attempting to parse
the runtime data.

The patch also fixes bugs in iallocator related to the RpcResult
conversion, changes the should_run to admin_up attribute name (as per
the internals change), and adds “-I” as a short option for
“--iallocator” in gnt-instance, gnt-backup and burnin.

Reviewed-by: ultrotter

Currently the DRBD code checks that the metadata devices are valid
before creation, initial disk attachment and add children.

However, the process for checking validity requires a free DRBD minor,
and this conflict with parallel checking.

There are at least three possible solutions:
  - serialize all checks, which means we reduce parallelism and need
    extra locks
  - don't pass a valid minor number, but one like “/dev/drbd256” (which
    is invalid); this works for current version of DRBD, but since it's
    not guaranteed to remain so it doesn't look nice
  - don't do the checking at all, and rely on “drbdsetup ... disk ...”
    to fail by itself

The reason for checking metadata was that in 1.2, this was much cheaper
than trying to activate devices (and the subsequent iteration over the
minors). However, in 2.0, they have the same cost, so we can choose
option 3: just remove the explicit checking and rely on drbdsetup and
the kernel to fail.

Since DRBD8._InitMeta still requires a minor number, the two places
where this is run are handled as follows:
  - Create: we just use our own (unused currently) minor number
  - AddChildren: we keep using FindUnusedMinor, with the caveat that
    this function (used by replace-disks -n ...) cannot be yet
    parallelized

Reviewed-by: ultrotter

This patch changes (significantly) the execution model in burnin:
  - for all runs, (almost) all instance mods in a single Burn* procedure
    are done as part of a job; so for example add disk, stop, remove
    disk, start are no longer done as separate jobs but as a single job
    consisting of four opcodes
  - for parallel runs, all Burn* procedures except the rename (which
    uses a single target name) run in parallel; before, only the
    creation was done in parallel
  - due to the single-job execution and also parallel execution, the
    logging messages are no longer happening synchronously with the
    execution, so they are more informative than an actual execution log

The end result is that burnin now tests properly multi-opcode jobs and
also tests all opcodes (except rename) for parallel execution.

Note: On a test cluster, parallelization reduces burnin time from 23m to
15m.

Reviewed-by: ultrotter

Currently the restrictions are too harsh: there is a time interval
between an instance gets a new disk and before it is added to the
configuration in which the restriction is not met. We solve this by
allowing temporary DRBD minors to match existing minors (for the same
instance), such that parallel creations/minor allocations are OK.

The change is done by moving the add of temporary minors to the
minor map after the instance minors are computed, and only considering
them as duplicate if the instance name doesn't match.

Reviewed-by: ultrotter

This patch enhances the duplicate DRBD minors checks (currently just a
few) and adds automatic checks of configuration consistency at
configuration file writing time.

In order to do so and show meaningful error messages, the
_UnlockedComputeDRBDMap function is changed to not raise errors in case
of duplicates, but instead return both the minors map and the duplicate
list, and its callers now raise the error. This allows the VerifyConfig
function to return a complete list of duplicates.

The new checks required some small updates to the unittests for the
config module.

Reviewed-by: ultrotter

When creating ‘fake’ results for offline nodes, we currently don't pass
the call attribute. This complicates debugging, so even though this
should not matter in practice, it's better to fix it.

Reviewed-by: imsnah

This removes some constraints:
  - only two disks supported, this is no longer true as the underlying
    functions can now compute size for a variable number of disks
  - error when the hypervisor was not being passed
  - typo error

Reviewed-by: imsnah

This is less of an actual issue for regular gnt-* clients, but it's
easily reproducible with burnin and possible with RAPI (depending on how
the program uses luxi.Client(s)).

In case of burnin, if we interrupt the client (^C) while it polls the
job, it will abort and raise an error. After that, burnin issues a
remove instance job, and at this point, we send the submit job (remove)
call but the first thing we read from the socket will be the response to
the previous poll job request, since that was queued already from the
master.

To solve this, whenever we detect an error in Transport.Call(), we close
that transport and re-create a new one, to start anew. The other
alternative would be to introduce a sequence to the protocol, but this
is something that would be design-level change and it's not recommended
at this stage.

Reviewed-by: imsnah

When an instance fails to shut down we currently log its whole object,
rather than just the instance name.

Reviewed-by: iustinp

If the KVM live migration ends up in a 'failed' state it has been
aborted at the kvm level, and the machine is still running locally.
We support also the 'cancelled' state even though there should be no way
of reaching it, without manual intervention.

Reviewed-by: iustinp

Reviewed-by: iustinp

The tcp port used for migrating KVM instances is selectable at
./configure time. We use a single port as nodes are locked anyway during
a migration, so no two migrations can happen at the same time to the
same node.

Reviewed-by: iustinp

Throughout the kvm code we very often look for the instance pidfile
name, read it, and check if the process is alive. Abstract this into a
private function and use that one instead.

This patch also changes RebootInstance to check whether the instance is
alive before trying to reboot it.

Reviewed-by: iustinp

RebootInstance was broken, because it just used to call StartInstance
with wrong parameters. With this patch we still stop the instance, but
use the saved kvm runtime to start it again.

Reviewed-by: iustinp

When we ask the instance to shutdown sometimes the command won't work,
especially if the instance isn't fully booted up. We'll wait for a bit,
and give it a few chances before giving up.

Reviewed-by: iustinp

These are used, for the xen hypervisor, to copy the xen config file to
the remote node. This breaks migration for instances which have been
migrated, but not restarted, with the old code, for which the config
file was just lost.

Reviewed-by: iustinp

This patch converts the DRBD minors reservation protocol from explicit
release to automatic release on the success paths. On the errors paths,
it's still needed to manual release.

The patch doesn't bring much by itself, but is needed for a future patch
which enhances the automatic verification of configuration consistency.

Reviewed-by: ultrotter

Two are real errors (invalid names) and one is style error (overriding
name from outer scope).

Reviewed-by: ultrotter

This was forgotten in the recent “switch to explicit ignore rules”.

Reviewed-by: imsnah

Currently the rpc module logs the error description and target node in
rpc calls logging, as such:

  2009-01-21 00:50:01,456:  pid=1051/Thread-21 ERROR RPC error from node
    node1.example.com: Connection failed (111: Connection
    refused)

but this doesn't help to understand which call caused this (here it's an
offline node which should not be contacted at all).

This patch adds the logging of the call too, so cases like the above can
be debugged easier.

Reviewed-by: imsnah, ultrotter

Due to historic reasons, the “should run or not” attribute of an
instance was denoted by its “status” attribute having a string value of
either ‘up’ or ‘down’. Checking this is in code was done via hardcoding
of the strings.

This was long done for a redo, and this patch changes this attribute to
“admin_up” having a boolean value. The patch is in fact shorter than I
expected, and passes burnin.

The patch also fixes an error in BuildInstanceHookEnvByObject where the
instance.os was passed as the status value.

Reviewed-by: ultrotter

MigrationInfo, AcceptInstance and AbortMigration are implemented as
hypervisor specific functions, and by default they do nothing (as
they're not always necessary).

This patch also converts hv_base.MigrateInstance docstring to epydoc,
adds a missing @type to the GetInstanceInfo docstring, and removes an
unneeded empty line.

Reviewed-by: iustinp

At instance startup time we save the kvm runtime, and at stop time we
delete it. This patch also includes a function to load the kvm runtime,
which is unused yet.

Reviewed-by: iustinp

Before we used to generate the kvm command line and then just run it.
With this patch we split the generation from the time it is run,
allowing us to save it and replay it at reboot.

We must take special care about instance nics:
  - We can't include them in the saved command line, as they point to
    temporary files
  - We can't just generate them at exec time, because we would apply
    those changes, but not all the other ones, to a running instance,
    thus making it inconsistent (for example if an instance had a memory
    increased and one more nic, in a soft reboot we would add the nic, but
    not the memory)
So we'll just save the instance nic data at the time the kvm runtime
data is generated, and transform it into actual parameters at execution
time.

Reviewed-by: iustinp

Currently the hypervisor is expected to do all the migration from the
source side. With this patch we also add the option of passing some
information to the target side, and starting some operation there.

As a bonus, a function to cleanup any started operation is included.

Reviewed-by: iustinp

With the addition of minors, this needs to show them too.

Reviewed-by: ultrotter

Currently we keep pid files and control files. In the conf dir we'll
also keep the data to start the instance anew, and the network
interface scripts. These will then be copied to a separate area (since
_CONF_DIR could be mounted 'noexec') and used to start the instance.

This patch also adds comments to state what the various directories are
used for.

Reviewed-by: iustinp

Abstract the monitor and serial socket naming in two functions, and
reuse them to cleanup the files after shutdown.

Reviewed-by: iustinp

Reviewed-by: iustinp

Reviewed-by: iustinp

When StopInstance raises an HypervisorError, report it in the logged
message to ease with debugging.

Reviewed-by: iustinp

Reviewed-by: iustinp

Reviewed-by: iustinp

(this is related to the master daemon log)

Currently it's not possible to follow (in the non-debug runs) the
logical execution thread of jobs. This is due to the fact that we don't
log the thread name (so we lose the association of log messages to jobs)
and we don't log the start/stop of job and opcode execution.

This patch adds a new parameter to utils.SetupLogging that enables
thread name logging, and promotes some log entries from debug to info.
With this applied, it's easier to understand which log messages relate
to which jobs/opcodes.

The patch also moves the "INFO client closed connection" entry to debug
level, since it's not a very informative log entry.

Reviewed-by: ultrotter

This way newly added files will be not be excluded by default. Fixes
also a small whitespace error in utils.py.

Reviewed-by: iustinp

This allows the rename failures to show the ouput of OS scripts.

Reviewed-by: ultrotter

As per Michael's comment, gitignore should not ignore a couple of real
files from the autotools/ directory.

Reviewed-by: ultrotter

The ConfigWriter.AllocateDRBDMinor requires the instance name, not the
instance object. The LUSetInstanceParms is passing wrongly the instance
object, which can cause breakage.

The patch also adds asserts to check for this mismatch in ConfigWriter.

Reviewed-by: ultrotter

The urllib2 module has very bad error handling. This patch changes to urllib
which is simpler, and we derive a custom class from the FancyURLopener. Burning
is no longer keeping sockets in CLOSE_WAIT state with this patch.

Reviewed-by: ultrotter

This patch adds support for verification of drbd minors space in cluster
verify: minors which belong to running instances and should be online
but are not, and minors which do not belong to any instace but are in
use.

The patch requires exposing some methods from bdev.DRBD8 and
config.ConfigWriter which were until now private methods.

Reviewed-by: ultrotter