Currently the iallocator execution takes place in the master, which is a
violation of the current architecture, and will create problems with a
threaded master daemon.

This patch moves the execution to the backend, similar to the hooks
runner, by:
  - introducing a new class that handles the execution in the backend
    (and could be used also for listing the allocators, etc.)
  - introducing a new rpc call
  - replacing the actual execution in IAllocator.Run() with a rpc call

This passes burnin with the dumb allocator

Reviewed-by: imsnah

This patch changes the definition of a job and introduces per-opcode
results.

First, the result and status fields of a job are condensed into a single
'status' attribute. Then, we introduce an opcode status and one result
list, that allow jobs to return values.

The gnt-job script is also modified to allow these new fields to be
queried.

Note that the patch changes the opcode field to op_list, and it changes
its return value from string to a list of (serialized) opcodes.

Reviewed-by: ultrotter

This patch adds checks for the master role and daemonize support to
ganeti-masterd.

The patch modifies the startup/shutdown of the server because:
  - we want bind()/listen() to the master socket to occur before forking
    so that we can return a correct exit code and write messages to
    stderr
  - but we want thread startup to occur after fork(), otherwise python
    threading gets confused

The patch also has some small cleanups:
  - remove the unix socket after closing it, so we don't need to remove
    it manually
  - instead of just telling the threads to terminate via the new_queue,
    we also join() them so that the logs show what thread clinging to
    life
  - the daemon logs to its own logfile now
  - there is command line parameter support :)

Reviewed-by: imsnah

Reviewed-by: ultrotter

This patch adds a very basic gnt-job script that allows job querying.
This goes on top of the previous master daemon patches.

Currently, because of the not-changed cmd lock, you can't query the jobs
as long as a job is running - you have to rm the cmd lock and then you
can query the jobs.

Reviewed-by: imsnah

Currently, in ganeti-noded we have the createDaemon function. Since
we'll need the same in other daemons, we move this function to utils.py

With the move, a few changes were also done:
  - change the name to Daemonize()
  - add a parameter, logfile, as different daemons will want to log to
    different files
  - remove the try.. except.. around the fork calls, since they were
    only re-raising the OS exception with less data; unless we want to
    actually handle fork error (not just re-raising), these try blocks
    are not useful
  - change the return style at the end of the function

Reviewed-by: imsnah

This patch adds a very in-progress master daemon. This needs to be
launched manually, does not background itself, but can be used for
opcode execution.

Also parts of this code should be moved to luxi.py.

Reviewed-by: ultrotter

This patch just removes an extraneous \n from the log message making it
nicer to view.

Reviewed-by: schreiberal

Currently in order to deal with tmpfs /var/run, we create the
BDEV_CACHE_DIR in the init script. However, that does not cover all the
cases, and it's not a proper place to deal with it: for example, dealing
with not initialized clusters and the master node is more complicated.

Therefore, this patch does:
  - make ganeti-noded create the directory automatically
  - make ganeti-noded error out if it can't create it or it's already
    there but not a directory
  - remove the creation from the init.d script

Reviewed-by: ultrotter

This patch modifies TcpPing and its callers to make the source address
selection optional. Usually, the kernel will know better what
source address to use, just in some cases we want to enforce a given
source address so it makes sense to make this optional.

Reviewed-by: ultrotter

This patch switches from the twisted usage for inter-node protocol to
simple BaseHTTPServer/httplib. The patch has more deletions because we
use no authentication, no encryption at all.

As such, this is just for trunk, and only for testing. What it brings is
the ability to use the rpc library from within multiple threads in
parallel (or it should so).

Since the changes are very few and non-intrusive, they can be reverted
without impacting the rest of the code.

This passes burnin. QA was not tested.

Reviewed-by: imsnah

This can be used for testing purposes.

Reviewed-by: ultrotter,imsnah

This patch modifies the watcher to run the ‘gnt-cluster verify-disks’
command and to log its output (if any).

Reviewed-by: imsnah

- When line wrapping is needed, move spaces to the next line.
- Remove embedded line breaks from error messages.

Reviewed-by: schreiberal

This patch adds logging of command failures to the debug log in case the
user either started the command (gnt-*) or the node daemon with the
debug flag.

Reviewed-by: imsnah

- Use constants for keys.
- Fix bug through which automatic instance restarts wouldn't be limited

Reviewed-by: iustinp

In order to do this for simplicity we leave the OSFromDisk function as-is and
we convert the eventual exception to an OS object in ganeti-noded. The
unmangling gets simplified and so does the code for checking whether the OS is
valid.

Reviewed-By: iustinp

The functions in ganeti-noded and rpc.py still deal with the fact that an
InvalidOS error could be returned by DiagnoseOS. As this is not the case
anymore simplify their code for the current behavior.

Reviewed-By: iustinp

Currently, troubleshooting DRBD problems involves a manual process of going
backwards from the DRBD device to the instance that owns it.

This patch adds a weak (i.e. not guaranteed to be correct or up-to-date)
cache of device to instance. The cache should be, in normal operation,
having correct information as the only time when devices change paths
are when they are started/stopped, and the code in backend.py adds cache
updates to exactly these operations.

The only drawback of this implementation is that we don't fully update
the cache on renames of devices (we clean the old entries but we don't
add new ones). Since the rename changes the path only for LVs (and not
drbd and md), this is less of a problem as the target of this code is
debugging DRBD and MD issues.

The patch writes files named bdev_drbd<N> (or bdev_md<N>,
bdev_xenvg_...) in /var/run/ganeti (more exactly, LOCALSTATEDIR/ganeti).
The files start with 'bdev_' and continue with the path of the device
under /dev/ (this prefix stripped), and contain the following values,
space separated:
  - instance name
  - primary or secondary (depending on how the device is on the primary
    or secondary node)
  - instance visible name: sda or sdb or not_visible, the latter case
    when the device is not the top-level device (i.e. remote_raid1
    templates will have sd[ab] for the md, but not_visible for drbd and
    logical volumes)

The cache is designed to not raise any errors, if there is an I/O error
it will only be logged in the node daemon log file. This is in order to
reduce the possible impact of the cache on the block device activation
and shutdown code.

Reviewed-by: imsnah

This patch add code for renaming a device; more precisely, for changing
the unique_id of the device. This means:
  - logical volumes, rename the volume
  - drbd8, change the remote peer

This is needed for the being able to replace disks for drbd8.

Reviewed-by: imsnah

The two calls mirror_addchild and mirror_removechild take only one child
for addition/removal. While this is enough for our md usage, for local
disk replacement in drbd8, we need to be able to specify both the data
and metadata device. This patch modifies these two rpc calls (and their
backend implementation and their usage in cmdlib) to take a list of
children to add/remove.

Reviewed-by: imsnah

This patch series implements the reboot command for gnt-instance. It
supports three types of reboot: soft (hypervisor reboot), hard (instance
config rebuild and reboot) and full (full instance shutdown and startup
again).

This patch contains the backend and rpc part of the patch.


Reviewed-by: iustinp

The creation of the log file for the node daemon lacks the mode
parameter, so after applying the current umask, the file got 0700
permissions. Restrict this to the correct 0600.

Reviewed-by: schreiberal

This patch completely  gets rid of fping
 - replace all fping invocations with TcpPing calls
 - update documentation accordingly.
 - associated cleanups (use constant for localhost IP, use more sensible
   defaults for TcpPing and _use_ those)

Reviewed-by: iustinp

Reviewed-by: iustinp

- Change format of watcher state file to JSON.
- Move log path for watcher script to constants.py.

Reviewed-by: iustinp

- Add NEWS file with major changes between versions.
- Bump RPC version number
- No longer serialize in RPC, but just convert to dict

Old Pickle based configuration files can be converted using the cfgupgrade
utility.

Reviewed-by: iustinp, ultrotter

We currently require that hostnames are FQDN not short names
(node1.example.com instead of node1). We can allow short names as long
as:
  - we always resolve the names as returned by socket.gethostname()
  - we rely on having a working resolver

These issues are not as big as may seem, as we only did gethostname() in
a few places in order to check for the master; we already required
working resolver all over the code for the other nodes names (and thus
requiring the same for the current node name is normal).  The patch
moves some resolver calls from within execution path to the checking
path (which can abort without any problems). It is important that after
this patch is applied, no name resolving is called from the execution
path (LU.Exec() or other code that is called from within those methods)
as in this case we get much better code flow.

This patch also changes the functions for doing name lookups and
encapsulates all functionality in a single class.

The final change is that, by requiring working resolver at all times, we
can change the 'return None' into an exception and thus we don't have to
check manually each time; only some special cases will check
(ganeti-daemon and ganeti-watcher which are not covered by the
generalized exception handling in cli.py). The code is cleaner this way.

Reviewed-by: imsnah

The EXIT_NODESETUP_ERROR is a useful constant and ganeti-watcher could
use it too. This patch moves it to constants.py and modifed the
ganeti-master script to use it from there.

Reviewed-by: imsnah

This patch adds support for instance rename operation at all remaining
layers: RPC, OpCode/LU and CLI.

Reviewed-by: imsnah

This was forgotten when the init script was changed.

Reviewed-by: imsnah

This changes the raising of exceptions from:
  raise Exception, value
to
  raise Exception(value)

as the first form will be removed in python-3000 and the second form is
preferred now.

The changes also involve a few cases of changing from raising standard
exceptions and use our own ones.

The new version also fixes many pylint-generated warnings, especially in
ganeti-noded where I changed many methods to @staticmethod.

There is no functionality changed (barring any bugs).

Reviewed-by: iustinp

Reviewed-by: iustinp

handling, as it can be static and outside of ganeti.

This also means we can get rid of a lot of infrastructure too:
  - the master/node config files checkers
  - one rpc function

Reviewed-by: imsnah

Reviewed-by: iustinp

- Create all --output options using a constant
- Put node checking code from opcodes into a single function
- Do the same for output fields

Reviewed-by: iustinp

  - move the master node name from the ConfigWriter to SimpleStore (all
    nodes need this, and it was the only thing pulled in from the
    ConfigWriter on nodes)
  - fix mcpu.py and the testing w.r.t. this change; for testing, rename
    the fake_config.py to mocks.py and add a FakeSStore object
  - then add a ganeti-master script which can be run on any node at boot
    and which will not do anything if not master on start (on stop it
    will still try to remove the ip address)
  - also add a new cluster-wide variable (master_netdev) that determines
    on which interface we add this ip address; it's customizable at
    cluster init time
  - also remove the cluster name file which was separately handled from
    ssconf (not needed anymore)
  - remove the master init.d links from the list of config files as this
    is not our responsibility now

Reviewed-by: iustinp