GitLab

This (big) patch reworks the master startup/shutdown and the fixes the
master failover.

What does the patch do?

For master start/stop:
  - remove the old ganeti-master script and its associated man page
  - moves the ip start/stop directly into the backend.(Start|Stop)Master
  - adds start/stop of the master/rapi daemon into these functions,
    selectively based on the start/stop arguments
  - makes the master call via rpc StartMaster(start_daemons=False) to
    the local node so that the master IP is started
  - and finally changes the example init.d script to directly start and
    stop all three daemons, since they do the right thing (depending on
    master/not master role)

For master failover:
  - moves the code from LUMasterFailover into bootstrap.MasterFailover,
    since we need to start/stop the master during this operation and
    thus it can't be executed from the master
  - removes the LUMasterFailover and its associated opcode

Notes: ubuntu's /etc/lsb-base-logging.sh is dumb, so the messages 'not
master' are not seen during startup on non-master nodes.

Reviewed-by: ultrotter

Since we need to compute this from outside utils.py, we change this to a
public function.

Reviewed-by: ultrotter

This patch moves the CheckMaster function from ganeti-masterd to ssconf
(most logical place, it cannot go in utils since we would have recursive
imports between ssconf and utils) and changes ganeti-rapi to also call
this function.

This is needed so that starting ganeti-rapi on a non-master node does
the right thing.

Reviewed-by: ultrotter

This patch adds a new, unused for now, parameter to the start and stop
master operations in backend. The idea behind it is that we need to be
able to control whether the IP (de)activation is coupled with daemon
startup/shutdown.

The callers are also modified to pass this parameter (even if unused for
now).

Reviewed-by: ultrotter

Reviewed-by: iustinp

The passed parameters were not correct.

Reviewed-by: iustinp, ultrotter

Reviewed-by: imsnah

Reviewed-by: imsnah

We cannot depend on all environments to have a start-stop-daemon or
similar tool. We instead implement a KillProcess function that behaves
similar to “start-stop-daemon --retry”.

Note that the attached unittest can hang in foreground if the child
misbehaves (doesn't write to the internal pipe). Since unittest are
either run in the foreground or are run with a timeout from an automated
framework, I think this is an acceptable trade-off (against of using
hardcoded timeouts in the test).

Reviewed-by: imsnah

We already have a function to test if a PID is alive, so it makes more
sense to use function composition that force calling (since we need to
read PIDs from files in other places too). Now IsProcessAlive returns
False for PIDs <= 0, since this is the error return from ReadPidFile.

The patch also adds a unittest for checking that WriteFile raises the
correct exception, and checks that an invalid or missing file causes
ReadPidFile to return zero. The unittest tearDown method will try to
cleanup the temp directory too (otherwise it leaves stuff after it).

Reviewed-by: ultrotter

All other daemons have their main code in themselves and not in a module.
This patch does the same to ganeti-rapi by moving the code from
lib/rapi/RESTHTTPServer.py to daemons/ganeti-rapi.

Reviewed-by: iustinp

The generic HTTP server doesn't know about httperror based exceptions
and would treat them as unknown exceptions, thereby not doing the right
thing with HTTP errors.

Reviewed-by: iustinp

Locking is not completeley right due to a deadlock when the job calls
UpdateJob after changing its status.

Reviewed-by: ultrotter

Reviewed-by: iustinp

Reviewed-by: ultrotter

It might come in handy at some point and makes the code a bit easier
to read.

Reviewed-by: iustinp

The set triggers exception on a list-tags command and RAPI calls for tags
since it is not serializable by JSON.

Reviewed-by: iustinp

Reviewed-by: imsnah

So far no error reporting to the client is done. Clients don't get
noticed if a job doesn't exist or couldn't be archived because of
its current status.

The internal cache is always cleaned when the preconditions didn't
fail to make sure that the actual disk status will be reread next
time.

Reviewed-by: iustinp

Reviewed-by: iustinp

A later patch will add a memory based job storage class, hence this
code is going into a separate class. It also changes the number format
to always use at least 10 digits, allowing up to 9'999'999'999 jobs to
be sorted without using a custom function.

Reviewed-by: iustinp

WritePidFile is a helper function that writes the current pid in a
pidfile within the ganeti run directory. RemovePidFile tries to delete
it.

Reviewed-by: iustinp

This helper function reads a pid from a file containing it and checks
whether it refers to a live process.

Reviewed-by: iustinp

An implementation mistake from the original design caused nodes to be
locked before instances, rather than after. This patch inverts the level
numbering, changing also the relevant unittests and the recursive
locking function starting point.

Reviewed-by: iustinp

Reviewed-by: iustinp

Reviewed-by: iustinp

The job ID is now a string, hence logging must use %s instead of %d.

Reviewed-by: iustinp

We can use zip for simplifying this function. Actually, at this point
I'm not sure if it needs to be a separate function at all.

Reviewed-by: imsnah

The docstring says that _NewSerialUnlocked returns “a string
representing the job identifier”. Until now it returned an
integer and this patch changes it.

Reviewed-by: iustinp

This patch adds distribution of the queue serial file after each write
to it (but before a new job is created and written with that ID, and
before a response is returned, so we should be safe from crashes in
between).

Currently it only logs if a node cannot be contacted, it should abort if
> 50% errors are seen.

Reviewed-by: imsnah

This will be needed for master failover. If we don't have a valid queue
directory, we need to reinitialize it, but we should keep the existing
serial number.

As such, we abstract the reading of the serial and if we find a valid
serial, we do not reset it.

Reviewed-by: imsnah

This was a TODO for 2.0

Reviewed-by: iustinp

Grab a lock for the instance we're working on, and update its params.

Reviewed-by: iustinp

When we set the instance params we're not adding a new instance, but
just updating an existing one, so why using AddInstance?

Reviewed-by: iustinp

For ConnectConsole we just need to lock the instance we're connecting
to. We make a few rpcs to its primary node, but node daemons can now
handle multiple queries and nodes cannot be removed till they have
instances on them anyway. Note that since we return the ssh command, and
that's executed outside of the ganeti daemon, without any locks held,
the instance can then be subject to operations while we're connected to
it, but that was the previous behavior as well.

Reviewed-by: iustinp

LUs that take an instance name as input and need to expand its name and
lock it can use it to simplify their ExpandNames call. Possibly, and
_ExpandAndLockNode will come as well.

Reviewed-by: iustinp

LUQueryClusterInfo and LUDumpClusterConfig can be made concurrent and
don't need to acquire any locks. In fact they don't interact with the
cluster at all, but just with its configuration, which is thread-safe by
design.

Reviewed-by: iustinp

Two top level definitions were separated only by one empty line.
Fixing this.

Reviewed-by: imsnah

Reviewed-by: imsnah

Not passing the argument means it has the value None. Iterating None
doesn't work:
  >>> "123" in None
  Traceback (most recent call last):
    File "<stdin>", line 1, in ?
  TypeError: iterable argument required

Hence I rename it to "exclude" instead of "exceptions", which may be
confusing, and make it mandatory. If one wants to clean all cache
entries, an empty list can be passed.

Reviewed-by: iustinp