GitLab

The RunWithLocks test assumed that gnt-debug delay would have the info
needed for interruption ready immediately after being run, and in some
situations this is not the case. This patch makes the test more patient
and forgiving by retrying a number of times.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Upon detection of a stale livelock, make WConfD also
try to clean it up. We can safely ignore any failures,
as the Ganeti resources are already freed, and the worst
thing that can happen is an empty file left over in the
file system.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

.. which will be needed soon for WConfd.

While at it, also ensure we have the required version of
base64-bytestring.
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

.. as the newest version of hlint adds additional checks
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Since the default GHC6 isn't usable for modern libraries and starting
from wheezy there is GHC7 anyway, download and install the current
vanilla version as well as the latest Cabal.

Use the current up-to-date libraries.

Also enhance the check for test-framework so that it works for the
recent version.
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

This speeds down transferring them to buildbots, at a negligible size
cost.
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

As jobs are currently running in masterd, use the masterd livelock
file.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

As livelock files are constructed in a systematic manner,
we can guess what the livelock file for a given owner is.
While this will not necessarily work perfectly, it will be
useful to simplify direct debugging of WConfD.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

...so that it can request resources from WConfd.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

...as this function sets up a much richer context than just
the job queue, including the current lock management.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

To request resources from WConfD, requesters have to provide
the name of a file they own an exclusive lock on. In this way,
their death can be detected. Add utility functions to obtain
such a file name.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

When requesting resources from WConfD, a file has to be
presented where an exclusive lock is owned on, so that
WConfD can detect when the requester dies. Add a path
to a directory where these files are kept in.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Add the cleanup task for locks of died owners to WConfD,
as a separate thread.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Provide a task that periodically checks for all locks
owners whether they are still alive. If not, it cleans
up their locks.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Add a constant for the time interval at which WConfD checks for
resource owners that have died and cleans up their resources.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Add a test verifying that the list of lock owners is not
too big in the sense that every subject contained in that
list owns at least one lock.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

In other words, verify that if all the lock owners release their
locks, the empty lock allocation is reached.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

To allow for clean up, add a function that provides the list
of all the owners holding at least one lock.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

With jobs running in processes different from the lock management soon,
we need to detect if a job holding locks dies, in order to release the
resources.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

* stable-2.11
  Convert int to float when checking config. consistency
  Updateing release date to tomorrow
  Revision bump for the 2.11.0~beta1 release
  Set release date of 2.11 beta1
  Stop watcher from restarting down instances during an opcode

Conflicts:
	NEWS: take both additions
	lib/watcher/__init__.py: manually apply the client
	    change to the version on 2.11

Semantical conflicts:
	configure.ac: undo suffix bump
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

When reading the configuration file from RPC JSON, values without a
floating point are parsed as 'int', not as 'float', and later the
consistency check fails.

This patch adds an automatic conversion from 'int' to 'float' during
checking so that the improper JSON parsing is fixed.
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Cherry-picked from 710a2863

 to fix
downgrades to 2.11.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

To compensate for the cases where a QA test is supposed to block when
a lock is present, add an additional option showing whether blocking is
supposed to happen or not.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

This patch adds threading to the RunWithTests function, allowing one
thread to execute the QA test, and the other to monitor if it is being
blocked by locks set up during the test. If it is, terminate the
blocking job, and let the QA continue, reporting the test failure at
the very end.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

The command logging performed by GetCommandOutput can be distracting
and unwelcome. This patch adds a switch allowing the command output to
be muted as needed.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

The SSH multiplexer is used by default in every connection made, and
when multiple threads are used, this causes strange issues when running
the QA locally. This patch allows the multiplexer to be bypassed,
making a new connection every time.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

This patch adds a QA utility function that acquires a set of locks, and
attempts to run a given function with the locks in place. Should the
given function block, this function does not detect this - later
patches will address the issue.

An example of its use is provided by having the move-instance test
modified to use it.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

This patch sorts the imports of qa_rapi.py to make things a little bit
more orderly.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

This patch aligns the timestamps output as a part of gnt-job info, and
performs minor refactorings in the process.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Aligning dictionary entries makes no difference to a YAML parser, but
makes the output much easier to read and compare. This patch adds the
possibility of specifying alignment groups to ordered dictionary
entries.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

This patch changes gnt-job info to use standard functions defined in
cli.py, and output valid YAML.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

The PrintGenericInfo function in cli.py did not handle tuples as
containers of items, making it impossible for these to be deserialized
automatically when a YAML parser is used. This patch adds separate
handling of tuples, including inlining them for readability when
possible.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

The move-instance QA test will have to be changed in the following
patches to allow testing opportunistic locking.

This patch retains the same functionality as before, but allows an
iallocator to be used, and splits the move back and forth into two
invocations, the first of which will be wrapped later on.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

The gnt-debug delay command could be useful as a means of acquiring
locks for testing purposes. In practice, to be useful it should be
interruptible, otherwise we risk race conditions or long delays.

This patch follows the examples of the move-instance command and the
gnt-debug test-jobqueue commands, and introduces a mechanism for
communicating with the delay command, allowing it to be interrupted
early.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Following the example of the job-queue test, the delay also defines a
log message type of its own.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

This patch allows the opcode option to be used through the gnt-debug
client.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

To prepare for the use of sockets to interrupt a running delay opcode,
this patch adds an option to the opcode for doing so.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

As the delay class will also have to start using domain sockets,
extract the functionality into a helper class.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>