GitLab

In this way we avoid a circular dependency: tryAlloc depends on
the notion of global N+1 redundancy and global N+1 redundancy for
plain instances depends on knowing how to allocate an instance on
a single node.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Some of the hspace tests where designed for a setting where N+1 redundancy
should only be taken into account for DRBD. So add the --no-capacity-check
option in order for hspace to continue to show this behavior in the tests.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

A new option --no-capacity-checks was added to hspace to tell
it to restrict N+1 redundancy checks to only DRBD. Describe it
in the man page.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

While it is generally a good idea to plan capacity to have a redundant
cluster, there are legitimate use cases, where we want to know how many
instances can be squeezed on the cluster at all, taking only N+1 redundancy
for DRBD into account. So add an option restoring the old behavior.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Some old tests were designed without global N+1 redundancy for
plain instances in mind. So, to make these tests compatible with
the upcoming changes, add the (already present) option --no-capacity-checks.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

While we cannot avoid data loss on node crashs if we
have plain instances, we can ensure that the cluster
has enough capacity to reinstall the instances on a
new node. Add a design describing how we enusre this.
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: Hrvoje Ribicic <riba@google.com>

...as they now do not behave differently any more.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

The change itself is entered atomically by WConfD. A @ConfigSync also
cannot help when synchronizing with other jobs, as the only thing that
this annotations reads before and writes after is the configuration---which
does not include the state of temporary reservations.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Signed-off-by: Lisa Velden <velden@google.com>
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

It is a frequent request to be able to specify desired locations
for particular instances. Describe a tag-based way of achieving
this without adding hard enforcements.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

.. so that we can build against it.
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

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

We'll later use it to a GHC 7.8 variant on Jessie
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Since GHC 7.8 installation comes in two variants, depending on the
version of libgmp, add the variant as an environment variable.
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

.. so that it matches the version of cabal-install.

If the versions don't match, the cabal part of the Makefile fails.
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

.. to the release so that in future different releases can have
different versions.
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Otherwise it pulls the recent version (1.0.1), which fails to build
against the other libraries.
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Instead of always allocating 20 instances at a time, make
a conservative estimation of the number of machines that
definitely should fit on the cluster without violating
global N+1 redundancy. Only if this yields a too small
number fall back to fixed step size.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

In the current computation of `iterateAlloc` most of the time
is spent in checking whether the obtained situation is globally
N+1 redundant. However, actually finding a non N+1 redundant
situation only happens at the very end in a long sequence of iterations
in an hspace-like use case. Therefore, we allocate a group of instances
in one block without checking global N+1 redundancy; if that happens to
yield a globally N+1 redundant situation, we can continue from there.
Only when that heuristics fails, we do allocation step by step checking
global N+1 redundancy in each allocation step.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Add a predicate telling if a list of nodes is globally N+1
redundant. It is, if each node is.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

* stable-2.15
  Restrict evacuation to a group
  Staying is not an option in evacuation
  Set release date of 2.15.0 rc1 to tomorrow
  Bump revision number to 2.15.0 rc1
  Adapt NEWS file for 2.15.0 rc1
  Set up Transport properly when communicating with metad
  Allow the Transport class to be used for non-master nodes

* stable-2.14
  Make htools tolerate missing "dtotal" and "dfree" on luxi

* stable-2.13
  Mention migration change in NEWS
  Move misplaced NEWS entry
  Add protection against daemons that may already be listening
  Attempt to cleanup failed migrations using a pidfile
  Add utility that gets the full command line of a process
  Introduce socat as a way of doing xl migrations
  fix typo in UPGRADE notes
  During start, only update job file locally

* stable-2.12
  Fix tests for setting (shared) file storage directory
  Add missing call for setting shared file storage directory
  Update ganeti-luxid synopsis
  Update ganeti-mond synopsis
  Update ganeti-confd synopsis
  Update copyright statement

Conflicts:
	NEWS: take all additions
	configure.ac: ignore suffix bump
	src/Ganeti/HTools/GlobalN1.hs: trivial
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Hrvoje Ribicic <riba@google.com>

* stable-2.14
  Make htools tolerate missing "dtotal" and "dfree" on luxi
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

N+1 redundancy was always a per-group property.
Do the same with shared-storage N+1 redundancy, i.e.,
restrict the escape candidates to nodes of the same
group.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

When checking whether a node can be evacuated, do not
consider the node itself as an escape location.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

If a cluster allows sharedfile as only disk template, the amount of
total and free disk space might not be available. This is perfectly
normal, hence make the luxi backend handle it gracefully and just report
0 available disk on 0 total disk.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

A new option was added to gnt-cluster verify-disks to restrict
verification to a single group only. Describe it in the man
page as well.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

And only act on that group instead of on all groups,
if the parameter is given.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

...to allow users to only verify the disks in a
certain node group.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

The corresponding optional values in python are encoded in
the poor-man's-Maybe way, i.e., null for Nothing and unmodified
for Just.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Added new wconfd RPC SetInstanceStatus. _SetInstanceStatus
is now using this RPC.

_SetInstanceStatus is used by many config mod functions.
These functions no longer use config lock.

Moved the original implementation of _SetInstanceStatus
to config_mock.
Signed-off-by: BSRK Aditya <bsrk@google.com>
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Add wconfd RPC AddTcpUdpPort. This rpc is
now being used by config mod function AddTcpUdpPort.

Removed config lock from AddTcpUdpPort.

Added a basic AddTcpUdpPort in config_mock
for testing purposes.
Signed-off-by: BSRK Aditya <bsrk@google.com>
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

.. to prepare for the new release.
Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

While probably not widely used by third-party tools,
the IAllocator protocol still is an external interface
of Ganeti. So mention the extension of the protocol and
describe how users can continue to use their third-party
IAllocator not aware of that extension.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

When converting an instance from plain to DRBD, a secondary
node has to be chosen. Allow using an IAllocator to do this
choice.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

If we change the template of an instance to DRBD, a secondary node
needs to be chosen. Instead of insisting that it be specified explicitly,
also support asking an IAllocator.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

In order to support automatic choice of the secondary node
in case of a disk-template conversion to DRBD, we need to
pass the IAllocator to be used. So extend the OpCode accordingly.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>