GitLab

An overview is available in the design document for this change,
doc/design-chained-jobs.rst.

When a job enters the job processor, the current opcode's dependencies
are evaluated. If a referenced job has not yet reached the desired
status, the current job is registered as a dependant. The job processor
will continue to work on other pending tasks. When a job finishes it
notifies any pending dependants by re-adding them to the workerpool.

A per-job processor lock is necessary for rare cases where the same job
can be re-added twice.

There is no way to view waiting jobs at the moment, but I plan to
export this information to “gnt-debug locks”.

A so-called dependency manager takes care of managing waiting jobs and
keeping track of their status.

Unittests are included.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

They're no longer necessary.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Commit 66bd7445

 added an assertion to ensure a finalized job has its
“end_timestamp” attribute set. Unfortunately it didn't cover a case when
the queue is recovering from an unclean master shutdown.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

This will be useful for assertions. GanetiLockManager._is_owned is
exported, too.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

The wrapper will connect to the console, and check in the background if
the instance is paused, unpausing it as necessary.
Signed-off-by: Stephen Shirley <diamond@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

The wrapper will connect to the console, and check in the background if
the instance is paused, unpausing it as necessary.
Signed-off-by: Stephen Shirley <diamond@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

This fixes the lint error:

E1120:1220:InstanceReboot: No value passed for parameter
'startup_paused' in function call
Signed-off-by: Stephen Shirley <diamond@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

This patch allows commands to be run on and files to be copied to all
nodes within a specific group.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

This patc changes cli.GetOnlineNodes to use query2, which does the
filtering in the master daemon, and adds a new parameter to filter by
node group.

Unittests were added for the old implementation and then adopted to
ensure no functionality was lost.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Explicitely defining “__call__” silences a pylint warning when wrapped
type check functions are used directly. I had no idea pylint is this
intelligent.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Places which receive floats can usually also deal with integers, e.g.
OpTestDelay. Tests are added and the new check function is used for the
aforementioned opcode and verifying query results.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Commit 009e73d0

 (September 2009) changed the job queue to generate
multiple job serials at once. Ever since it would return one more than
requested.

The “serial” file in the job queue directory is defined to contain the
“last job ID used” (design-2.0). With the change above, the serial file
would always contain the next serial number. The first value returned by
the generating function was the one contained in the file, so during the
switch in 2009 one job may have been overwritten.

This patch changes the code to always return the exact number of
serials, to keep the last used serial on disk and adds an assertion.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

This reverts commits 030a9cb8 and
ae082df0

.

There are two problems:

- Makefile.am breakage, which is trivial to revert
- unittest breakage, which honestly I'm not sure how to fix and how
  serial consoles interact with the unpause helper

After the reset, the startup --paused still works but won't unpause
the instance automatically (if I understood the code correctly).

Furthermore, the code also fixes a style issue in hv_kvm.py (too long
line) introduced by the next commit after the above two.
Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

Creates the instance, but pauses execution before booting. This combined
with 'gnt-instance console' unpausing instances means that the entire
boot process can be viewed and monitored.
Signed-off-by: Stephen Shirley <diamond@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

The wrapper will connect to the console, and check in the background if
the instance is paused, unpausing it as necessary.
Signed-off-by: Stephen Shirley <diamond@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

The wrapper will connect to the console, and check in the background if
the instance is paused, unpausing it as necessary.
Signed-off-by: Stephen Shirley <diamond@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

- Use constants and an assertion
- Update documentation for node migration
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

LUNodeEvacStrategy has been replaced with LUNodeEvacuate.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

The change is not backwards compatible, see the updated NEWS file.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

By default it'll now evacuate all instances from the node, not
just secondaries.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

This new opcode will replace LUNodeEvacStrategy, which used to return a
list of instances and new secondary nodes. With the new opcode the
iallocator (if available) is tasked to generate the necessary operations
in the form of opcodes. This moves some logic from the client to the
master daemon.

At the same time support is added to evacuate primary instances, which
are also evacuated by default.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

Am I the only one to make that mistake 10 times a week?
Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

There were some implicit assertions in the code that all node groups
have nodes, which is not necessarily true.

Additionally, the patch does a wrapping change.
Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

The new functionality in 2.4.2 for recreate-disks to change nodes is
broken for DRBD instances: it simply changes the nodes without caring
for the DRBD minors mapping, which will lead to conflicts in non-empty
clusters.

This patch changes Exec() method of this LU significantly, to both fix
the DRBD minor usage and make sure that we don't have partial
modification to the instance objects:

- the first half of the method makes all the checks and computes the
  needed configuration changes
- the second half then performs the configuration changes and
  recreates the disks

This way, instances will either be fully modified or not at all;
whether the disks are successfully recreate is another point, but at
least we'll have the configuration sane.
Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Patch db8e5f1c

 removed the use of feedback_fn, hence pylint warn
now.
Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

Commit 5d9bfd87

 moved tap interface handling from KVM to Ganeti, partly
to also solve the problem of routed interfaces getting configured too
early during live migrations, causing network anomalies. In that
direction, configuration of NICs of incoming instances was deferred to
FinalizeMigration time.

However, this causes minor issues with bridged interfaces; KVM sends out
an ARP-like packet upon migration finish, which is lost because the tap
interface is not yet configured. As a consequence, intermediate network
equipment (i.e. switches) does not get notified about the topology
change, until the instance transmits another packet after the bridge has
been configured, or the switch's ARP cache expires.

The proper solution to that is to support different phases in network
configuration (pre/post migration), which also requires separate ifup
scripts. Until then we fall back to configuring bridged interfaces on
incoming instances at migration start, instead of finish.
Signed-off-by: Apollon Oikonomopoulos <apollon@noc.grnet.gr>
Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Currently the drbd8 replace-disks on the same node (i.e. -p or -s) has
a bug in that it does modify the instance disk temporarily before
changing it back to the same value. However, we don't need to, and
shouldn't do that: what this operation do is simply change the LVM
configuration on the node, but otherwise the instance disks keep the
same configuration as before.

In the current code, this change back-and-forth is fine *unless* we
fail during attaching the new LVs to DRBD; in which case, we're left
with a half-modified disk, which is entirely wrong.

So we change the code in two ways:

- use temporary copies of the disk children in the old_lvs var
- stop updating disk.children

Which means that the instance should not be modified anymore (except
maybe for SetDiskID, which is a legacy and unfortunate decision that
will have to cleaned up sometime).
Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

This function is a copy of bootstrap._InitFileStorage with the following
differences:
  - check constants.ENABLE_SHARED_FILE_STORAGE and not
    constants.ENABLE_FILE_STORAGE
  - use different local variable names
  - one different error string

Thus:
  - move the constant check outside of the function call
  - change error string so it's clear where the error is
  - call the same function twice
Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Under newer kvm this prevents the vm from starting.
Ah, change!
Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

When using the pool security model, _ExecuteKVMRuntime was storing the
instance's UID using str(uid), which would result in storing the
LockedUid.__repr__() result:

 $ cat /var/run/ganeti/kvm-hypervisor/uid/xxxxxxxxxxxxx
 <ganeti.uidpool.LockedUid object at 0x1f30610>

This patch restores the intended behaviour, by using LockedUid.AsStr().
Signed-off-by: Apollon Oikonomopoulos <apollon@noc.grnet.gr>
Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

This was left out during the fix/refactoring
Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

- Move the calculation at the beginning of CheckPrereq, since it doesn't
  modify any state, but still keeps locks
- Only perform the calculation if the actual disk template is filebased
- Error out if there is no defined file storage dir
- Only join the optional --file-storage-dir extra-path if one is passed
Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

As the manpage says, and the code does, self.op.file_storage_dir is an
additional relative path under the cluster file storage dir. As such it
should not be absolute.
Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

This patch removes all occurrences of the “multi-relocate” iallocator
mode. Commit 25ee7fd8

 updated the design document and introduced
separate modes, “change-group” and “node-evacuate”. The constants aren't
removed yet as they're still used by htools.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: René Nussbaumer <rn@google.com>

Chained jobs need to look at previous jobs, including archived ones. A
nice side-effect of this change is the ability to look at archived jobs
using “gnt-job info <id>” as long as the ID is known.
Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>