GitLab

These cluster parameters were introduced in 2da9f556
without being added to the parameters implicitly upgraded
if missing. This now causes problems when upgrading pre-2.6
clusters to 2.11 or later where this parameter is assumed
to be a dictionary. So add the implicit upgrade.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Fixes issue #1015
Signed-off-by: Aaron Karper <akarper@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Cherry-picked from 19a5c6c3Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Hrvoje Ribicic <riba@google.com>

Our current hlint version cannot find them yet.
Signed-off-by: Niklas Hambuechen <niklash@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Cherry-picked from: d05f1c86

Conflicts:
	src/Ganeti/Utils.hs (trivial)
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Niklas Hambuechen <niklash@google.com>

For generated files, it is enough to check the headers of the
files they are generated from. Moreover, the generated files
have the shebang-line of the target system, which might be
different from the generic one checked for.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Niklas Hambuechen <niklash@google.com>

While most of our top-level python scripts used to be generated
in the Makefile already, the scripts under tools used to be
"ready to use". However, in our current situation where the
python 2 interpreter is called different on different platforms
we still need to fix the shebang line. This patch achieves this
without moving source files around.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

...in our automatically generated top-level python files.
Instead use the automatically detected python path.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

...this will allow inspecting, e.g., the effective compiler
flags without actually running the compiler. For example,
to see the value of the HFLAGS,

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

Unless you develop Ganeti, it is not a problem if some of
the functions used are deprecated in you newer haskell
environment. So allow those builds for normal users.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

So that we can have fatal warnings in development, but
people can still build old Ganeti versions with newer
compilers despite some functions are deprecated by now.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

if using python2-sphinx '/usr/bin/sphinx-build2 --version' prints
"Sphinx (sphinx-build2) 1.2.3" which didn't match the regex
Signed-off-by: Neal Oakey <neal.oakey@googlemail.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

use the Python interpreter which has been set via PYTHON env-var at
configure time
Signed-off-by: Neal Oakey <neal.oakey@googlemail.com>
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

When running gnt-instance console for an instance name that doesn't
exist, I got the following raw error:

...
    cl.QueryInstances([instance_name], ["console", "oper_state"], False)
ValueError: need more than 0 values to unpack

In this case we should raise the OpPrereqError to notify that the
instance which having that name doesn't exist.
This patch adds a check for the return value of the QueryInstances and
raise the OpPrereqError when the result of the query is a blank list.
Signed-off-by: Yuto KAWAMURA(kawamuray) <kawamuray.dadada@gmail.com>
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

* stable-2.9
  Ganeti.Daemon: always install SIGHUP handler
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Newer GnuTLS versions (>= 3.3.0) use a library constructor for
initialization and open /dev/urandom on library load, way before we
fork(). Closing /dev/urandom on fork causes a failure to re-seed GnuTLS's
random number generator during the first ganeti.http.client request, which
in turn causes the process to silently abort(3).

For more background on this behavior, see this thread at the GnuTLS
mailing list:

http://lists.gnupg.org/pipermail/gnutls-help/2014-April/003429.html

Note that calling pycurl.global_init() at the correct place (as we do) is not
enough, as it does not cause a re-initialization of the GnuTLS library.

As we cannot reliably detect neither the GnuTLS version, nor the socket, we
work our way around this by keeping all fds referring to /dev/urandom open
after fork. We do so using the /proc/self/fd interface.

This fixes issues #961 and #964.

Note that this would not affect the Haskell daemons using cURL + GnuTLS,
because we don't close all file descriptors on fork there.
Signed-off-by: Apollon Oikonomopoulos <apoikos@gmail.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Install the SIGHUP handler regardless of the daemonization status. This
fixes issue #755.
Signed-off-by: Apollon Oikonomopoulos <apoikos@gmail.com>
Signed-off-by: Petr Pudlak <pudlak@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Commit 742dc8de revealed a bug concerning DRBD version check during
cluster verify: In case non VM cabables nodes have DRBD not
installed, we get a version mismatch warning.

VerifyNode() updates the NV_DRBDVERSION key in the response message
only if a node is VM capable. Even if noded fails to retrieve the
DRBD version, the error string is returned.

Thus we must first check in _VerifyGroupDRBDVersion() if the
NV_DRBDVERSION key of the response message is updated and
then verify if we have different versions across nodes.
Signed-off-by: Dimitris Aragiorgis <dimara@grnet.gr>
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

When we still don't have any os scripts installed, running gnt-os
diagnose or gnt-os info foo(whatever) shows an error message "Can't get
the OS list" even if the request has been succeeded.
The OpOsDiagnose can return a blank list, so we should check whether
the result is None.
Signed-off-by: Yuto KAWAMURA(kawamuray) <kawamuray.dadada@gmail.com>
Reviewed-by: Hrvoje Ribicic <riba@google.com>

This patch renames the constant 'NV_VMNODES' to
'NV_NONVMNODES' as it is actually used to store
a list of *non* vm-capable nodes.
Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Hrvoje Ribicic <riba@google.com>

This patch populates the list 'NV_VMNODES' with node
names and not node UUIDs, as the backend only uses this
list to identify VM-capable nodes by querying for the
node name.
Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Hrvoje Ribicic <riba@google.com>

It was decided that Ganeti is relicensed under the 2-clause
BSD license. Update the license statements accordingly (issue #936).
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

From the man page, the --online option is supposed to mark an
instance down only if it is already offline. Otherwise it should
fail. With this patch we avoid undesired transitions to ADMIN_down
state while the instance is already up and running.

Fix the corresponding QA test as well (TestInstanceModify).
Signed-off-by: Dimitris Aragiorgis <dimara@grnet.gr>
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Changing certain hvparams while an instance is running leads to a
dangerous situation for the instance. When the instance undergoes live
migration without having undergone a reboot that would have applied the
parameters, it is setup with the modified hvparams, and loaded with the
state relying on the old hvparams. Depending on what is used and how,
this might lead to a crash of the instance.

While a proper fix should go into a stable version of Ganeti as this is
not a trivial change, the least we can do is warn users that hvparam
changes may be dangerous.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

This patch makes minor additions to the Ceph design doc documentation
to clarify that only some parts of the design have been implemented.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

During cluster verification, Ganeti would complain about
orphan LVs in VGs other than the one specified in its configuration.

This commit fixes the backend layer for nodegroup verification
to only report LVs in the VG passed by the master node as part
of the node verification parameters.
Signed-off-by: Vangelis Koukis <vkoukis@grnet.gr>
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Network query mechanism fills inst_list field with the list of
instances that are connected to the corresponding network.
Fix the case of multi NIC instances where the NIC scan stopped
once a matching network was found.
Signed-off-by: Dimitris Aragiorgis <dimara@grnet.gr>
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Make gnt-cluster upgrade refuse to upgrade if an upgrade to be resumed
is present. For the convenience of the user, consider an upgrade
command to the same target version as the upgrade to be resumed as an
equivalent way of saying that the pending upgrade is to be resumed.
Fixes issue 906.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

As also suggested in the manual upgrade instructions,
pause the watcher (for 1h) during the upgrade. Fixes
issue 905.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

The patch 3c260845 fixed a bug where
adding a new disk to an instance with deactivated disks activated the
disk. However, it also introduced some erroneous behaviour, preventing
disks from being added to online instances with --no-wait-for-sync.

In line with the original meaning of the patch, this patch modifies the
check to disallow adding disks to shutdown instances with
--no-wait-for-sync, and allow doing so for online instances.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

Note the security issue with config backups
and other changes.
Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

This fixes a lint error of the previous patch complaining
about an unused variable.
Signed-off-by: Helga Velroyen <helgav@google.com>
Reviewed-by: Michele Tartara <mtartara@google.com>

Since the config backup archive contains sensitive information and is
written in world-readable locations (/var/lib by default), it should be
created in a safe way and with strict permissions.

This commit uses a temporary file to tackle two issues: the relaxed
permissions of the archive which respected the umask of the user running
`gnt-cluster upgrade' and a (possible) collision attack using a
pre-created file with the predictable backup filename.
Signed-off-by: Apollon Oikonomopoulos <apoikos@gmail.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Ganeti versions 2.11 and 2.12 have both had patches related to strange
Xen domU states which are still legitimate running states, but Ganeti
did not recognize them as such. The previous two cherry-picked patches
bring some of these changes, but the final patch needed (79cb7ece)
cannot be cherry-picked because of a conflict with the other two
patches. The changes were made on different branches, and merged
together at a later time.

This patch manually adds the change, merging together the descriptions
of why certain states are allowed, and putting all of this into the
function docstring.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

... which means that a domain is in the queue behind other domains
waiting to be scheduled.  The domain is runnable but it is not
currently running.
Signed-off-by: Jose A. Lopes <jabolopes@google.com>
Reviewed-by: Hrvoje Ribicic <riba@google.com>

Cherry-picked from 30350df5.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

The QA test 'TestClusterBurnin' pauses instances and the Xen state
domain 'paused' was not being handled properly by Ganeti because it
was not being classified as either a running or shutdown state.  This
patch adds the 'paused' state to the set of running states by
extending the function 'ganeti.hypervisor.hv_xen._IsInstanceRunning'
with the corresponding Xen state string '--p---'.

This fixes the QA test 'TestClusterBurnin'.
Signed-off-by: Jose A. Lopes <jabolopes@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Cherry-picked from fd201010.
Signed-off-by: Hrvoje Ribicic <riba@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

Commit 2243b133 removed the old, positional, passing of mode
and link arguments. Adapt the documentation string accordingly.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

Commit 6b47261a exported the VLAN to the KVM NIC configuration scripts.
However, in doing so it assumed that the VLAN is always present in the
runtime, which is not true for instances being migrated but started on
Ganeti versions prior to 2.10.6.

We fix this by exporting the VLAN nicparam only if it actually exists.
This closes issue #893.
Signed-off-by: Apollon Oikonomopoulos <apoikos@gmail.com>
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Klaus Aehlig <aehlig@google.com>

The test framework we use sets PYTHONPATH for tests so that
the python module under test is used. Unfortunately, the paths
in PYTHONPATH are relative paths. So we need to expand them for
this test. Here we just update our fragile work around, but the
only way the test is called is, well, in testing, so we can build
it for our test environment.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Helga Velroyen <helgav@google.com>

* stable-2.9
  Make ganeti-cleaner switch to a save working directory
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>

For some reason, find does stat the current working directory,
which, when invoked via cron, may or may not be readable by the
process (issue 880). Therefore, first change to a directory that
is definitely accessible to the process; the directory it logs
into is such a safe place.
Signed-off-by: Klaus Aehlig <aehlig@google.com>
Reviewed-by: Petr Pudlak <pudlak@google.com>