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

Currently this only supports the new instance creation request data
format version 1, but support for the old version can be easily
implemented.

Most arguments are optional and documented in the RAPI documentation.

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

As mentioned in commit d975f482, the current way of creating an
instance via RAPI is not very flexible. With this patch, a new
instance creation request data format is introduced and documented.
Support can be detected by checking the list of features returned
by the /2/features resource for the value "instance-create-reqv1".

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

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

Also add a separate function to retrieve body parameters. Having
it separate (independent of a class instance) will make it easier
to unittest users of this function.

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

The /2/features RAPI resource can be used to detect optional
features implemented by the RAPI server. This will be used
to recognize servers implementing a new request format for
instance creation requests.

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

Node daemon logs will be printed on system console if writing on the log file
fails.

Signed-off-by: Luca Bigliardi <shammash@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Signed-off-by: Luca Bigliardi <shammash@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Add a parameter to SetupLogging to use LogFileHandler (default is off)

Signed-off-by: Luca Bigliardi <shammash@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Add a log handler class which logs to /dev/console in case of logging errors.

Signed-off-by: Luca Bigliardi <shammash@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Signed-off-by: Luca Bigliardi <shammash@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Signed-off-by: Luca Bigliardi <shammash@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Signed-off-by: Luca Bigliardi <shammash@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Signed-off-by: Luca Bigliardi <shammash@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Add Mlockall() utility to lock current process' virtual adress space into RAM.

Signed-off-by: Luca Bigliardi <shammash@google.com>
Reviewed-by: Michael Hanselmann <hansmi@google.com>

Signed-off-by: Luca Bigliardi <shammash@google.com>
Reviewed-by: Guido Trotter <ultrotter@google.com>

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

The way the resource /2/instances expects its request data (e.g.
instance name, disks, NICs) to be formatted in a dict is not
very extensible. HV and BE parameters are interleaved with all
other values. In commit 495cfdf0 a new parameter “mode” was added
which can be misunderstood as the instance creation mode, but is
actually a mode for the network interface.

Short of bumping the whole RAPI version (currently 2), which would
involve many, many other changes and potentially break clients,
we can not change the request format easily as it needs to stay
backwards and future-compatible as far as possible. This is not an
internal API, but one used by several external applications.

By introducing a data format version per request, we can still
support the old request format, and thereby not break existing users,
while we can add a more flexible request format which will, for
example, support multiple NICs and move HV/BE parameters to separate
containers, avoiding conflicts.

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

For passwords we require strict oneliners, we're a bit more lax with pid
and uid files.

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Read the first non-empty file line. When strict is set, abort if more
than one line is non-empty. Some unittests inspired by the reverted ones
from commit b774bb10, and some new ones.

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

This partially reverts commit b774bb10.
Unittests unrelated to that particular functionality but introduced in
that commit are left untouched. Since the temporary directory is now
only used to check for a non-existing file, it is removed as well, and
a convenient very-very-unlikely-to-exist file is used instead.

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

The 0.7 version was just an example, but since we're on 0.9 now,
updating the number cannot be wrong. The new admonition helps KVM users
to configure debootstrap correctly so that instances can be shutdown
cleanly.

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

Xen: Just specify 3.0 or above as a requirement, as versions above 3.1
have been tested and work fine with Ganeti. Also note that the Xen
dependency is optional, if one wants to work with KVM.

KVM: list as an alternative, specify that 72 (the one shipped with
lenny) is the bare minimum, but newer qemu-kvm is recommended.

Signed-off-by: Guido Trotter <ultrotter@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

My previous patch, commit 857705e8, helped in one particular
situation where the exception didn't contain any arguments
(pyOpenSSL reporting a WantReadError, which shouldn't occur with a
blocking socket anyway). With this patch, more common and uncommon
errors should be easy to recognize.

SSL errors without any of these patches:
“ganeti.rapi.client.CertificateError: SSL issue:
[('SSL routines', 'SSL3_GET_SERVER_CERTIFICATE',
'certificate verify failed')]”

SSL errors with both patches:
“ganeti.rapi.client.CertificateError: SSL issue:
[('SSL routines', 'SSL3_GET_SERVER_CERTIFICATE',
'certificate verify failed')] (<OpenSSL.SSL.Error
instance at 0x…>)”

WantReadError without any of these two patches:
“ganeti.rapi.client.CertificateError: SSL issue: ”

WantReadError with both patches:
“ganeti.rapi.client.CertificateError: SSL issue:
(<OpenSSL.SSL.WantReadError instance at 0x…>)”

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

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

Only the server knows which node roles can be set via RAPI.
Constants are provided for convenience.

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

- The RAPI client shouldn't check the replacement mode as only
  the server knows which ones are valid (constants are still
  provided for convenience)
- Disks shouldn't be a list of names, but of indexes

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

When modifying a storage unit, the “allocatable” flag should default
to “no modification”. This replicates the behaviour of the command
line interface.

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

Boolean values must be passed as 0 or 1. None should be an empty
value ("").

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

Only the RAPI server (actually masterd) knows which storage types
are valid. The exception can no longer be raised as the type is
only checked in the job.

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

This reverts a60e3cb0 partially by moving the RAPI version into a constant.

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

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

The RAPI tests will depend on the certificate.

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

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

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: David Knowles <dknowles@google.com>

The RAPI client module shouldn't depend on any Ganeti module, yet it's
useful to have some Ganeti-specific code, like a PollJob function for
RAPI.

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: David Knowles <dknowles@google.com>

Having the HTTP error code allows users of the class to act differently based
on the error code.

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: David Knowles <dknowles@google.com>

This allows testing error conditions (e.g. if a job wasn't found).

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: David Knowles <dknowles@google.com>

Signed-off-by: Michael Hanselmann <hansmi@google.com>
Reviewed-by: David Knowles <dknowles@google.com>

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