The DRBD driver can ignore dead disks but not dead meta devices (for
which it refuses to configure the minor). To handle this case, we check
that the meta device is readable and if not we ignore it (the same as
when backend._RecursiveAssembleBD didn't find it).

A needed change is the move of some checks and initialisers before this
test (which is before the super().__init__ call), but that should be
fine.

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

A newer version of pylint, more warnings…

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

We will need to enumerate selectively the PVs of (possible) many VGs and
not only the allocatable ones. For this we make the VG selection and the
allocatable filtering optional. The two callers are modified for this
new calling syntax.

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

While ':' is not actually a supporte character in PV names (it has a
special meaning for commands like lvcreate), we should throw specific
errors for this case instead of generic “Can't create LV”.

This patch does two things:

- modifies the separator used when listing PVs to be '|' such that we
  can actually parse ':' as part of PV names
- check if any of the discovered PVs have ':' in their name when
  creating LVs, and if so throw a specific error

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

Also replaces a hardcoded limit of 15 seconds with 1/4
of NET_RECONFIG_TIMEOUT.

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

This survived QA, burnin and unittests.

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

In case there are gaps in /proc/drbd related to the minors sequence,
this can lead to empty lines (see the attached data file). In this case,
we currently fail to parse the file correctly.

The patch fixes this by skipping empty lines in
bdev.BaseDRBD._MassageProcData and adds a data file and unittest for
this condition.

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

This allows us to report “uncertain” states (LDS_UNKNOWN) for cases
where the code can't easily detect or report what's wrong with a
block device.

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

This patch changes the return type for backend.BlockdevFind to an object
(objects.BlockDevStatus). Before a tuple was used. Adding more values to
this tuple causes a lot of work.  Converting the result to an object with
properties will make this a bit simpler.

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

This patch adds a GetActualSize for block devices that returns the
actual disk size. It is done using blockdev (and stat for file storage).

While this could be done via reading /sys/block/N/size, that is not as
simple as running blockdev, as the correspondence between an LV and its
sys entry is not straightforward.

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

This patch changes the DRBD8 class (the only one to use the size in
Assemble) to ignore the size in Assemble when a zero size is passed.
This will allow activation of disks even when the size recorded in the
configuration is wrong.

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

LVM has issues when growing stripped volumes, so it's best to specify
the growth in exact multiples of the full stripe size (as precise as
possible). For this we need to do a couple of changes:
  - in LVM Attach(), we query additionally the VG extent size and the LV
    stripe count; since this makes lvs return a (possibly) multi-line
    output, we now split it into lines and only take the last one
  - in LVM Grow(), we round up the increase in multiples of the full
    stripe size

The patch also sets the correct target size in DRBD growth.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Olivier Tharan <olive@google.com>

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

With the change to stripped LVs, the actual size of a meta device (which
is small) can be more than we expected (for non-stripped LVs). This
patch increases from 160MB to 1GB the accepted size, and updates the
comment with the rationale behind this change.

Note that we do want even meta devices stripped, since it can increase
metadata update.

Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Olivier Tharan <olive@google.com>

This patch enables stripped LVs, falling back to non-stripped if the
stripped creation fails. If the configure-time lvm-stripecount is 1,
this patch becomes a noop (with an insignificant python-level overhead,
but no extra lvm calls).

The effect of this patch is that new instances will get stripped LVs
from the start, whereas old instances will have their LVs stripped as
soon as replace-disks is run for them.

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

This patch adds constants for the connection status, peer roles and disk
status, and it changes the rules for when the disk is considered as
“resyncing” - previously it was only for syncsource/synctarget, but
there are many other transient statuses which could be misinterpreted as
‘degraded’ (because they where not considered as resyncing, but the disk
is not consistent in these statuses).

Furthermore, cmdlib.py:WaitForSync determines if a device is syncing or
not based on sync_percent being not none. Not all DRBD resync statuses
offer a percent done, so if we are syncing but don't have a sync
percent, we'll report a zero sync percent (and no time estimate).

The patch also removes a few unused variables (is_sync_target,
peer_sync_target, is_resync) whose value doesn't make sense anymore with
the new sync rules.

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

This patch changes DRBD disk attachment to force the wanted size, as opposed to
letting the device auto-discover its size.

This should make the disks more resilient with regard to small differences in
size (e.g. due to LVM rounding). This still works with regard to disk
growth, but the instances needs to be fully restarted (including disks)
in that case.

This passes a full burning without problems, but it's still a tricky
change - if the config.data is not synced with the reality, we might
tell DRBD a wrong size. At least this will fail outright (and not
introduce silent errors), as DRBD (per a quick check at the sources)
tracks the size in the meta-dev and also does not allow shrinking
consistent devices.

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

This patch changes all the bdev.BlockDev constructors to take an
additional ‘size’ parameter, all the backend functions that call those
functions to pass it and also changes backend.BlocdevCreate() to not use
the size passed via the rpc call but instead directly disk.size (this is
the only way it's called).

Note that this patch doesn't do anything with this parameter, just
stores it on the blockdev objects.

With the patch, we actually have a more uniform init sequence (before
create had the parameter, but the other functions not).

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

DRBD 8.3 changes two more things compared to 8.2:
  - /proc/drbd format changed in multiple ways; the part we're
    interested is the ‘st:’ to ‘ro:‘ change (in the changelog named as
    “Renamed 'state' to 'role'”
  - “drbdsetup /dev/drbdN show” changed the ‘device’ stanza from:
      device "/dev/drbd0";
    to:
      device                  minor 0;

This patch fixes these both and adds data files and unittests for DRBD
8.3.1.

Signed-off-by: Iustin Pop <iustin@google.com>

This patch adds (and suppresses) the extra ipv4/ipv6 words before the
actual address that newer DRBD versions add.

[iustin@google.com: slightly changed the patch to conform to style
guide, and changed the commit message]
Signed-off-by: Iustin Pop <iustin@google.com>
Reviewed-by: Iustin Pop <iustin@google.com>

If /proc/drbd can't be opened, this raises an IOError, but all the
error-handling behaviour in backend treats only BlockDeviceErrors. This
creates a plain failure in cluster verify and in other RPC calls.

This patch simply converts EnvironmentErrors into BlockDeviceErrors, and
also changes the RPC result for NV_DRBDLIST and its handling to be able
to show the error. The other RPC calls work by default now, due the
existing error handling.

Reviewed-by: ultrotter

In FileStorage there is a TODO:
 decide whether we should check for existing files and
 abort or not
After Ganeti ate my instance data I decided. Let's abort.
In general there is no reason we should overwrite existing files, and
doing it can be very harmful for preexisting files on the host.

Reviewed-by: iustinp

Reviewed-by: imsnah

This patch converts the _CheckMetaSize method to raise exceptions
instead of logging and returning False. This fits now in the new rpc
return types, so it's a cheap change.

Reviewed-by: ultrotter

This big patch converts the bdev Assemble() methods and the supporting
functions to raise exceptions instead of returning False. This is a big
patch, since the assembly functions touch other functions: add children,
creation, etc. However, the patch does not add much new code, rather it
reworks existing code.

One of the biggest changes is in the rework of the DRBD8._SlowAssemble()
method (one of the most complicated/ugly ones). Hopefully the new
version is a little bit more readable.

Reviewed-by: ultrotter

Currently, the Remove() methods of block devices return True/False.
This doesn't permit any error detail reporting.

This patch changes the return type to None for success, and raises
BlockDeviceError in case of failure. This permits the details to be
passed up the stack.

The patch also simplifies a little the Remove method of file-based
devices (no stat first, just try unlink).

Reviewed-by: ultrotter

Currently, the Shutdown() methods of block devices return True/False.
This doesn't permit any error detail reporting.

This patch changes the return type to None for success, and raises
BlockDeviceError in case of failure. This permits the details to be
passed up the stack.

For LVM and file-backed devices, this is a simple change. For DRBD, we
first remove the shutdown of disks in case of network activation
failures (since with static minors the minor is used anyway, we don't
gain anything by clearing it), and the we simply change _ShutdownAll()
to raise an exception.

Reviewed-by: ultrotter

This patch adds two utility functions for raising BlockDeviceError
exceptions and for running functions while ignoring this error. Most of
the manual “raise errors.BlockDeviceError” cases are converted to
_ThrowError, as this makes the code clearer.

We also change most of the DRBD error messages to include the minor
number because with the parallel execution of commands it's not longer
possible to identify the failed DRBD just from the timestamp, and the
minor number can be mapped back to the instance easier.

Reviewed-by: ultrotter

Currently the DRBD code checks that the metadata devices are valid
before creation, initial disk attachment and add children.

However, the process for checking validity requires a free DRBD minor,
and this conflict with parallel checking.

There are at least three possible solutions:
  - serialize all checks, which means we reduce parallelism and need
    extra locks
  - don't pass a valid minor number, but one like “/dev/drbd256” (which
    is invalid); this works for current version of DRBD, but since it's
    not guaranteed to remain so it doesn't look nice
  - don't do the checking at all, and rely on “drbdsetup ... disk ...”
    to fail by itself

The reason for checking metadata was that in 1.2, this was much cheaper
than trying to activate devices (and the subsequent iteration over the
minors). However, in 2.0, they have the same cost, so we can choose
option 3: just remove the explicit checking and rely on drbdsetup and
the kernel to fail.

Since DRBD8._InitMeta still requires a minor number, the two places
where this is run are handled as follows:
  - Create: we just use our own (unused currently) minor number
  - AddChildren: we keep using FindUnusedMinor, with the caveat that
    this function (used by replace-disks -n ...) cannot be yet
    parallelized

Reviewed-by: ultrotter

This patch adds support for verification of drbd minors space in cluster
verify: minors which belong to running instances and should be online
but are not, and minors which do not belong to any instace but are in
use.

The patch requires exposing some methods from bdev.DRBD8 and
config.ConfigWriter which were until now private methods.

Reviewed-by: ultrotter

In order to prevent errors with old, in-use DRBD minors, we check and
abort at create time if our minor is already in use. For this we need to
also modify DRBD8Status to be able to parse cs:Unconfigured devices.

Reviewed-by: ultrotter

Currently when creation LVM-based instances, we always get the
extremely-confusing message "ERROR Can't find LV /dev/xenvg/..." which
is actually expected. This behaviour was introduced before we had
UUID-style LV names, since at that point it was not a unexpected to have
such volumes laying around after a failed creation.

Today, it's much more of an error to see existing volumes, and it's
better to abort with a failure. Since bdev.LogicalVolume.Create() method
will raise an error in case it exists, we can remove this check in
backend before creating the device.

The Create methods for DRBD and FileStorage currently don't raise
exception, as behaviour is not very well defined here.

We also change some exception types raised in bdev so that all
exceptions raised by device creation are a subclass of GenericError.

Reviewed-by: ultrotter

This is a modified forward-port of DrbdNetReconfig and their associated
RPCs. In Ganeti 2.0, these functions will be used for two things:
  - live migration (as in 1.2)
  - and for other network reconfiguration tasks, since DRBD8.Attach()
    doesn't do them anymore

Because of the Attach() changes, we can now implement the
AttachNet/DisconnectNet functions as independent entities, and we don't
need the cache anymore.

Note these functions are copies of the latest 1.2 code, and not
cherry-picks of the (many) patches that went into 1.2.

Reviewed-by: ultrotter

Since now the Assemble function is different than Attach, we rename this
backend function to show that the intent is to fully assemble the device
(and it's always allowed to modify the device).

Reviewed-by: ultrotter

Currently, both the Attach and Assemble methods for DRBD8 devices will use and
alter the device state. This is suboptimal, and it has been worked
around in 1.2 via a special cache in the node daemon so that we don't
need to call Attach() again in migration, for example.

Since in 2.0 we have static minors, we can change these functions so
that:
  - Attach() does not affect the device in any way, and only checks if
    the minor is already in use or not
  - Assemble() has two logic paths, one for startup from unused minor
    (the old Assemble, now renamed _FastAssemble) and one for
    re-checking/fixing an in-use minor (the old Attach, now renamed
    _SlowAttach)

Basically Attach was renamed to _SlowAttach, Assemble to _FastAssemble,
and we have a new, simple Assemble that calls one or the other based on
the result of the new Attach.

The LUReplaceDisks (with new secondary) is relying on the special
semantics of Attach modifying the device and is broken until the end of
the patch series.

Reviewed-by: ultrotter

The caller of dev.Assemble() (backend._RecursiveAssembleBD) is doing an
explicit recursion over all the children of the device, with better
error reporting. As such, we don't need this repeated assembly inside
the base BlockDev class.

Reviewed-by: ultrotter

This is modified forward-port of commit 1544 on the 1.2 branch:

  When DRBD is doing its dance to establish a connection with its
  peer, it also sends the synchronization speed over the wire. In
  some cases setting the sync speed only after setting up both
  sides can race with DRBD connecting, hence we set it here before
  telling DRBD anything about its peer.

  Reviewed-by: iustinp

The modification we make is that we split SetSyncSpeed in two so that we
don't need to modify our minor temporarily, and the fact that we call
this function from within _AssembleNet (right before enabling network),
instead of Assemble()/Attach().

Original-Author: imsnah

This is plain copy of the 1.2 ReAttachNet and DisconnectNet methods on
the DRBD8 device, with the logger to logging module changes and the
ReAttachNet method renamed to AttachNet.

These methods are not used anywhere right now, but will be used for
migration and a simpler disk-replace.

The code was originally committed on the 1.2 branch as revision numbers
1165 and 1204.

Originally-Reviewed-by: imsnah, ultrotter

This patch should fix all outstanding epydoc parsing errors; as such, we
switch epydoc into verbose mode so that any new errors will be visible.

Reviewed-by: imsnah

We changed a while ago the protocol for opening block devices, but
FileStorage was not changed. This patch makes it work again.

Reviewed-by: imsnah