bundlewrap/items/zfs_pool.py

from collections import Counter
from pipes import quote

from bundlewrap.exceptions import BundleError
from bundlewrap.items import Item
from bundlewrap.utils.text import mark_for_translation as _


def prepare_blockdevice(node, device):
    # To increase our chances of success, we run partprobe beforehand to
    # make the kernel re-scan all devices.
    node.run("partprobe", may_fail=True)

    # Try to find out if the device already contains some filesystem.
    # Please note that there is no 100% reliable way to do this.
    res = node.run("lsblk -rndo fstype {}".format(quote(device)))
    detected = res.stdout.decode('UTF-8').strip()
    if detected != "":
        raise Exception(_("Device {} to be used for ZFS, but it is not empty! Has '{}'.").format(
            device, detected))

    res = node.run("lsblk -rndo type {}".format(device))
    if res.stdout.decode('UTF-8').strip() == "disk":
        # We create a new partition on a raw disk. That partition will
        # be used for ZFS. Yes, this is on purpose. No, creating a pool
        # on raw disks does not work properly on Linux.
        node.run("parted -s {} mklabel gpt".format(quote(device)))
        node.run("parted -s {} mkpart -a optimal primary 0% 100%".format(quote(device)))
        node.run("partprobe")

        # Simply append a "1" to get to the first partition.
        #
        # XXX I know that this fails if you're using /dev/disk/by-*.
        # Yes, this is a problem if your device names are not
        # predictable. Yes, we could use "lsblk" to try to find the
        # first partition ... but "lsblk" still reports it as
        # "/dev/vdb1" instead of "/dev/disk/by-foo/bar-part1".
        #
        # This is an unsolved problem. Please configure your VMs to use
        # predictable device names.
        if device.find('nvme') != -1:  # NVME Devices have the partitions with the prefix pX
            partition = "{}p1".format(device)

        else:
            partition = "{}1".format(device)

        return partition
    else:
        return device


class ZFSPool(Item):
    """
    Creates ZFS pools and the required partitions.
    """
    BUNDLE_ATTRIBUTE_NAME = "zfs_pools"
    ITEM_ATTRIBUTES = {
        'config': None,
        'autotrim': None,
        'autoreplace': None,
        'autoexpand': None,
    }
    ITEM_TYPE_NAME = "zfs_pool"

    def __repr__(self):
        return "<ZFSPool name:{} autoexpand:{} autoreplace:{} autotrim:{} config:{}>".format(
            self.name,
            self.attributes['autoexpand'],
            self.attributes['autoreplace'],
            self.attributes['autotrim'],
            self.attributes['config'],
        )

    def cdict(self):
        ret = {}
        for i in {'autoexpand', 'autoreplace', 'autotrim'}:
            if self.attributes.get(i):
                ret[i] = self.attributes[i]
        return ret

    @property
    def devices_used(self):
        devices = set()
        for option in self.attributes['config']:
            for device in option['devices']:
                devices.add(device)
        return sorted(devices)

    def fix(self, status):
        if status.must_be_created:
            cmdline = []
            for option in self.attributes['config']:
                if option.get('type'):
                    cmdline.append(option['type'])
                    if option['type'] == 'log' and len(option['devices']) > 1:
                        cmdline.append('mirror')
                for device in sorted(option['devices']):
                    cmdline.append(quote(prepare_blockdevice(node, device)))

            self.run('zpool create {} {}'.format(quote(self.name), ' '.join(cmdline)))

        for attr in {'autoexpand', 'autoreplace', 'autotrim'}:
            if attr in status.keys_to_fix:
                state_str = 'on' if status.cdict[attr] else 'off'
                self.run('zpool set {}={} {}'.format(attr, state_str, quote(self.name)))

    def sdict(self):
        status_result = self.run('zpool list {}'.format(quote(self.name)), may_fail=True)
        if status_result.return_code != 0:
            return {}

        autoexpand_state = self.run('zpool get autoexpand -H -o value {}'.format(quote(self.name)), may_fail=True).stdout.decode().strip()
        autoreplace_state = self.run('zpool get autoreplace -H -o value {}'.format(quote(self.name)), may_fail=True).stdout.decode().strip()
        autotrim_state = self.run('zpool get autotrim -H -o value {}'.format(quote(self.name)), may_fail=True).stdout.decode().strip()
        return {
            'autoexpand': (autoexpand_state == 'on'),
            'autoreplace': (autoreplace_state == 'on'),
            'autotrim': (autotrim_state == 'on'),
        }

    def test(self):
        duplicate_devices = [
            item for item, count in Counter(self.devices_used).items() if count > 1
        ]
        if duplicate_devices:
            raise BundleError(_(
                "{item} on node {node} uses {devices} more than once as an underlying device"
            ).format(
                item=self.id,
                node=self.node.name,
                devices=_(" and ").join(duplicate_devices),
            ))

        # Have a look at all other ZFS pools on this node and check if
        # multiple pools try to use the same device.
        for item in self.node.items:
            if (
                item.ITEM_TYPE_NAME == "zfs_pool" and
                item.name != self.name and
                set(item.devices_used).intersection(set(self.devices_used))
            ):
                raise BundleError(_(
                    "Both the ZFS pools {self} and {other} on node {node} "
                    "try to use {devices} as the underlying storage device"
                ).format(
                    self=self.name,
                    other=item.name,
                    node=self.node.name,
                    devices=_(" and ").join(set(item.devices_used).intersection(set(self.devices_used))),
                ))

    @classmethod
    def validate_attributes(cls, bundle, item_id, attributes):
        if not isinstance(attributes['config'], list):
            raise BundleError(_(
                "{item} on node {node} 'config'"
            ).format(
                item=item_id,
                node=bundle.node.name,
            ))

        for config in attributes['config']:
            if not config.get('type', '') in {'', 'mirror', 'raidz', 'raidz2', 'raidz3', 'cache', 'log'}:
                raise BundleError(_(
                    "{item} on node {node} has invalid type '{type}', "
                    "must be one of (unset), 'mirror', 'raidz', 'raidz2', "
                    "'raidz3', 'cache', 'log'"
                ).format(
                    item=item_id,
                    node=bundle.node.name,
                    type=config['type'],
                ))

            if not config.get('devices', set()):
                raise BundleError(_(
                    "{item} on node {node} uses no devices!"
                ).format(
                    item=item_id,
                    node=bundle.node.name,
                ))

            if config.get('type') == 'log':
                if not 0 < len(config['devices']) < 3:
                    raise BundleError(_(
                        "{item} on node {node} type 'log' must use exactly "
                        "one or two devices"
                    ).format(
                        item=item_id,
                        node=bundle.node.name,
                    ))