PlanB automates remote SSH+rsync backups


Keywords
backup, django-application, rsync, rsync-backups, zfs, zfs-snapshots
License
GPL-3.0+
Install
pip install planb==1.7

Documentation

PlanB - automating remote backups and snapshots with zfs/rsync

PlanB backs up your remote files to a local ZFS storage. Manage many hosts and host groups. Automate hourly, daily, weekly, monthly and yearly backups with snapshots.

The following data transfer methods are supported:

  • ssh+rsync (built-in);
  • ssh+rsync of Kubernetes volume mounts (through kubersync), like Rook managed Ceph;
  • snapshots of ZFS (encrypted) datasets (through planb-zfssync);
  • snapshots of ZFS volumes (through planb-zfssync);
  • copies of (large) OpenStack Swift containers (through planb-swiftsync);
  • custom transfer (through your own custom transfer_exec script).

What it looks like

At the moment, the interface is just a Django admin interface:

A list of hosts configured in PlanB with most recent backup status

The files are stored on ZFS storage. It uses ZFS snapshots to keep earlier versions of files. See this example shell transscript:

# zfs list | grep mongo2
tank/BACKUP/experience-mongo2   9,34G  1,60T   855M  /srv/backups/experience-mongo2

# ls -l /srv/backups/experience-mongo2/data/srv/mongodb
total 646610
-rw------- 1 planb nogroup   67108864 jun 17 17:03 experience.0
-rw------- 1 planb nogroup  134217728 jun  9 16:01 experience.1
...

Those are the "current" files in the workspace. But you can go back in time:

# zfs list -r -t all tank/BACKUP/experience-mongo2 | head -n4
NAME                                                 USED  AVAIL  REFER  MOUNTPOINT
tank/BACKUP/experience-mongo2                       9,34G  1,60T   855M  /srv/backups/experience-mongo2
tank/BACKUP/experience-mongo2@planb-20170603T1147Z      0      -   809M  -
tank/BACKUP/experience-mongo2@planb-20170603T1211Z      0      -   809M  -

# cd /srv/backups/experience-mongo2/.zfs/
# ls -1
planb-20170603T1147Z
planb-20170603T1211Z
planb-20170604T0001Z
planb-20170605T0002Z
...

# ls planb-20170603T1147Z/data/srv/mongodb -l
total 581434
-rw------- 1 planb nogroup   67108864 jun  2 18:21 experience.0
-rw------- 1 planb nogroup  134217728 mei 29 14:38 experience.1
...

Requirements / setup

PlanB can be installed as a standalone Django application, or it can be integrated in another Django project.

See requirements.txt or setup.py for up-to-date dependencies/requirements.

Basically, you'll need: ZFS storage, ssh and rsync, a webserver (nginx), python hosting (uwsgi), a database (mysql), a communication/cache bus (redis) and a few python packages.

For more detailed steps, see Setting it all up below.

TODO

  • Encryption: right now, encryption keys are still a bit of a mess:
    • stuff is stored in tank/_local; should use some kind of vault;
    • when removing/renaming, those keys are not updated alongside;
    • planb-zfssync.sh does not clean up snapshots created before send/recv failure (e.g. because remote did not support --raw)
    • add key rotation example scripts?
  • Docs: add documentation for sync from previous unencrypted filesets?
  • Docs: add a bit of documentation on how to work with encrypted filesets
  • Consider: move the hostgroup contents to separate filesets, so as to create a more readable fileset listing. tank/HOSTGROUP/FILESET instead of tank/HOSTGROUP-FILESET.
  • RFE: Add post-backup.d directory somewhere where we can place post-backup-done scripts to manually do X or Y.
  • RFE: Add planb group for better permission management.
  • RFE: Also store user/group permissions on/after rsync (using xattr extended attributes?).
  • BUG: Items added to /exclude list are not deleted from destination if they have already been backed up once. The rsync job would need some way to keep track of changes in include/exclude settings, and run a cleanup in case they are changed. (See metadata storage like planb-swiftsync.* files.)
  • RFE: Standardize stdout/stderr output from Rsync/Exec success (and prepend "> " to output) to be more in line with failure.
  • RFE: Add possibility to feed back snapshot size from the individual Transport instead of using dutree. Parsing the swiftsync listings is fast after all.
  • FIX: Add uwsgi-uid==djangoq-uid check?
  • Replace the exception mails for common errors (like failing rsync) to use mail_admins style mail.
  • After using mail_admins style mail, we can start introducing mail digests instead: daily summary of backup successes and failures.
  • Replace the "daily report" hack with a signal-receiver.
  • Clarify why there's a /contrib/ and a /planb/contrib/ directory.

WARNING

The Django-Q task scheduler is highly configurable from the /admin/-view. With a little effort it will run user-supplied python code directly. Any user with access to the schedulers will have tremendous powers

Recommendation: don't give your users powers to edit the schedulers. Use the fine-grained permissions of the Django-admin systems to limit them to Hosts and HostGroups only.

Perhaps we should disable web-access to it altogether.

Setting it all up

If you follow the HOWTO below, you'll set up PlanB as a standalone project. Those familiar with Django will know how to integrate it into their own project.

The setup below assumes you'll be using the planb user. You're free to change that consistently of course.

Setting up a ZFS pool

You should really do your own research on this. If you're lucky, your operating system has native support for ZFS, and then this is relatively easy.

Please read README-zpool.rst for a quick introduction. When you're done, things should look somewhat like this:

# zpool status
  pool: tank
 state: ONLINE
  scan: none requested
config:

  NAME                                  STATE
  tank                                  ONLINE
    raidz2-0                            ONLINE
      scsi-SSEAGATE_ST10000NM0226_6351  ONLINE
      scsi-SSEAGATE_ST10000NM0226_0226  ONLINE
      scsi-SSEAGATE_ST10000NM0226_8412  ONLINE
      scsi-SSEAGATE_ST10000NM0226_...   ONLINE
      ...
    raidz2-1                            ONLINE
      scsi-SSEAGATE_ST10000NM0226_0123  ONLINE
      scsi-SSEAGATE_ST10000NM0226_...   ONLINE
      scsi-SSEAGATE_ST10000NM0226_...   ONLINE
      scsi-SSEAGATE_ST10000NM0226_...   ONLINE
      ...
  spares
    scsi-SSEAGATE_ST10000NM0226_9866    AVAIL
    scsi-SSEAGATE_ST10000NM0226_5992    AVAIL

Setting up the project

This section assumes you know a little about Python, pip and virtual envs. Details may vary a slight bit across distro versions.

Set up a virtualenv (optional):

mkdir -p /srv/virtualenvs
echo 'WORKON_HOME=/srv/virtualenvs' >>~/.bashrc
apt-get install python3-virtualenv python3-pip virtualenvwrapper
# you may need to log in/out once after this

# you may need /usr/share/bash-completion/completions/virtualenvwrapper
# sources in your bashrc
mkvirtualenv planb --python=$(which python3) --system-site-packages
workon planb

mkdir /etc/planb
cd /etc/planb
pwd >$VIRTUAL_ENV/.project  # or the src dir, if you're going to edit a lot

Install PlanB prerequisites:

apt-get install redis-server  # and: mysql-server or postgresql

Install PlanB dependencies through apt (optional):

apt-get install python3-redis python3-setproctitle
# .. and: python3-mysqldb or python3-psycopg2

Install PlanB (including depedencies) from PyPI:

pip3 install planb

Install PlanB (including dependencies) from git:

pip3 install git+https://github.com/ossobv/planb.git@master#egg=planb

Set up a local planb user:

adduser planb --disabled-password --home=/var/spool/planb \
  --shell=/bin/bash --system

sudo -H -u planb ssh-keygen -t ed25519      # use elliptic curve
sudo -H -u planb ssh-keygen -t rsa -b 8192  # or use RSA if you're old

Note

You may want to back that ssh key up somewhere.

Set up the local environment:

cat >/etc/planb/envvars <<EOF
USER=planb
PYTHONPATH=/etc/planb
DJANGO_SETTINGS_MODULE=settings
EOF

Note

PlanB looks for an environment file in the locations: - env PLANB_ENVFILE - /etc/planb/envvars - ./envvars The first file that can be loaded will be used.

Set up the local configuration:

cp ${VIRTUAL_ENV:-/usr/local}/share/planb/example_settings.py \
  /etc/planb/settings.py
${EDITOR:-vi} /etc/planb/settings.py

Replace all *FIXME* entries in the ``settings.py``

Note

For development you only need the settings module which can be placed in the project root. cp -n example_settings.py settings.py You can use python setup.py develop to install planb in develop mode. This links the source directory to python site-packages and is especially useful for production hacking.

Make sure the SQL database exists. How to do that is beyond the scope of this readme.

At this point, you should be able to run the planb script.

Set up the database and a web-user:

planb migrate
planb createsuperuser

Set up uwsgi planb.ini:

[uwsgi]
plugin = python3
workers = 4

chdir = /
virtualenv = /srv/virtualenvs/planb
wsgi-file = /srv/virtualenvs/planb/share/planb/wsgi.py

uid = planb
gid = www-data
chmod-socket = 660

for-readline = /etc/planb/envvars
   env = %(_)
endfor =

Set up static path, static files and log path:

# see the STATIC_ROOT entry in your settings.py
install -o planb -d /srv/http/YOURHOSTNAME/static

planb collectstatic

install -o planb -d /var/log/planb

Set up nginx config:

server {
    listen 80;
    server_name YOURHOSTNAME;

    root /srv/http/YOURHOSTNAME;

    location / {
        uwsgi_pass unix:/run/uwsgi/app/planb/socket;
        include uwsgi_params;
    }
    location = /favicon.ico {
        return 404;
    }
    location /static/ {
    }
}

Give PlanB sudo access to ZFS tools and fix paths:

cat >/etc/sudoers.d/planb <<EOF
planb ALL=NOPASSWD: /sbin/zfs, /bin/chown
EOF

zfs create tank/BACKUP -o mountpoint=/srv/backups
chown planb /srv/backups
chmod 700 /srv/backups

(Note that setting up a different mount point is optional. See also README-zpool.rst for additional tips.

Set up qcluster for scheduled tasks:

# (in the source, this file is in rc.d)
cp ${VIRTUAL_ENV:-/usr/local}/share/planb/planb-queue.service \
  /etc/systemd/system/

${EDITOR:-vi} /etc/systemd/system/planb-queue.service

systemctl daemon-reload &&
  systemctl enable planb-queue &&
  systemctl start planb-queue &&
  systemctl status planb-queue

Set up the qcluster for dutree tasks. If you do not use dutree or if you want to run dutree on the default qcluster you can set Q_DUTREE_QUEUE='PlanB' in /etc/planb/settings.py.:

cp ${VIRTUAL_ENV:-/usr/local}/share/planb/planb-queue-dutree.service \
  /etc/systemd/system/

${EDITOR:-vi} /etc/systemd/system/planb-queue-dutree.service

systemctl daemon-reload &&
  systemctl enable planb-queue-dutree &&
  systemctl start planb-queue-dutree &&
  systemctl status planb-queue-dutree

Install automatic jobs:

planb loaddata planb_jobs

Don't forget a logrotate config:

cat >/etc/logrotate.d/planb <<EOF
/var/log/planb/*.log {
        weekly
        missingok
        rotate 52
        compress
        delaycompress
        notifempty
        create 0644 planb www-data
        sharedscripts
}
EOF

Create aliases to quickly mount/unmount the current working directory in your ~/.bashrc:

alias zfs-quick-mount="zfs load-key -L \
    "'"file:///tank/_local/zfskeys/${PWD#/}/_key.bin" "${PWD#/}" &&
    zfs mount "${PWD#/}" && cd .'
alias zfs-quick-umount='cd / && if zfs umount "${OLDPWD#/}"
    then zfs unload-key "${OLDPWD#/}"; cd "${OLDPWD}"
    else cd "${OLDPWD}"; false; fi'

Warning

WARNING: The example above uses local key files! This will be fixed/replaced in upcoming commits.

Configuring a remote host

Create a remotebackup user on the remote host (or encbackup for backups encrypted at the source [1] [2], which is beyond the scope of this document):

useradd -m remotebackup

Configure sudo access using visudo -f /etc/sudoers.d/remotebackup:

# Backup user needs to be able to get the files
remotebackup ALL=NOPASSWD: /usr/bin/rsync --server --sender *
remotebackup ALL=NOPASSWD: /usr/bin/ionice -c2 -n7 /usr/bin/rsync --server --sender *
remotebackup ALL=NOPASSWD: /usr/bin/ionice -c3 /usr/bin/rsync --server --sender *

# Optional, for planb-zfsync.sh (only destroy snapshots with @ in the name)
remotebackup ALL=NOPASSWD: /sbin/zfs destroy *@*
remotebackup ALL=NOPASSWD: /sbin/zfs list *
remotebackup ALL=NOPASSWD: /sbin/zfs send *
remotebackup ALL=NOPASSWD: /sbin/zfs set *
remotebackup ALL=NOPASSWD: /sbin/zfs snapshot *

Observe how the --server --sender makes the rsync read-only.

Set up the ssh key like you'd normally do:

mkdir -p ~remotebackup/.ssh
cat >>~remotebackup/.ssh/authorized_keys <<EOF
... ssh public key from /var/spool/planb/.ssh/id_rsa.pub goes here ...
EOF

chmod 640 ~remotebackup/.ssh/authorized_keys
chown remotebackup -R ~remotebackup/.ssh

When you use this pattern, you can tick use_sudo and set the remote user to remotebackup.

Adding post-backup notification

Do you want a notification when a backup succeeds? Or when it fails?

You can add something like this to your settings:

from datetime import datetime
from subprocess import check_call
from django.dispatch import receiver
from planb.signals import backup_done

@receiver(backup_done)
def notify_zabbix(sender, fileset, success, **kwargs):
    if success:
        key = 'planb.get_latest[{}]'.format(fileset.unique_name)
        val = datetime.now().strftime('%s')
        cmd = (
            'zabbix_sender', '-c', '/etc/zabbix/zabbix_agentd.conf',
            '-k', key, '-o', val)
        check_call(cmd)

That combines nicely with a backup host discovery rule using blist:

# Machine discovery (redirects stderr to mail).
UserParameter=planb.discovery, \
  ( planb blist --zabbix 3>&2 2>&1 1>&3 \
  | mail -E -s 'ERROR: planb.discovery (zabbix)' root ) 2>&1

Doing daily jobs

A quick hack to get daily reports up and running, is by placing something like this in /etc/planb/planb_custom.py:

from planb.contrib.billing import BossoBillingPoster, daily_hostgroup_report

def daily_billing_report():
    """
    This function is added into: Home >> Task Queue >> Scheduled task
    As: "Report to Billing" <planb_custom.daily_bosso_report>
    """
    daily_hostgroup_report(BossoBillingPoster('http://my.url.here/'))

F.A.Q.

Can I use the software and customize it to my own needs?
It is licensed under the GNU GPL version 3.0 or higher. See the LICENSE file for the full text. That means: probably yes, but you may be required to share any changes you make. But you were going to do that anyway, right?
Mails for backup success are sent, but mails for failure are not.
Check the DEBUG setting. At the moment, error-mails are sent through the logging subsystem and that is disabled when running in debug-mode.
Where are the ssh host fingerprints (known_hosts files) stored?

They're in ~planb/.ssh/known_hosts.d/. If you want to ssh manually, you can add this to ~planb/.profile:

ssh() {
    for arg in "$@"; do
        case $arg in
        -*) ;;
        *) break ;;
        esac
    done
    if test -n "$arg"; then
        host=${arg##*@}
        echo "(adding: \
-o UserKnownHostsFile=$HOME/.ssh/known_hosts.d/$host)" >&2
        /usr/bin/ssh -o HashKnownHosts=no \
          -o UserKnownHostsFile=$HOME/.ssh/known_hosts.d/$host "$@"
    else
        /usr/bin/ssh "$@"
    fi
}
Can I use a jump host?
You can add -e 'ssh -J jumpuser@jumphost' to the rsync transport flags. Observe that the known hosts file of target will contain the fingerprint of the jump host.
Are bandwidth limits in place?
Yes, the default for the rsync transport is 10MB/s (megabyte). You can lower or raise this by adding --bwlimit=10M to the transport flags.
I've increased the bwlimit, but it's still slow.

If you notice that you're limited by ssh encryption CPU speed, you can consider setting the preferred ciphers in ~planb/.ssh/config:

Host *
    # The default is:
    #
    #   chacha20-poly1305@openssh.com,
    #   aes128-ctr,aes192-ctr,aes256-ctr,
    #   aes128-gcm@openssh.com,aes256-gcm@openssh.com
    #
    # The available ciphers may be obtained using "ssh -Q cipher".
    # (Adding a non-existent one will yield a "Bad SSH2 cipher spec".)
    #
    # The AES ciphers are commonly hardware/CPU accelerated.
    #
    Ciphers aes128-ctr,aes128-gcm@openssh.com,aes256-ctr,\
        aes256-gcm@openssh.com,chacha20-poly1305@openssh.com,3des-cbc
Removing a fileset does not wipe the filesystem from disk, what should I do?

This is done intentionally. You should periodically use planb slist --stale to check for stale filesystems.

You can them remove them manually using zfs destroy [-r] FILESYSTEM.

Rsync complains about failed to stat or mkdir failed.

If rsync returns these messages:

rsync: recv_generator: failed to stat "...": Permission denied (13)
rsync: recv_generator: mkdir "..." failed: Permission denied (13)

Then you may be looking at parent directories with crooked permissions, like 077. Fix the permissions on the remote end.

However, many of these problems have likely been fixed by the addition of the --chmod=Du+rwx rsync option.

Rsync complains about Invalid or incomplete multibyte or wide character.

If rsync returns with code 23 and says this:

rsync: recv_generator: failed to stat "...\#351es-BCS 27-09-11.csv":
  Invalid or incomplete multibyte or wide character (84)

Then you might be backing up old hosts with legacy Latin-1 encoding on the filesystem. Adding --iconv=utf8,latin1 to the rsync transport flags should fix it.

You may need rsync version 3 or higher for that.

Right now we opt to not implement any of these workarounds:

  • Patch rsync to cope with EILSEQ (84) "Illegal byte sequence".
  • Cope with error code 23 and pretend that everything went fine.

Instead, you should install a recent rsync and/or fix the filenames on your remote filesystem.

The mkvirtualenv said locale.Error: unsupported locale setting.
You need to install the right locales until perl -e setlocale is silent. How depends on your system and your config. See locale and e.g. locale-gen en_US.UTF-8.
The uwsgi log complains about "No module named site".

If your uwsgi fails to start, and the log looks like this:

Python version: 2.7.12 (default, Nov 19 2016, 06:48:10)
Set PythonHome to /srv/virtualenvs/planb
ImportError: No module named site

Then your uWSGI is missing the Python 3 module. Go install uwsgi-plugin-python3.

Authors

PlanB was started in 2013 as "OSSO backup" by Alex Boonstra at OSSO B.V. Since then, it has been evolved into PlanB. When it was Open Sourced by Walter Doekes in 2017, the old commits were dropped to ensure that any private company information was not disclosed. Since then, Harm Geerts has also been busy on the project.

Footnotes

[1] If you want your data encrypted before it gets sent to the PlanB server, check out the OSSO blog: on the fly encrypted backups using gocryptfs (2020)
[2] An older OSSO blog about on the fly encryption at the source: on the fly encrypted backups using encfs (2015)