Tape Backup =========== Proxmox tape backup provides an easy way to store datastore content onto magnetic tapes. This increases data safety because you get: - an additional copy of the data - to a different media type (tape) - to an additional location (you can move tapes offsite) In most restore jobs, only data from the last backup job is restored. Restore requests further decline the older the data gets. Considering this, tape backup may also help to reduce disk usage, because you can safely remove data from disk once archived on tape. This is especially true if you need to keep data for several years. Tape backups do not provide random access to the stored data. Instead, you need to restore the data to disk before you can access it again. Also, if you store your tapes offsite (using some kind of tape vaulting service), you need to bring them onsite before you can do any restore. So please consider that restores from tapes can take much longer than restores from disk. Tape Technology Primer ---------------------- .. _Linear Tape Open: https://en.wikipedia.org/wiki/Linear_Tape-Open As of 2021, the only broadly available tape technology standard is `Linear Tape Open`_, and different vendors offers LTO Ultrium tape drives, autoloaders and LTO tape cartridges. There are a few vendors offering proprietary drives with slight advantages in performance and capacity, but they have significant disadvantages: - proprietary (single vendor) - a much higher purchase cost So we currently do not test such drives. In general, LTO tapes offer the following advantages: - Durable (30 years) - High Capacity (12 TB) - Relatively low cost per TB - Cold Media - Movable (storable inside vault) - Multiple vendors (for both media and drives) - Build in AES-CGM Encryption engine Please note that `Proxmox Backup Server` already stores compressed data, so we do not need/use the tape compression feature. Supported Hardware ------------------ Proxmox Backup Server supports `Linear Tape Open`_ genertion 4 (LTO4) or later. In general, all SCSI2 tape drives supported by the Linux kernel should work, but feature like hardware encryptions needs LTO4 or later. Tape changer support is done using the Linux 'mtx' command line tool. So any changer device supported by that tool should work. Drive Performance ~~~~~~~~~~~~~~~~~ Current LTO-8 tapes provide read/write speeds up to 360MB/s. This means, that it still takes a minimum of 9 hours to completely write or read a single tape (even at maximum speed). The only way to speed up that data rate is to use more than one drive. That way you can run several backup jobs in parallel, or run restore jobs while the other dives are used for backups. Also consider that you need to read data first from your datastore (disk). But a single spinning disk is unable to deliver data at this rate. We measured a maximum rate of about 60MB/s to 100MB/s in practice, so it takes 33 hours to read 12TB to fill up an LTO-8 tape. If you want to run your tape at full speed, please make sure that the source datastore is able to deliver that performance (e.g, by using SSDs). Terminology ----------- :Tape Labels: are used to uniquely indentify a tape. You normally use some sticky paper labels and apply them on the front of the cartridge. We additionally store the label text magnetically on the tape (first file on tape). .. _Code 39: https://en.wikipedia.org/wiki/Code_39 .. _LTO Ultrium Cartridge Label Specification: https://www.ibm.com/support/pages/ibm-lto-ultrium-cartridge-label-specification .. _LTO Barcode Generator: lto-barcode/index.html :Barcodes: are a special form of tape labels, which are electronically readable. Most LTO tape robots use an 8 character string encoded as `Code 39`_, as definded in the `LTO Ultrium Cartridge Label Specification`_. You can either buy such barcode labels from your cartridge vendor, or print them yourself. You can use our `LTO Barcode Generator`_ App for that. .. Note:: Physical labels and the associated adhesive shall have an environmental performance to match or exceed the environmental specifications of the cartridge to which it is applied. :Media Pools: A media pool is a logical container for tapes. A backup job targets one media pool, so a job only uses tapes from that pool. The pool additionally defines how long a backup job can append data to tapes (allocation policy) and how long you want to keep the data (retention policy). :Media Set: A group of continuously written tapes (all from the same media pool). :Tape drive: The decive used to read and write data to the tape. There are standalone drives, but drives often ship within tape libraries. :Tape changer: A device which can change the tapes inside a tape drive (tape robot). They are usually part of a tape library. .. _Tape Library: https://en.wikipedia.org/wiki/Tape_library :`Tape library`_: A storage device that contains one or more tape drives, a number of slots to hold tape cartridges, a barcode reader to identify tape cartridges and an automated method for loading tapes (a robot). People als call this 'autoloader', 'tape robot' or 'tape jukebox'. :Inventory: The inventory stores the list of known tapes (with additional status information). :Catalog: A media catalog stores information about the media content. Tape Quickstart --------------- 1. Configure your tape hardware (drives and changers) 2. Configure one or more media pools 3. Label your tape cartridges. 4. Start your first tape backup job ... Configuration ------------- Please note that you can configure anything using the graphical user interface or the command line interface. Both methods results in the same configuration. Tape changers ~~~~~~~~~~~~~ Tape changers (robots) are part of a `Tape Library`_. You can skip this step if you are using a standalone drive. Linux is able to auto detect those devices, and you can get a list of available devices using:: # proxmox-tape changer scan ┌─────────────────────────────┬─────────┬──────────────┬────────┐ │ path │ vendor │ model │ serial │ ╞═════════════════════════════╪═════════╪══════════════╪════════╡ │ /dev/tape/by-id/scsi-CC2C52 │ Quantum │ Superloader3 │ CC2C52 │ └─────────────────────────────┴─────────┴──────────────┴────────┘ In order to use that device with Proxmox, you need to create a configuration entry:: # proxmox-tape changer create sl3 --path /dev/tape/by-id/scsi-CC2C52 Where ``sl3`` is an arbitrary name you can choose. .. Note:: Please use stable device path names from inside ``/dev/tape/by-id/``. Names like ``/dev/sg0`` may point to a different device after reboot, and that is not what you want. You can show the final configuration with:: # proxmox-tape changer config sl3 ┌──────┬─────────────────────────────┐ │ Name │ Value │ ╞══════╪═════════════════════════════╡ │ name │ sl3 │ ├──────┼─────────────────────────────┤ │ path │ /dev/tape/by-id/scsi-CC2C52 │ └──────┴─────────────────────────────┘ Or simply list all configured changer devices:: # proxmox-tape changer list ┌──────┬─────────────────────────────┬─────────┬──────────────┬────────────┐ │ name │ path │ vendor │ model │ serial │ ╞══════╪═════════════════════════════╪═════════╪══════════════╪════════════╡ │ sl3 │ /dev/tape/by-id/scsi-CC2C52 │ Quantum │ Superloader3 │ CC2C52 │ └──────┴─────────────────────────────┴─────────┴──────────────┴────────────┘ The Vendor, Model and Serial number are auto detected, but only shown if the device is online. To test your setup, please query the status of the changer device with:: # proxmox-tape changer status sl3 ┌───────────────┬──────────┬────────────┬─────────────┐ │ entry-kind │ entry-id │ changer-id │ loaded-slot │ ╞═══════════════╪══════════╪════════════╪═════════════╡ │ drive │ 0 │ vtape1 │ 1 │ ├───────────────┼──────────┼────────────┼─────────────┤ │ slot │ 1 │ │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ slot │ 2 │ vtape2 │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ ... │ ... │ │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ slot │ 16 │ │ │ └───────────────┴──────────┴────────────┴─────────────┘ Tape libraries usually provide some special import/export slots (also called "mail slots"). Tapes inside those slots are acessible from outside, making it easy to add/remove tapes to/from the library. Those tapes are considered to be "offline", so backup jobs will not use them. Those special slots are auto-detected and marked as ``import-export`` slot in the status command. It's worth noting that some of the smaller tape libraries don't have such slots. While they have something called "Mail Slot", that slot is just a way to grab the tape from the gripper. But they are unable to hold media while the robot does other things. They also do not expose that "Mail Slot" over the SCSI interface, so you wont see them in the status output. As a workaround, you can mark some of the normal slots as export slot. The software treats those slots like real ``import-export`` slots, and the media inside those slots is considered to be 'offline' (not available for backup):: # proxmox-tape changer update sl3 --export-slots 15,16 After that, you can see those artificial ``import-export`` slots in the status output:: # proxmox-tape changer status sl3 ┌───────────────┬──────────┬────────────┬─────────────┐ │ entry-kind │ entry-id │ changer-id │ loaded-slot │ ╞═══════════════╪══════════╪════════════╪═════════════╡ │ drive │ 0 │ vtape1 │ 1 │ ├───────────────┼──────────┼────────────┼─────────────┤ │ import-export │ 15 │ │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ import-export │ 16 │ │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ slot │ 1 │ │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ slot │ 2 │ vtape2 │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ ... │ ... │ │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ slot │ 14 │ │ │ └───────────────┴──────────┴────────────┴─────────────┘ Tape drives ~~~~~~~~~~~ Linux is able to auto detect tape drives, and you can get a list of available tape drives using:: # proxmox-tape drive scan ┌────────────────────────────────┬────────┬─────────────┬────────┐ │ path │ vendor │ model │ serial │ ╞════════════════════════════════╪════════╪═════════════╪════════╡ │ /dev/tape/by-id/scsi-12345-nst │ IBM │ ULT3580-TD4 │ 12345 │ └────────────────────────────────┴────────┴─────────────┴────────┘ In order to use that drive with Proxmox, you need to create a configuration entry:: # proxmox-tape drive create mydrive --path /dev/tape/by-id/scsi-12345-nst .. Note:: Please use stable device path names from inside ``/dev/tape/by-id/``. Names like ``/dev/nst0`` may point to a different device after reboot, and that is not what you want. If you have a tape library, you also need to set the associated changer device:: # proxmox-tape drive update mydrive --changer sl3 --changer-drivenum 0 The ``--changer-drivenum`` is only necessary if the tape library includes more than one drive (The changer status command lists all drivenums). You can show the final configuration with:: # proxmox-tape drive config mydrive ┌─────────┬────────────────────────────────┐ │ Name │ Value │ ╞═════════╪════════════════════════════════╡ │ name │ mydrive │ ├─────────┼────────────────────────────────┤ │ path │ /dev/tape/by-id/scsi-12345-nst │ ├─────────┼────────────────────────────────┤ │ changer │ sl3 │ └─────────┴────────────────────────────────┘ .. NOTE:: The ``changer-drivenum`` value 0 is not stored in the configuration, because that is the default. To list all configured drives use:: # proxmox-tape drive list ┌──────────┬────────────────────────────────┬─────────┬────────┬─────────────┬────────┐ │ name │ path │ changer │ vendor │ model │ serial │ ╞══════════╪════════════════════════════════╪═════════╪════════╪═════════════╪════════╡ │ mydrive │ /dev/tape/by-id/scsi-12345-nst │ sl3 │ IBM │ ULT3580-TD4 │ 12345 │ └──────────┴────────────────────────────────┴─────────┴────────┴─────────────┴────────┘ The Vendor, Model and Serial number are auto detected, but only shown if the device is online. For testing, you can simply query the drive status with:: # proxmox-tape status --drive mydrive ┌───────────┬────────────────────────┐ │ Name │ Value │ ╞═══════════╪════════════════════════╡ │ blocksize │ 0 │ ├───────────┼────────────────────────┤ │ status │ DRIVE_OPEN | IM_REP_EN │ └───────────┴────────────────────────┘ .. NOTE:: Blocksize should always be 0 (variable block size mode). This is the default anyways. Media Pools ~~~~~~~~~~~ A media pool is a logical container for tapes. A backup job targets one media pool, so a job only uses tapes from that pool. .. topic:: Media Set A media set is a group of continuously written tapes, used to split the larger pool into smaller, restorable units. One or more backup jobs write to a media set, producing an ordered group of tapes. Media sets are identified by an unique ID. That ID and the sequence number is stored on each tape of that set (tape label). Media sets are the basic unit for restore tasks, i.e. you need all tapes in the set to restore the media set content. Data is fully deduplicated inside a media set. .. topic:: Media Set Allocation Policy The pool additionally defines how long backup jobs can append data to a media set. The following settings are possible: - Try to use the current media set. This setting produce one large media set. While this is very space efficient (deduplication, no unused space), it can lead to long restore times, because restore jobs needs to read all tapes in the set. .. NOTE:: Data is fully deduplicated inside a media set. That also means that data is randomly distributed over the tapes in the set. So even if you restore a single VM, this may have to read data from all tapes inside the media set. Larger media sets are also more error prone, because a single damaged media makes the restore fail. Usage scenario: Mostly used with tape libraries, and you manually trigger new set creation by running a backup job with the ``--export`` option. .. NOTE:: Retention period starts with the existence of a newer media set. - Always create a new media set. With this setting each backup job creates a new media set. This is less space efficient, because the last media from the last set may not be fully written, leaving the remaining space unused. The advantage is that this procudes media sets of minimal size. Small set are easier to handle, you can move sets to an off-site vault, and restore is much faster. .. NOTE:: Retention period starts with the creation time of the media set. - Create a new set when the specified Calendar Event triggers. .. _systemd.time manpage: https://manpages.debian.org/buster/systemd/systemd.time.7.en.html This allows you to specify points in time by using systemd like Calendar Event specifications (see `systemd.time manpage`_). For example, the value ``weekly`` (or ``Mon *-*-* 00:00:00``) will create a new set each week. This balances between space efficency and media count. .. NOTE:: Retention period starts when the calendar event triggers. Additionally, the following events may allocate a new media set: - Required tape is offline (and you use a tape library). - Current set contains damaged of retired tapes. - Media pool encryption changed - Database consistency errors, e.g. if the inventory does not contain required media info, or contain conflicting infos (outdated data). .. topic:: Retention Policy Defines how long we want to keep the data. - Always overwrite media. - Protect data for the duration specified. We use systemd like time spans to specify durations, e.g. ``2 weeks`` (see `systemd.time manpage`_). - Never overwrite data. .. topic:: Hardware Encryption LTO4 (or later) tape drives support hardware encryption. If you configure the media pool to use encryption, all data written to the tapes is encrypted using the configured key. That way, unauthorized users cannot read data from the media, e.g. if you loose a media while shipping to an offsite location. .. Note:: If the backup client also encrypts data, data on tape will be double encrypted. The password protected key is stored on each media, so it is possbible to `restore the key `_ using the password. Please make sure you remember the password in case you need to restore the key. .. NOTE:: FIXME: Add note about global content namespace. (We do not store the source datastore, so it is impossible to distinguish store1:/vm/100 from store2:/vm/100. Please use different media pools if the source is from a different name space) The following command creates a new media pool:: // proxmox-tape pool create --drive [OPTIONS] # proxmox-tape pool create daily --drive mydrive Additional option can be set later using the update command:: # proxmox-tape pool update daily --allocation daily --retention 7days To list all configured pools use:: # proxmox-tape pool list ┌───────┬──────────┬────────────┬───────────┬──────────┐ │ name │ drive │ allocation │ retention │ template │ ╞═══════╪══════════╪════════════╪═══════════╪══════════╡ │ daily │ mydrive │ daily │ 7days │ │ └───────┴──────────┴────────────┴───────────┴──────────┘ Tape Jobs ~~~~~~~~~ Administration -------------- Many sub-command of the ``proxmox-tape`` command line tools take a parameter called ``--drive``, which specifies the tape drive you want to work on. For convenience, you can set that in an environment variable:: # export PROXMOX_TAPE_DRIVE=mydrive You can then omit the ``--drive`` parameter from the command. If the drive has an associated changer device, you may also omit the changer parameter from commands that needs a changer device, for example:: # proxmox-tape changer status Should displays the changer status of the changer device associated with drive ``mydrive``. Label Tapes ~~~~~~~~~~~ By default, tape cartidges all looks the same, so you need to put a label on them for unique identification. So first, put a sticky paper label with some human readable text on the cartridge. If you use a `Tape Library`_, you should use an 8 character string encoded as `Code 39`_, as definded in the `LTO Ultrium Cartridge Label Specification`_. You can either bye such barcode labels from your cartidge vendor, or print them yourself. You can use our `LTO Barcode Generator`_ App for that. Next, you need to write that same label text to the tape, so that the software can uniquely identify the tape too. For a standalone drive, manually insert the new tape cartidge into the drive and run:: # proxmox-tape label --changer-id [--pool ] You may omit the ``--pool`` argument to allow the tape to be used by any pool. .. Note:: For safety reasons, this command fails if the tape contain any data. If you want to overwrite it anways, erase the tape first. You can verify success by reading back the label:: # proxmox-tape read-label ┌─────────────────┬──────────────────────────────────────┐ │ Name │ Value │ ╞═════════════════╪══════════════════════════════════════╡ │ changer-id │ vtape1 │ ├─────────────────┼──────────────────────────────────────┤ │ uuid │ 7f42c4dd-9626-4d89-9f2b-c7bc6da7d533 │ ├─────────────────┼──────────────────────────────────────┤ │ ctime │ Wed Jan 6 09:07:51 2021 │ ├─────────────────┼──────────────────────────────────────┤ │ pool │ daily │ ├─────────────────┼──────────────────────────────────────┤ │ media-set-uuid │ 00000000-0000-0000-0000-000000000000 │ ├─────────────────┼──────────────────────────────────────┤ │ media-set-ctime │ Wed Jan 6 09:07:51 2021 │ └─────────────────┴──────────────────────────────────────┘ .. NOTE:: The ``media-set-uuid`` using all zeros indicates an empty tape (not used by any media set). If you have a tape library, apply the sticky barcode label to the tape cartridges first. Then load those empty tapes into the library. You can then label all unlabeled tapes with a single command:: # proxmox-tape barcode-label [--pool ] Run Tape Backups ~~~~~~~~~~~~~~~~ To manually run a backup job use:: # proxmox-tape backup [OPTIONS] The following options are available: --eject-media Eject media upon job completion. It is normally good practice to eject the tape after use. This unmounts the tape from the drive and prevents the tape from getting dirty with dust. --export-media-set Export media set upon job completion. After a sucessful backup job, this moves all tapes from the used media set into import-export slots. The operator can then pick up those tapes and move them to a media vault. Restore from Tape ~~~~~~~~~~~~~~~~~ Restore is done at media-set granularity, so you first need to find out which media set contains the data you want to restore. This information is stored in the media catalog. If you do not have media catalogs, you need to restore them first. Please note that you need the catalog to find your data, but restoring a complete media-set does not need media catalogs. The following command shows the media content (from catalog):: # proxmox-tape media content ┌────────────┬──────┬──────────────────────────┬────────┬────────────────────────────────┬──────────────────────────────────────┐ │ label-text │ pool │ media-set-name │ seq-nr │ snapshot │ media-set-uuid │ ╞════════════╪══════╪══════════════════════════╪════════╪════════════════════════════════╪══════════════════════════════════════╡ │ TEST01L8 │ p2 │ Wed Jan 13 13:55:55 2021 │ 0 │ vm/201/2021-01-11T10:43:48Z │ 9da37a55-aac7-4deb-91c6-482b3b675f30 │ ├────────────┼──────┼──────────────────────────┼────────┼────────────────────────────────┼──────────────────────────────────────┤ │ ... │ ... │ ... │ ... │ ... │ ... │ └────────────┴──────┴──────────────────────────┴────────┴────────────────────────────────┴──────────────────────────────────────┘ A restore job reads the data from the media set and moves it back to data disk (datastore):: // proxmox-tape restore # proxmox-tape restore 9da37a55-aac7-4deb-91c6-482b3b675f30 mystore Update Inventory ~~~~~~~~~~~~~~~~ Restore Catalog ~~~~~~~~~~~~~~~ Encryption Key Management ~~~~~~~~~~~~~~~~~~~~~~~~~ Creating a new encryption key:: # proxmox-tape key create --hint "tape pw 2020" Tape Encryption Key Password: ********** Verify Password: ********** "14:f8:79:b9:f5:13:e5:dc:bf:b6:f9:88:48:51:81:dc:79:bf:a0:22:68:47:d1:73:35:2d:b6:20:e1:7f:f5:0f" List existing encryption keys:: # proxmox-tape key list ┌───────────────────────────────────────────────────┬───────────────┐ │ fingerprint │ hint │ ╞═══════════════════════════════════════════════════╪═══════════════╡ │ 14:f8:79:b9:f5:13:e5:dc: ... :b6:20:e1:7f:f5:0f │ tape pw 2020 │ └───────────────────────────────────────────────────┴───────────────┘ To show encryption key details:: # proxmox-tape key show 14:f8:79:b9:f5:13:e5:dc:...:b6:20:e1:7f:f5:0f ┌─────────────┬───────────────────────────────────────────────┐ │ Name │ Value │ ╞═════════════╪═══════════════════════════════════════════════╡ │ kdf │ scrypt │ ├─────────────┼───────────────────────────────────────────────┤ │ created │ Sat Jan 23 14:47:21 2021 │ ├─────────────┼───────────────────────────────────────────────┤ │ modified │ Sat Jan 23 14:47:21 2021 │ ├─────────────┼───────────────────────────────────────────────┤ │ fingerprint │ 14:f8:79:b9:f5:13:e5:dc:...:b6:20:e1:7f:f5:0f │ ├─────────────┼───────────────────────────────────────────────┤ │ hint │ tape pw 2020 │ └─────────────┴───────────────────────────────────────────────┘ The ``paperkey`` subcommand can be used to create a QR encoded version of a tape encryption key. The following command sends the output of the ``paperkey`` command to a text file, for easy printing:: proxmox-tape key paperkey --output-format text > qrkey.txt .. _restore_encryption_key: Restoring Encryption Keys ^^^^^^^^^^^^^^^^^^^^^^^^^ You can restore the encryption key from the tape, using the password used to generate the key. First, load the tape you want to restore into the drive. Then run:: # proxmox-tape key restore Tepe Encryption Key Password: *********** If the password is correct, the key will get imported to the database. Further restore jobs automatically use any availbale key. Tape Cleaning ~~~~~~~~~~~~~ LTO tape drives requires regular cleaning. This is done by loading a cleaning cartridge into the drive, which is a manual task for standalone drives. For tape libraries, cleaning cartridges are identified using special labels starting with letters "CLN". For example, our tape library has a cleaning cartridge inside slot 3:: # proxmox-tape changer status sl3 ┌───────────────┬──────────┬────────────┬─────────────┐ │ entry-kind │ entry-id │ changer-id │ loaded-slot │ ╞═══════════════╪══════════╪════════════╪═════════════╡ │ drive │ 0 │ vtape1 │ 1 │ ├───────────────┼──────────┼────────────┼─────────────┤ │ slot │ 1 │ │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ slot │ 2 │ vtape2 │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ slot │ 3 │ CLN001CU │ │ ├───────────────┼──────────┼────────────┼─────────────┤ │ ... │ ... │ │ │ └───────────────┴──────────┴────────────┴─────────────┘ To initiate a cleaning operation simply run:: # proxmox-tape clean This command does the following: - find the cleaning tape (in slot 3) - unload the current media from the drive (back to slot1) - load the cleaning tape into the drive - run drive cleaning operation - unload the cleaning tape (to slot 3) Configuration Files ------------------- ``media-pool.cfg`` ~~~~~~~~~~~~~~~~~~ File Format ^^^^^^^^^^^ .. include:: config/media-pool/format.rst Options ^^^^^^^ .. include:: config/media-pool/config.rst ``tape.cfg`` ~~~~~~~~~~~~~~~~~~ File Format ^^^^^^^^^^^ .. include:: config/tape/format.rst Options ^^^^^^^ .. include:: config/tape/config.rst Command Syntax -------------- ``proxmox-tape`` ---------------- .. include:: proxmox-tape/synopsis.rst ``pmt`` -------- .. include:: pmt/options.rst .... .. include:: pmt/synopsis.rst ``pmtx`` -------- .. include:: pmtx/synopsis.rst