obnam man page

obnam — make, restore, and manipulate backups


obnam [--always-restore-setuid] [--no-always-restore-setuid] [--checkpoint=SIZE] [--checksum-algorithm=CHECKSUM] [--chunk-bag-size=SIZE] [--chunk-cache-size=SIZE] [--chunk-size=SIZE] [--chunkids-per-group=NUM] [--client-name=CLIENT-NAME] [--compress-with=PROGRAM] [--config=FILE] [--dump-config] [--dump-setting-names] [--generate-manpage=TEMPLATE] [-h] [--help] [--help-all] [--list-config-files] [--version] [--no-default-configs] [--crash-limit=COUNTER] [--critical-age=AGE] [--deduplicate=MODE] [--dir-bag-size=SIZE] [--dir-cache-size=SIZE] [--dump-memory-profile=METHOD] [--dump-repo-file-metadata] [--no-dump-repo-file-metadata] [--encrypt-with=ENCRYPT-WITH] [--exclude=EXCLUDE] [--exclude-caches] [--no-exclude-caches] [--exclude-from=FILE] [--fsck-fix] [--no-fsck-fix] [--fsck-ignore-chunks] [--no-fsck-ignore-chunks] [--fsck-ignore-client=NAME] [--fsck-last-generation-only] [--no-fsck-last-generation-only] [--fsck-rm-unused] [--no-fsck-rm-unused] [--fsck-skip-checksums] [--no-fsck-skip-checksums] [--fsck-skip-dirs] [--no-fsck-skip-dirs] [--fsck-skip-files] [--no-fsck-skip-files] [--fsck-skip-generations] [--no-fsck-skip-generations] [--fsck-skip-per-client-b-trees] [--no-fsck-skip-per-client-b-trees] [--fsck-skip-shared-b-trees] [--no-fsck-skip-shared-b-trees] [--fuse-opt=FUSE] [--generation=GENERATION] [--gnupghome=HOMEDIR] [--idpath-bits=IDPATH-BITS] [--idpath-depth=IDPATH-DEPTH] [--idpath-skip=IDPATH-SKIP] [--include=INCLUDE] [--keep=KEEP] [--key-details] [--no-key-details] [--keyid=KEYID] [--leave-checkpoints] [--no-leave-checkpoints] [--lock-timeout=TIMEOUT] [--log=FILE] [--log-keep=N] [--log-level=LEVEL] [--log-max=SIZE] [--log-mode=MODE] [--lru-size=SIZE] [--memory-dump-interval=SECONDS] [--node-size=SIZE] [--one-file-system] [--no-one-file-system] [--output=FILE] [--pretend] [--dry-run] [--no-act] [--no-pretend] [--no-dry-run] [--no-no-act] [--pretend-time=TIMESTAMP] [--pure-paramiko] [--no-pure-paramiko] [--quiet] [--silent] [--no-quiet] [--no-silent] [-rURL] [--repository=URL] [--repository-format=FORMAT] [--root=URL] [--sftp-delay=SFTP-DELAY] [--small-files-in-btree] [--no-small-files-in-btree] [--ssh-command=EXECUTABLE] [--ssh-host-keys-check=VALUE] [--ssh-key=FILENAME] [--ssh-known-hosts=FILENAME] [--strict-ssh-host-keys] [--no-strict-ssh-host-keys] [--symmetric-key-bits=BITS] [--testing-fail-matching=REGEXP] [--to=TO] [--trace=TRACE] [--upload-queue-size=SIZE] [--verbose] [--no-verbose] [--verify-randomly=N] [--warn-age=AGE] [--weak-random] [--no-weak-random]

obnam [options] _lock
obnam [options] add-key [CLIENT-NAME]...
obnam [options] backup [DIRECTORY|URL]...
obnam [options] client-keys
obnam [options] clients
obnam [options] diff [GENERATION1]GENERATION2
obnam [options] dump-repo
obnam [options] force-lock
obnam [options] forget [GENERATION]...
obnam [options] fsck
obnam [options] generations
obnam [options] genids
obnam [options] help
obnam [options] help-all
obnam [options] kdirstat [FILE]...
obnam [options] list-errors
obnam [options] list-formats
obnam [options] list-keys
obnam [options] list-toplevels
obnam [options] ls [FILE]...
obnam [options] mount [ROOT]
obnam [options] nagios-last-backup-age
obnam [options] remove-client [CLIENT-NAME]...
obnam [options] remove-key [CLIENT-NAME]...
obnam [options] restore [DIRECTORY]...
obnam [options] verify [DIRECTORY]...


obnam makes, restores, manipulates, and otherwise deals with backups. It can store backups on a local disk or to a server via sftp. Every backup generation looks like a fresh snapshot, but is really incremental: the user does not need to worry whether it's a full backup or not. Only changed data is backed up, and if a chunk of data is already backed up in another file, that data is re-used.

The place where backed up data is placed is called the backup repository. A repository may be, for example, a directory on an sftp server, or a directory on a USB hard disk. A single repository may contain backups from several clients. Their data will intermingle as if they were using separate repositories, but if one client backs up a file, the others may re-use the data.

obnam command line syntax consists of a command possibly followed by arguments. The commands are list below.

backup makes a new backup. The first time it is run, it makes a full backup, after that an incremental one.
restore is the opposite of a backup. It copies backed up data from the backup repository to a target directory. You can restore everything in a generation, or just selected files.
clients lists the clients that are backed up to the repository.
generations lists every backup generation for a given client, plus some metadata about the generation.
genids lists the identifier for every backup generation for a given client. No other information is shown. This can be useful for scripting.
ls lists the contents of a given generation, similar to ls -lAR.
kdirstat lists the contents of a given generation, in a format which is compatible with the kdirstat cache file format, which can then be used to visualise the contents of a backup.
verify compares data in the backup with actual user data, and makes sure they are identical. It is most useful to run immediately after a backup, to check that it actually worked. It can be run at any time, but if the user data has changed, verification fails even though the backup is OK.
forget removes backup generations that are no longer wanted, so that they don't use disk space. Note that after a backup generation is removed the data can't be restored anymore. You can either specify the generations to remove by listing them on the command line, or use the --keep option to specify a policy for what to keep (everything else will be removed).
fsck checks the internal consistency of the backup repository. It verifies that all clients, generations, directories, files, and all file contents still exists in the backup repository. It may take quite a long time to run.
force-lock removes lock files for aall clients in the repository. You should only force a lock if you are sure no client is accessing the repository. A dangling lock might happen, for example, if "obnam backup" loses its network connection to the backup repository.
client-keys lists the encryption key associated with each client.
list-keys lists the keys that can access the repository, and which toplevel directories each key can access. Some of the toplevel directories are shared between clients, others are specific to a client.
list-toplevels is like list-keys, but lists toplevels and which keys can access them.
add-key adds an encryption key to the repository. By default, the key is added only to the shared toplevel directories, but it can also be added to specific clients: list the names of the clients on the command line. They key is given with the --keyid option. Whoever has access to the secret key corresponding to the key id can access the backup repository (the shared toplevels plus specified clients).
remove-key removes a key from the shared toplevel directories, plus any clients specified on the command line.
nagios-last-backup-age is a check that exits with non-zero return if a backup age exceeds a certain threshold. It is suitable for use as a check plugin for nagios. Thresholds can be given the --warn-age and --critical-age options.
diff compares two generations and lists files differing between them. Every output line will be prefixed either by a plus sign (+) for files that were added, a minus sign (-) for files that have been removed or an asterisk (*) for files that have changed. If only one generation ID is specified on the command line that generation will be compared with its direct predecessor. If two IDs have been specified, all changes between those two generations will be listed.

mount makes the backup repository available via a read-only FUSE filesystem. Each backup generation is visible as a subdirectory, named after the generation id. This means you can look at backed up data using normal tools, such as your GUI file manager, or command line tools such as ls(1), diff(1), and cp(1). You can't make new backups with the mount subcommand, but you can restore data easily.

You need to have the FUSE utilities and have permission to use FUSE for this to work. The details will vary between operating systems; in Debian, install the package fuse and add yourself to the fuse group (you may need to log out and back in again).

Making backups

When you run a backup, obnam uploads data into the backup repository. The data is divided into chunks, and if a chunk already exists in the backup repository, it is not uploaded again. This allows obnam to deal with files that have been changed or renamed since the previous backup run. It also allows several backup clients to avoid uploading the same data. If, for example, everyone in the office has a copy of the same sales brochures, only one copy needs to be stored in the backup repository.

Every backup run is a generation. In addition, obnam will make checkpoint generations every now and then. These are exactly like normal generations, but are not guaranteed to be a complete snapshot of the live data. If the backup run needs to be aborted in the middle, the next backup run can continue from the latest checkpoint, avoiding the need to start over from scratch.

If one backup run drops a backup root directory, the older generations will still keep it: nothing changes in the old generations just because there is a new one. If the root was dropped by mistake, it can be added back and the next backup run will re-use the existing data in the backup repository, and will only back up the file metadata (filenames, permissions, etc).

Verifying backups

What good is a backup system you cannot rely on? How can you rely on something you cannot test? The obnam verify command checks that data in the backup repository matches actual user data. It retrieves one or more files from the repository and compares them to the user data. This is essentially the same as doing a restore, then comparing restored files with the original files using cmp(1), but easier to use.

By default, verification happens on all files. You can also specify the files to be verified by listing them on the command line. You should specify the full paths to the files, not relative to the current directory.

The output lists files that fail verification for some reason. If you verify everything, it is likely that some files (e.g., parent directories of backup root) may have changed without it being a problem. Note that you will need to specify the whole path to the files or directories to be verified, not relative to the backup root. You still need to specify at least one of the backup roots on the command line or via the --root option so that obnam will find the filesystem, in case it is a remote one.

URL syntax

Whenever obnam accepts a URL, it can be either a local pathname, or an sftp URL. An sftp URL has the following form:


where domain is a normal Internet domain name for a server, user is your username on that server, port is an optional port number, and path is a pathname on the server side. Like bzr(1), but unlike the sftp URL standard, the pathname is absolute, unless it starts with /~/ in which case it is relative to the user's home directory on the server.

See the Example section for examples of URLs.

You can use sftp URLs for the repository, or the live data (root), but note that due to limitations in the protocol, and its implementation in the paramiko library, some things will not work very well for accessing live data over sftp. Most importantly, the handling of of hardlinks is rather suboptimal. For live data access, you should not end the URL with /~/ and should append a dot at the end in this special case.

Generation specifications

When not using the latest generation, you will need to specify which one you need. This will be done with the --generation option, which takes a generation specification as its argument. The specification is either the word latest, meaning the latest generation (also the default), or a number. See the generations command to see what generations are available, and what their numbers are.

Policy for keeping and removing backup generations

The forget command can follow a policy to automatically keep some and remove other backup generations. The policy is set with the --keep=POLICY option.

POLICY is comma-separated list of rules. Each rule consists of a count and a time period. The time periods are h, d, w, m, and y, for hour, day, week, month, and year.

A policy of 30d means to keep the latest backup for each day when a backup was made, and keep the last 30 such backups. Any backup matched by any policy rule is kept, and any backups in between will be removed, as will any backups older than the oldest kept backup.

As an example, assume backups are taken every hour, on the hour: at 00:00, 01:00, 02:00, and so on, until 23:00. If the forget command is run at 23:15, with the above policy, it will keep the backup taken at 23:00 on each day, and remove every other backup that day. It will also remove backups older than 30 days.

If backups are made every other day, at noon, forget would keep the 30 last backups, or 60 days worth of backups, with the above policy.

Note that obnam will only inspect timestamps in the backup repository, and does not care what the actual current time is. This means that if you stop making new backups, the existing ones won't be removed automatically. In essence, obnam pretends the current time is just after the latest backup when forget is run.

The rules can be given in any order, but will be sorted to ascending order of time period before applied. (It is an error to give two rules for the same period.) A backup generation is kept if it matches any rule.

For example, assume the same backup frequency as above, but a policy of 30d,52w. This will keep the newest daily backup for each day for thirty days, and the newest weekly backup for 52 weeks. Because the hourly backups will be removed daily, before they have a chance to get saved by a weekly rule, the effect is that the 23:00 o'clock backup for each day is saved for a month, and the 23:00 backup on Sundays is saved for a year.

If, instead, you use a policy of 72h,30d,52w, obnam would keep the last 72 hourly backups, and the last backup of each calendar day for 30 days, and the last backup of each calendar week for 52 weeks. If the backup frequency was once per day, obnam would keep the backup of each calendar hour for which a backup was made, for 72 such backups. In other words, it would effectively keep the last 72 daily backups.

Sound confusing? Just think how confused the developer was when writing the code.

If no policy is given, forget will keep everything.

A typical policy might be 72h,7d,5w,12m, which means: keep the last 72 hourly backups, the last 7 daily backups, the last 5 weekly backups and the last 12 monthly backups. If the backups are systematically run on an hourly basis, this will mean keeping hourly backups for three days, daily backups for a week, weekly backups for a month, and monthly backups for a year.

The way the policy works is a bit complicated. Run forget with the --pretend option to make sure you're removing the right ones.

Using encryption

obnam can encrypt all the data it writes to the backup repository. It uses gpg(1) to do the encryption. You need to create a key pair using gpg --gen-key (or use an existing one), and then tell obnam about it using the --encrypt-with option. You may optionally use a separate home directory using the --gnupghome option. By default, the default directory for gpg(1) will be used.

Configuration files

obnam will look for configuration files in a number of locations. See the Files section for a list. All these files together are treated as one big file with the contents of all files concatenated.

The files are in INI format, and only the [config] section is used (any other sections are ignored).

The long names of options are used as keys for configuration variables. Any setting that can be set from the command line can be set in a configuration file, in the [config] section.

For example, the options in the following command line:

obnam --repository=/backup --exclude='.wav$' backup

could be replaced with the following configuration file:

repository: /backup
exclude: .wav$

(You can use either foo=value or foo: value syntax in the files.)

The only unusual thing about the files is the way options that can be used many times are expressed. All values are put in a single logical line, separated by commas (and optionally spaces as well). For example:

exclude = foo, bar, \.mp3$

The above has three values for the exclude option: any files that contain the words foo or bar anywhere in the fully qualified pathname, or files with names ending with a period and mp3 (because the exclusions are regular expressions).

A long logical line can be broken into several physical ones, by starting a new line at white space, and indenting the continuation lines:

exclude = foo,

The above example adds three exclusion patterns.

Multiple clients and locking

obnam supports sharing a repository between multiple clients. The clients can share the file contents (chunks), so that if client A backs up a large file, and client B has the same file, then B does not need to upload the large file to the repository a second time. For this to work without confusion, the clients use a simple locking protocol that allows only one client at a time to modify the shared data structures. Locks do not prevent read-only access at the same time: this allows you to restore while someone else is backing up.

Sometimes a read-only operation happens to access a data structure at the same time as it is being modified. This can result in a crash. It will not result in corrupt data, or incorrect restores. However, you may need to restart the read-only operation after a crash.


restore setuid/setgid bits in restored files, even if not root or backed up file had different owner than user running restore
opposite of --always-restore-setuid
use CHECKSUM for checksum algorithm (not for repository format 6); one of: sha512, sha224, sha256, sha384
name of client (defaults to hostname)
use PROGRAM to compress repository with (one of none, deflate)
for nagios-last-backup-age: maximum age (by default in hours) for the most recent backup before statis is critical. Accepts one char unit specifier (s,m,h,d for seconds, minutes, hours, and days.
dump metadata about files?
opposite of --dump-repo-file-metadata
fill in manual page TEMPLATE
which generation to restore
-h, --help
show this help message and exit
policy for what generations to keep when forgetting
when locking in the backup repository, wait TIMEOUT seconds for an existing lock to go away before giving up
write output to FILE, instead of standard output
--pretend, --dry-run, --no-act
do not actually change anything (works with backup, forget and restore only, and may only simulate approximately real behavior)
--no-pretend, --no-dry-run, --no-no-act
opposite of --pretend
--quiet, --silent
be silent: show only error messages, no progress updates
--no-quiet, --no-silent
opposite of --quiet
-r, --repository=URL
name of backup repository (can be pathname or supported URL)
use FORMAT for new repositories; one of "6", "green-albatross-20160813"
where to restore or FUSE mount; for restores, must be empty or must not exist
be verbose: tell the user more of what is going on and generally make sure the user is aware of what is happening or at least that something is happening and also make sure their screen is getting updates frequently and that there is changes happening all the time so they do not get bored and that they in fact get frustrated by getting distracted by so many updates that they will move into the Gobi desert to live under a rock
opposite of --verbose
verify N files randomly from the backup (default is zero, meaning everything)
show program's version number and exit
for nagios-last-backup-age: maximum age (by default in hours) for the most recent backup before status is warning. Accepts one char unit specifier (s,m,h,d for seconds, minutes, hours, and days.

Backing up

make a checkpoint after a given SIZE
find duplicate data in backed up data and store it only once; three modes are available: never de-duplicate, verify that no hash collisions happen, or (the default) fatalistically accept the risk of collisions
regular expression for pathnames to exclude from backup (can be used multiple times)
exclude directories (and their subdirs) that contain a CACHEDIR.TAG file (see http://www.brynosaurus.com/cachedir/spe… for what it needs to contain, and http://liw.fi/cachedir/ for a helper tool)
opposite of --exclude-caches
read exclude patterns from FILE
regular expression for pathnames to include from backup even if it matches an exclude rule (can be used multiple times)
leave checkpoint generations at the end of a successful backup run
opposite of --leave-checkpoints
exclude directories (and their subdirs) that are in a different filesystem
opposite of --one-file-system
what to backup
this is available only for backwards compatibility; do not use it, and remove it from your configuration
opposite of --small-files-in-btree

Configuration files and settings

add FILE to config files
write out the entire current configuration
write out all names of settings and quit
show all options
list all possible config files
clear list of configuration files to read

Development of Obnam itself

artificially crash the program after COUNTER files written to the repository; this is useful for crash testing the application, and should not be enabled for real use; set to 0 to disable (disabled by default)
pretend it is TIMESTAMP (YYYY-MM-DD HH:MM:SS); this is only useful for testing purposes
add an artificial delay (in milliseconds) to all SFTP transfers
development testing helper: simulate failures during backup for files that match the given regular expressions
add to filename patters for which trace debugging logging happens


PGP key with which to encrypt data in the backup repository
home directory for GPG
show additional user IDs for all keys
opposite of --key-details
PGP key id to add to/remove from the backup repository
size of symmetric key, in bits
use /dev/urandom instead of /dev/random to generate symmetric keys
opposite of --weak-random

Integrity checking (fsck)

should fsck try to fix problems? Implies --fsck-rm-unused
opposite of --fsck-fix
ignore chunks when checking repository integrity (assume all chunks exist and are correct)
opposite of --fsck-ignore-chunks
do not check repository data for client NAME
check only the last generation for each client
opposite of --fsck-last-generation-only
should fsck remove unused chunks?
opposite of --fsck-rm-unused
do not check checksums of files
opposite of --fsck-skip-checksums
do not check anything about directories and their files
opposite of --fsck-skip-dirs
do not check anything about files
opposite of --fsck-skip-files
do not check any generations
opposite of --fsck-skip-generations
do not check per-client B-trees
opposite of --fsck-skip-per-client-b-trees
do not check shared B-trees
opposite of --fsck-skip-shared-b-trees


write log entries to FILE (default is to not write log files at all); use "syslog" to log to system log, "stderr" to log to the standard error output, or "none" to disable logging
keep last N logs (10)
log at LEVEL, one of debug, info, warning, error, critical, fatal (default: info)
rotate logs larger than SIZE, zero for never (default: 0)
set permissions of new log files to MODE (octal; default 0600)

Mounting with FUSE

options to pass directly to Fuse


make memory profiling dumps using METHOD, which is one of: none, simple, or meliae (default: simple)
make memory profiling dumps at least SECONDS apart

Performance tweaking

size of chunks of file data backed up
encode NUM chunk ids per group
chunk id level size
depth of chunk id mapping
chunk id mapping lowest bits skip
size of LRU cache for B-tree nodes
size of B-tree nodes on disk; only affects new B-trees so you may need to delete a client or repository to change this for existing repositories
length of upload queue for B-tree nodes

Repository format green-albatross

approximate maximum size of bag combining many chunk objects
size of in-memory cache for file data chunk objects
approximate maximum size of bags combining many DIR objects
size of in-memory cache for DIR objects


do not use openssh even if available, use paramiko only instead
opposite of --pure-paramiko
alternative executable to be used instead of "ssh" (full path is allowed, no arguments may be added)
If "yes", require that the ssh host key must be known and correct to be accepted. If "no", do not require that. If "ask", the user is interactively asked to accept new hosts. The default ("ssh-config") is to rely on the settings of the underlying SSH client
use FILENAME as the ssh RSA private key for sftp access (default is using keys known to ssh-agent)
filename of the user's known hosts file
DEPRECATED, use --ssh-host-keys-check instead
opposite of --strict-ssh-host-keys

Option values

The SIZE value in options mentioned above specifies a size in bytes, with optional suffixes to indicate kilobytes (k), kibibytes (Ki), megabytes (M), mebibyts (Mi), gigabytes (G), gibibytes (Gi), terabytes (T), tibibytes (Ti). The suffixes are case-insensitive.

Exit Status

obnam will exit with zero if everything went well, and non-zero otherwise.


obnam will pass on the environment it gets from its parent, without modification. It does not obey any unusual environment variables, but it does obey the usual ones when running external programs, creating temporary files, etc.



Configuration files for obnam. It is not an error for any or all of the files to not exist.


To back up your home directory to a server:

obnam backup --repository sftp://your.server/~/backups $HOME

To restore your latest backup from the server:

obnam restore --repository sftp://your.server/~/backups \
--to /var/tmp/my.home.dir

To restore just one file or directory:

obnam restore --repository sftp://your.server/~/backups \
--to /var/tmp/my.home.dir $HOME/myfile.txt

Alternatively, mount the backup repository using the FUSE filesystem (note that the --to option is necessary):

mkdir my-repo
obnam mount --repository sftp://your.server/~/backups \
--to my-repo
cp my-repo/latest/$HOME/myfile.txt
fusermount -u my-repo

To check that the backup worked:

obnam verify --repository sftp://your.server/~/backups \

To remove old backups, keeping the newest backup for each day for ten years:

obnam forget --repository sftp://your.server/~/backups \
--keep 3650d

To verify that the backup repository is OK:

obnam fsck --repository sftp://your.server/~/backups

To view the backed up files in the backup repository using FUSE:

obnam mount --to my-fuse
ls -lh my-fuse
fusermount -u my-fuse

See Also

obnam comes with a manual in HTML and PDF forms. See /usr/share/doc/obnam if you have Obnam installed system-wide, or in the subdirectory manual in the source tree.


Referenced By

obnam-viewprof(1), seivot(1), seivots-summary(1).