doveadm-backup(1) - Dovecot's one-way mailbox synchronization feature

SYNOPSIS

doveadm [GLOBAL OPTIONS] sync [-u user | -A | -F file | --no-userdb-lookup] [-S socket_path] [-1fPR] [-l secs] [-r rawlog path] [-m mailbox] [-g mailbox guid] [-n namespace | -N] [-x exclude] [-a all mailbox] [-s state] [-T secs] [-t start date] [-e end date] [-O sync flag] [-I max size] -d | destination

doveadm [GLOBAL OPTIONS] backup [-u user | -A | -F file | --no-userdb-lookup] [-S socket_path] [-fPR] [-l secs] [-r rawlog path] [-m mailbox] [-g mailbox guid] [-n namespace | -N] [-x exclude] [-a all mailbox] [-s state] [-T secs] [-t start date] [-e end date] [-O sync flag] [-I max size] -d | destination

DESCRIPTION

dsync (short for doveadm sync) is Dovecot's mailbox synchronization feature. It can be used for several different use cases: Two-way synchronization of mailboxes, creating backups of mails, and convert mailboxes from/to different mailbox formats. All of these can be used within the same server or between different servers (via ssh(1) or tcp connections). Remote mailboxes can be accessed also via IMAP protocol, which allows using dsync for mailbox migration purposes.

You can run dsync in one of three modes:

• doveadm backup performs one-way synchronization. If there are any changes in the destination they will be reverted, so the destination will look exactly like the source.

• doveadm sync performs two-way synchronization. It merges all changes without losing anything. Both the mailboxes will end up looking identical after the synchronization is finished.

• doveadm sync -1 performs one-way synchronization. If there are any changes in the destination, they will be preserved and the new changes will be merged on top of them. This merging doesn't currently work perfectly, so its use should be limited. Its main purpose is that during mailbox migration you can run doveadm backup multiple times, then switch mails to be delivered to the new mailbox and run doveadm sync -1 once more to transfer any last new mails from the old mailbox.

  • The one-way algorithm is the same as two-way dsync algorithm except the source account is not modified. It fetches the message's GUID (Global UID), which is used to identify any conflicting UIDs in messages. As long as the source and destination side has matching UID<->GUID mapping, those emails are assumed to be synced correctly. Only after the first mismatch will changes begin.

    Example: Source mailbox has messages UID 1..5; source mailbox is sync'd using doveadm backup to the destination. Subsequently, UID 6 is delivered to the source mailbox and UID 1 is expunged from the destination mailbox. In this example, UID 1 is kept removed (in destination) because UID 1..5 have identical Date+Message-ID headers. UID 6 is not seen in destination so it's copied.

    If both source and destination have UID 6, but the messages are different, the headers don't match and both the messages are kept in the destination but they're given new UIDs 7 and 8 just to be sure any client didn't get confused about what UID 6 actually was. Thus, one-way sync begins to quickly diverge from the source mailbox once changes start to occur on either side; one-way sync should therefore normally only be used within a short period of time after a doveadm backup or doveadm sync command was used to synchronize the mailboxes.

There are also three different synchronization algorithms:

• Full synchronization (-f parameter) scans through all the messages in all the mailboxes. This guarantees that everything will be synchronized, but it's unnecessarily slow for incremental synchronization.

• Fast synchronization (default) first attempts to find mailboxes that have changed, and synchronize only those. This is done by checking the mailboxes' metadata (NEXTUID and HIGHESTMODSEQ). Usually this works fine, especially with one-way synchronization, but if both sides do exactly the same number of changes, the metadata may end up containing the same values even if the changes were different.

• Stateful synchronization (-s parameter) is the most efficient way to synchronize mailboxes. It relies on having the earlier dsync run's state saved somewhere and being passed to the next dsync run. Based on this state dsync can send only the changes that happened after the previous dsync run. As long as the state or the mailboxes aren't corrupted this algorithm should work perfectly.

The syncing is done as perfectly as possible: an IMAP or a POP3 client shouldn't be able to notice any differences between the two mailboxes. Two-way syncing means that it's safe to do any kind of modifications in both sides, and dsync will merge the changes without losing any changes done on either side. This is possible because dsync can access Dovecot's index logs that keep track of changes. It's of course possible to have conflicts during merging, these are resolved in a safe way. See the dsync design document for more information.

dsync uses the same configuration files as the rest of Dovecot (via doveconf(1)). The entire configuration can be changed by giving -c parameter to another configuration file, or using -o parameter to override specific settings. When executing a remote dsync program it works the same way: it uses its own local configuration.

dsync can be run completely standalone. It doesn't require any Dovecot server processes to be running, except when using -u parameter to do a userdb lookup from auth process.

dsync can sync either one or multiple users using the -u or -A parameters.

GLOBAL OPTIONS

Global doveadm(1) options:

-D

Enables verbosity and debug messages.

-O

Do not read any config file, just use defaults. The dovecot_storage_version defaults to the latest version, but can be overridden with -o.

-k

Preserve entire environment for doveadm, not just import_environment.

-v

Enables verbosity, including progress counter.

-i instance-name

If using multiple Dovecot instances, choose the config file based on this instance name.

See instance_name for more information.

-c config-file

Read configuration from the given config-file. By default it first reads config socket, and then falls back to /etc/dovecot/dovecot.conf. You can also point this to config socket of some instance running compatible version.

-o setting= value

Overrides the configuration setting from /etc/dovecot/dovecot.conf and from the userdb with the given value. In order to override multiple settings, the -o option may be specified multiple times.

OPTIONS

-A

If the -A option is present, the command will be performed for all users. Using this option in combination with system users from userdb { driver = passwd } is not recommended, because it contains also users with a lower UID than the one configured with the first_valid_uid setting.

When the SQL userdb module is used make sure that the iterate_query setting in /etc/dovecot/dovecot-sql.conf.ext matches your database layout. When using the LDAP userdb module, make sure that the iterate_attrs and iterate_filter settings in /etc/dovecot/dovecot-ldap.conf.ext match your LDAP schema. Otherwise doveadm(1) will be unable to iterate over all users.

-F file

Execute the command for all the users in the file. This is similar to the -A option, but instead of getting the list of users from the userdb, they are read from the given file. The file contains one username per line.

-1

Do one-way synchronization instead of two-way synchronization.

-f

Do full synchronization.

-N

Synchronize all the available namespaces. By default only namespaces that don't have explicit location setting are synchronized.

--no-userdb-lookup

Do not perform userdb lookup. Use the USER environment variable to specify the username.

-P

Run a doveadm-purge(1) for the destination (remote) storage after synchronization.

-R

Do a reverse sync. Normally, messages would be pushed from the local system to the destination (remote). This option reverses the flow, and will instead pull messages from the remote to the local storage.

-S socket_path

The option's argument is either an absolute path to a local UNIX domain socket, or a hostname and port (hostname:port), in order to connect a remote host via a TCP socket.

This allows an administrator to execute doveadm(1) mail commands through the given socket.

-T secs

Specify the time in seconds, how long doveadm(1) should wait for stalled I/O operations. The default timeout is 600 seconds.

-d

Use the default destination, which is looked up from the mail_replica userdb extra field.

-g mailbox_guid

Same as -m, but find the mailbox to be synchronized by its GUID instead of by name.

-l secs

Lock the dsync for this user. Wait for maximum secs before giving up. This parameter should be used to avoid broken synchronization if it's possible that dsync is being run concurrently for the same user.

-m mailbox

Synchronize only this mailbox name.

-n namespace

Synchronize only the specified namespace. This parameter can be used multiple times.

-a all mailbox

Name for the "All mails" virtual mailbox. If specified, mails are attempted to be copied from this mailbox instead of being saved separately. This may reduce the total disk space usage as well as disk IO.

-t start date

Skip any mails whose received-timestamp is older than the specified time.

-e end date

Skip any mails whose received-timestamp is newer than the specified time.

-O sync flag

Sync only mails that have the specified flag. If the flag name begins with "-", sync all mails except the ones with the specified flag.

-I max size

Skip any mails larger than the specified size.

-r rawlog_path

Running dsync remotely, write the remote input/output traffic to the specified log file.

-s previous_state

Use stateful synchronization. If the previous state is unknown, use an empty string. The new state is always printed to standard output.

-u user/mask

Run the command only for the given user. It's also possible to use '*' and '?' wildcards (e.g. -u *@example.org).

-x mailbox_mask

Exclude the specified mailbox name/mask. The mask may contain "?" and "*****" wildcards. The mask can also be a special-use name (e.g. \Trash). This parameter can be used multiple times.

ARGUMENTS

destination

This argument specifies the synchronized destination.

It can be one of:

location

Same as mail_location setting, e.g. maildir:~/Maildir

remote: login@host

Uses dsync_remote_cmd setting to connect to the remote host (usually via ssh)

remoteprefix:login@host

This is the same as remote, except "user@domain\n" is sent before dsync protocol starts. This allows implementing a trusted wrapper script that runs doveadm dsync-server by reading the username from the first line.

tcp: host[:port]

Connects to remote doveadm server via TCP. The default port is specified by doveadm_port setting.

tcps: host[:port]

This is the same as tcp, but with SSL.

command [arg1 [, arg2, ...]]

Runs a local command that connects its standard input & output to a dsync server.

EXIT STATUS

dsync will exit with one of the following values:

0

Synchronization was done perfectly.

2

Synchronization was done without errors, but some changes couldn't be done, so the mailboxes aren't perfectly synchronized. Running dsync again usually fixes this. Typically this occurs for message modification sequences with newly created mailboxes. It can also occur if one of the mailboxes change during the syncing.

1, >2

Synchronization failed.

See also doveadm(1) for other possible error codes.

EXAMPLE

SYNCHRONIZATION

Synchronize mailboxes with a remote server. Any errors are written to stderr:

$ doveadm sync -u username@example.com remote:server-replica.example.com

If you need more complex parameters to ssh, you can use e.g.:

$ doveadm sync -u username@example.com ssh -i id_dsa.dovecot \
      mailuser@example.com doveadm dsync-server -u username@example.com

CONVERTING

Assuming that the mail_location setting in /etc/dovecot/conf.d/10-mail.conf is set to: mail_location = mdbox:~/mdbox, a logged in system user may convert her/his mails from its Maildir in her/his home directory to the mdbox mailbox format. The user has to execute the command:

$ doveadm sync maildir:~/Maildir

If you want to do this without any downtime, you can do the conversion one user at a time. Initially:

• Configuration uses mail_location = maildir:~/Maildir

• Set up the possibility of doing per-user mail location using userdb extra fields.

Then for each user:

1. Run doveadm sync once to do the initial conversion.

2. Run doveadm sync again, because the initial conversion could have taken a while and new changes could have occurred during it. This second time only applies changes, so it should be fast.

3. Update mail extra field in userdb to mdbox:~/mdbox. If you're using auth cache, you need to flush it, e.g. doveadm auth cache flush.

4. Wait for a few seconds and then kill (doveadm kick) the user's all existing imap and pop3 sessions (that are still using maildir).

5. Run doveadm sync once more to apply final changes that were possibly done. After this there should be no changes to Maildir, because the user's mail location has been changed and all existing processes using it have been killed.

Once all users have been converted, you can set the default mail_location to mdbox and remove the per-user mail locations from userdb.

REPORTING BUGS

Report bugs, including doveconf -n output, to the Dovecot Mailing List dovecot@dovecot.org. Information about reporting bugs is available at: https://dovecot.org/bugreport.html

SEE ALSO

doveadm(1), doveadm-purge(1)

Additional resources:

• Dsync Design