NAME

muchsync - synchronize maildirs and notmuch databases

SYNOPSIS

muchsync options
muchsync options server-name server-options
muchsync options –init maildir server-name server-options

DESCRIPTION

muchsync synchronizes the contents of maildirs and notmuch tags across machines. Any given execution runs pairwise between two replicas, but the system scales to an arbitrary number of replicas synchronizing in arbitrary pairs. For efficiency, version vectors and logical timestamps are used to limit synchronization to items a peer may not yet know about.

To use muchsync, both muchsync and notmuch should be installed someplace in your PATH on two machines, and you must be able to access the remote machine via ssh.

In its simplest usage, you have a single notmuch database on some server SERVER and wish to start replicating that database on a client, where the client currently does not have any mailboxes. You can initialize a new replica in $HOME/inbox by running the following command:

muchsync --init $HOME/inbox SERVER

This command may take some time, as it transfers the entire contents of your maildir from the server to the client and creates a new notmuch index on the client. Depending on your setup, you may be either bandwidth limited or CPU limited. (Sadly, the notmuch library on which muchsync is built is non-reentrant and forces all indexing to happen on a single core at a rate of about 10,000 messages per minute.)

From then on, to synchronize the client with the server, just run:

muchsync SERVER

Since muchsync replicates the tags in the notmuch database itself, you should consider disabling maildir flag synchronization by executing:

notmuch config set maildir.synchronize_flags=false

The reason is that the synchronize_flags feature only works on a small subset of pre-defined flags and so is not all that useful. Moreover, it marks flags by renaming files, which is not particularly efficient. muchsync was largely motivated by the need for better flag synchronization. If you are satisfied with the synchronize_flags feature, you might consider a tool such as offlineimap as an alternative to muchsync.

Synchronization algorithm

muchsync separately synchronizes two classes of information: the message-to-directory mapping (henceforth link counts) and the message-id-to-tag mapping (henceforth tags). Using logical timestamps, it can detect update conflicts for each type of information. We describe link count and tag synchronization in turn.

Link count synchronization consists of ensuring that any given message (identified by its collision-resistant content hash) appears the same number of times in the same subdirectories on each replica. Generally a message will appear only once in a single subdirectory. However, if the message is moved or deleted on one replica, this will propagate to other replicas.

If two replicas move or copy the same file between synchronization events (or one moves the file and the other deletes it), this constitutes an update conflict. Update conflicts are resolved by storing in each subdirectory a number of copies equal to the maximum of the number of copies in that subdirectory on the two replicas. This is conservative, in the sense that a file will never be deleted after a conflict, though you may get extra copies of files. (muchsync uses hard links, so at least these copies will not use too much disk space.)

For example, if one replica moves a message to subdirectory .box1/cur and another moves the same message to subdirectory .box2/cur, the conflict will be resolved by placing two links to the message on each replica, one in .box1/cur and one in .box2/cur. To respect the structure of maildirs, subdirectories ending new and cur are special-cased; conflicts between sibling new and cur subdirectories are resolved in favor of cur without creating additional copies of messages.

Message tags are synchronized based on notmuch’s message-ID (usually the Message-ID header of a message), rather than message contents. On conflict, tags are combined as follows. Any tag in the notmuch configuration parameter muchsync.and_tags is removed from the message unless it appears on both replicas. Any other tag is added if it appears on any replica. In other words, tags in muchsync.and_tags are logically anded, while all other flags are logically ored. (This approach will give the most predictable results if muchsync.and_tags has the same value in all your replicas. The --init option ensures uniform configurations initially, but subsequent changes to muchsync.and_tags must be manually propagated.)

If your configuration file does not specify a value for muchsync.and_tags, the default is to use the set of tags specified in the new.tags configuration option. This should give intuitive results unless you use a two-pass tagging system such as the afew tool, in which case new.tags is used to flag input to the second pass while you likely want muchsync.and_tags to reflect the output of the second pass.

File deletion

Because publishing software that actually deletes people’s email is a scary prospect, muchsync for the moment never actually deletes mail files. Though this may change in the future, for the moment muchsync moves any deleted messages to the directory .notmuch/muchsync/trash under your mail directory (naming deleted messages by their content hash). If you really want to delete mail to reclaim disk space or for privacy reasons, you will need to run the following on each replica:

cd "$(notmuch config get database.path)"
rm -rf .notmuch/muchsync/trash

OPTIONS

-C file, --config file

Specify the path of the notmuch configuration file to use. If none is specified, the default is to use the contents of the environment variable $NOTMUCH_CONFIG, or if that variable is unset, the value $HOME/.notmuch-config. (These are the same defaults as the notmuch command itself.)

-F

Check for modified files. Without this option, muchsync assumes that files in a maildir are never edited. -F disables certain optimizations so as to make muchsync at least check the timestamp on every file, which will detect modified files at the cost of a longer startup time. If muchsync dies with the error “message received does not match hash,” you likely need to run it with the -F option.

Note that if your software regularly modifies the contents of mail files (e.g., because you are running offlineimap with “synclabels = yes”), then you will need to use -F each time you run muchsync. Specify it as a server option (after the server name) if the editing happens server-side.

-r /path/to/muchsync

Specifies the path to muchsync on the server. Ordinarily, muchsync should be in the default PATH on the server so this option is not required. However, this option is useful if you have to install muchsync in a non-standard place or wish to test development versions of the code.

-s ssh-cmd

Specifies a command line to pass to /bin/sh to execute a command on another machine. The default value is “ssh -CTaxq”. Note that because this string is passed to the shell, special characters including spaces may need to be escaped.

-v

The -v option increases verbosity. The more times it is specified, the more verbose muchsync will become.

--help

Print a brief summary of muchsync’s command-line options.

--init maildir

This option clones an existing mailbox on a remote server into maildir on the local machine. Neither maildir nor your notmuch configuration file (see --config above) should exist when you run this command, as both will be created. The configuration file is copied from the server (adjusted to reflect the local maildir), while maildir is created as a replica of the maildir you have on the server.

--nonew

Ordinarily, muchsync begins by running “notmuch new”. This option says not to run “notmuch new” before starting the muchsync operation. It can be passed as either a client or a server option. For example: The command “muchsync myserver --nonew” will run “notmuch new” locally but not on myserver.

--noup, --noupload

Transfer files from the server to the client, but not vice versa.

--upbg

Transfer files from the server to the client in the foreground. Then fork into the background to upload any new files from the client to the server. This option is useful when checking new mail, if you want to begin reading your mail as soon as it has been downloaded while the upload continues.

--self

Print the 64-bit replica ID of the local maildir replica and exit. Potentially useful in higher-level scripts, such as the emacs notmuch-poll-script variable for identifying on which replica one is running, particularly if network file systems allow a replica to be accessed from multiple machines.

--newid

Muchsync requires every replica to have a unique 64-bit identifier. If you ever copy a notmuch database to another machine, including the muchsync state, bad things will happen if both copies use muchsync, as they will both have the same identifier. Hence, after making such copy and before running muchsync to synchronize mail, run muchsync --newid to change the identifier of one of the copies.

--version

Report on the muchsync version number

EXAMPLES

To initialize a the muchsync database, you can run:

muchsync -vv

This first executes “notmuch new”, then builds the initial muchsync database from the contents of your maildir (the directory specified as database.path in your notmuch configuration file). This command may take several minutes the first time it is run, as it must compute a content hash of every message in the database. Note that you do not need to run this command, as muchsync will initialize the database the first time a client tries to synchronize anyway.

muchsync --init ~/maildir myserver

First run “notmuch new” on myserver, then create a directory ~/maildir containing a replica of your mailbox on myserver. Note that neither your configuration file (by default ~/.notmuch-config) nor ~/maildir should exist before running this command, as both will be created.

To create a notmuch-poll script that fetches mail from a remote server myserver, but on that server just runs notmuch new, do the following: First, run muchsync --self on the server to get the replica ID. Then take the ID returned (e.g., 1968464194667562615) and embed it in a shell script as follows:

#!/bin/sh
self=$($HOME/muchsync --self) || exit 1
if [ "$self" = 1968464194667562615 ]; then
    exec notmuch new
else
    exec $HOME/muchsync -r ./muchsync --upbg myserver
fi

The path of such a script is a good candidate for the emacs notmuch-poll-script variable.

Alternatively, to have the command notmuch new on a client automatically fetch new mail from server myserver, you can place the following in the file .notmuch/hooks/post-new under your mail directory:

#!/bin/sh
muchsync --nonew --upbg myserver

FILES

The default notmuch configuration file is $HOME/.notmuch-config.

muchsync keeps all of its state in a subdirectory of your top maildir called .notmuch/muchsync.

SEE ALSO

notmuch(1).

BUGS

muchsync expects initially to create replicas from scratch. If you have created a replica using another tool such as offlineimap and you try to use muchsync to synchronize them, muchsync will assume every file has an update conflict. This is okay if the two replicas are identical; if they are not, it will result in artifacts such as files deleted in only one replica reappearing. Ideally notmuch needs an option like --clobber that makes a local replica identical to the remote one without touching the remote one, so that an old version of a mail directory can be used as a disposable cache to bootstrap initialization.

muchsync never deletes directories. If you want to remove a subdirectory completely, you must manually execute rmdir on all replicas. Even if you manually delete a subdirectory, it will live on in the notmuch database.

To synchronize deletions and re-creations properly, muchsync never deletes content hashes and their message IDs from its database, even after the last copy of a message has disappeared. Such stale hashes should not consume an inordinate amount of disk space, but could conceivably pose a privacy risk if users believe deleting a message removes all traces of it.

Message tags are synchronized based on notmuch’s message-ID (usually the Message-ID header of a message), rather than based on message contents. This is slightly strange because very different messages can have the same Message-ID header, meaning the user will likely only read one of many messages bearing the same Message-ID header. It is conceivable that an attacker could suppress a message from a mailing list by sending another message with the same Message-ID. This bug is in the design of notmuch, and hence not something that muchsync can work around. muchsync itself does not assume Message-ID equivalence, relying instead on content hashes to synchronize link counts. Hence, any tools used to work around the problem should work on all replicas.

Because notmuch and Xapian do not keep any kind of modification time on database entries, every invocation of muchsync requires a complete scan of all tags in the Xapian database to detect any changed tags. Fortunately muchsync heavily optimizes the scan so that it should take well under a second for 100,000 mail messages. However, this means that interfaces such as those used by notmuch-dump are not efficient enough (see the next paragraph).

muchsync makes certain assumptions about the structure of notmuch’s private types notmuch_message_t and notmuch_directory_t. In particular, it assumes that the Xapian document ID is the second field of these data structures. Sadly, there is no efficient and clean way to extract this information from the notmuch library interface. muchsync also makes other assumptions about how tokens are named in the Xapian database. These assumptions are necessary because the notmuch library interface and the notmuch dump utility are too slow to support synchronization every time you check mail.