4. Remote synchronisation of directories

For using the remsync facility, besides sharutils of course, you also need perl, GNU tar, GNU findutils and gzip, all installed. You also need a sum program which is BSD-compatible, for example the one from GNU textutils.

The remsync program tries to maintain up-to-date copies of whole hierarchy of files over many loosely connected sites, provided there is at least some slow electronic mail between them. It prepares and sends out specially packaged files called synchronization packages, and is able to processes them after reception.

There is no master site, each site has an equal opportunity to modify files, and modified files are propagated. Among many other commands, the broadcast command prepares and sends a synchronization package from the current site to all others, while the process command is used to apply synchronization packages locally after reception from remote sites. remsync will never send a file to another site without being asked to with the broadcast command, and besides the project synchronization state files (always named `.remsync'), it will never modify a file locally without being asked to with the process command.

The unit of transmission is a file, whatever its size may be. Nothing less than whole files are being transmitted. People deciding to cooperate in keeping a synchronized set of files must have trust each other, as each participant has the power of modifying the contents of files at other sites. When remsync is used by a single individual travelling between many sites, as it is often the case, this confidence problem should be easier to resolve :-).

The process command will modify a file without asking confirmation, as long as there is no reason to believe that the file has been modified at more than one place. When some confusion arises from the fact many people independently modified a single file, the receiving user of conflicting files will have the duty of resolving them into a merged version. So, the merging has to be done at the site where the discrepancy is observed, from where it is propagated again to others participants. There is no locking mechanism, so people should use other means, like electronic mail, for telling each other what they do, and which part of a project they are working on.

4.1 Quick start at using remsync

If you are in a real hurry, you can follow the recipe given here, and postpone studying this manual further. However, we will consider only a simple case. In any case, it is good to read the full example, as it gives a good picture of the overall usage of remsync.

For any sizeable project, it might not be convenient to start with one site having it all and the other site having nothing, because this would cause the first synchronization to be huge. It is more practical to move over a copy of the project by other means, might it be diskettes, tapes, or mailshar. So let's presume both sites have a copy of the project, not necessarily identical, but close.

For the following example, we presume that under the same domain `champignac.land', there are two machines named `spirou' and `fantasio'. Further, the participating user on `spirou@spirou.champignac.land' has `spirou' for a login name, and similarily, the participating user on `fantasio.champignac.land' has `fantasio' for a login name. On the `spirou' machine, user `spirou' keeps the project under his home, in directory `spirou-copy', while on the `fantasio' machine, user `fantasio' keeps the project under his home, in directory `fantasio-copy'. Of course, user names might be the same, as well as the directories containing the project. We use different names here just to make the example clearer.

Here is a full transcript of the initialization session, normally executed only once, and slightly edited to make it more suitable for this manual. The example is broken down in little parts, allowing explanations and comments.

% cd ~/spirou-copy
% remsync
remsync (format *.*) - GNU sharutils *.*

>> mode init

init>> remote fantasio@fantasio.champignac.land ~/fantasio-copy
* Directory `~/spirou-copy is not ready for synchronization
Should I prepare it for its first time (y/n)? [y]
Please enter a short project description: Zorglub project
What is your full email address, here? [spirou@spirou.champignac.land]

These commands prepare the `~/spirou-copy' hierarchy for synchronization. You should be located at the top directory of the hierarchy at the time the command remsync is called.

The `mode init' command instructs remsync that no files should be sent in the synchronization package, only their checksum. The goal here is to inform the other site of what we have, and what we don't, somewhat disregarding the fact the other site still looks like it has nothing yet.

The remote command is the key in establishing a synchronization link. It has two parameters, the first being the email address of the partner at the other site (as seen from here, if this matters), the second being the location of the directory where the package should reside on the remote site (as seen from there).

Because there is no `.remsync' file in the project's top-level directory, remsync concludes this is a first synchronization, and so, ask a few questions, often telling in square brackets what answer would be implied by a mere Return or Enter. If the default reply seems inappropriate, just give the correct information.

init>> broadcast

  Broadcasting to address `fantasio@fantasio.champignac.land'
  Studying local files for their signature
  Registering file `file1'
  Registering file `file2'
  Registering file `file3'
* There were new registrations, please check them
Should I resume the current command (y/n)? [y]
Mailing shar to fantasio@fantasio.champignac.land
Message queued
  Command `broadcast' done

init>> quit

%

The broadcast command produces an inventory of the project's files at this end, and mail it to the other partners. But before doing so, because some new files were registered into the synchronization, the user is given the opportunity of interrupting the command, if it is felt that some registered file should really not be there.

The quit command exits remsync, but only once it created the `.remsync' file on disk.

Then, on `fantasio.champignac.land', user `fantasio' will receive the synchronization package, easily recognizable by the fact the string `.remsync.tar.gz' appears in the Subject header of the message. Let's assume `fantasio' saves the whole message as file `/tmp/synchro-message'. Then, `fantasio' might use the following recipe:

% cd /tmp
% unshar synchro-message
uudecoding file .remsync.tar.gz
% remsync process
  Exploding archive `/tmp/.remsync.tar.gz'

  Package being received:
    from address `spirou@spirou.champignac.land'
    for project `Zorglub project'
  Visiting directory `~/fantasio-copy', remote was `~/spirou-copy'
  Initializing file `.remsync' from received information
  Studying local files for their signature
  Command `process' done

In that `remsync process' call, the process command is being given non-interactively, so remsync avoids unneeded interactions and exits right away once the command is done. But equivalently, remsync might be called without arguments, the process command given interactively, and a quit command later required to get out of remsync.

When receiving a synchronization package, remsync should be executed in the directory where the file `.remsync.tar.gz' has been unpacked, which might be quite unrelated to the project itself. Here, `fantasio' executed remsync in `/tmp/', while the project resides in `~/fantasio-project'. The synchronization package itself contains enough information for remsync to automatically visit the proper directory.

After this operation, `fantasio.champignac.land' has a `.remsync' file in `~/fantasio-copy', and the remote synchronization initialization is completed. Either `spirou' or `fantasio' may then modify files on their respective machine. If `spirou' modifies `file2' in the project, `spirou' may execute:

% cd ~/spirou-copy
% remsync broadcast
  Reading configuration for project `Zorglub project'

  Broadcasting to address `fantasio@fantasio.champignac.land'
  Studying local files for their signature
  Packaging file `file2'
shar: Saving file2 (gzipped)
Mailing shar to fantasio@fantasio.champignac.land
Message queued
  Command `broadcast' done

In fact, any time a participant later feel like sending modified files to all partners, s/he just have to change the directory to the top of the project hierarchy, then call `remsync broadcast'. Any time a synchronization package is later received, at either end, the receiving user should apply unshar to related electronic messages for reconstructing the synchronization package `.remsync.tar.gz', then call `remsync process' in the directory containing this package.

4.2 The remsync command and arguments

At the shell prompt, calling the command remsync without any parameters initiates an interactive dialog, in which the user types commands and receives feedback from the program.

The command remsync, given at the shell prompt, may have arguments, in which case these arguments taken together form one remsync interactive command. However, `--help' and `--version' options are interpreted especially, with their usual effect in GNU. Once this command has been executed, no more commands are taken from the user and remsync terminates execution. This allows for using remsync in some kind of batch mode. It is unwise to redirect remsync standard input, because user interactions might often be needed in ways difficult to predict in advance.

The two most common usages of remsync are the commands:

remsync b
remsync p

The first example executes the broadcast command, which sends synchronization packages to all connected remote sites for the current local directory tree.

The second example executes the process command, which studies and complies with a synchronisation package saved in the current directory (not necessarily into the synchronized directory tree), under the usual file name `remsync.tar.gz'.

4.3 Automatic mechanisms in the remsync program

The following points apply to many of the remsync commands. We describe them here once and for all.

• The file `.remsync' describes the various properties for the current synchronization. It is kept right in the top directory of a synchronized directory tree. Some commands may be executed without any need for this file. The program waits as far as possible before reading it.
• If the `.remsync' file is not found when required, and only then, the user is interactively asked to fill a questionnaire about it.
• If the `.remsync' file has been logically modified after having been read, or if it just has been created, the program will save it back on disk. But it will do so only before reading another `.remsync' file, or just before exit. A preexisting `.remsync' will be renamed to `.remsync.bak' before it is rewritten, when this is done, any previous `.remsync.bak' file is discarded.
• Many commands refer to previously entered information by repeating this information. For example, one can refer to a particular scan statement by entering the wildcard to be scanned by this statement. An alternative method of specifying a statement consists in using the decimal number which appears between square brackets in the result of a list command.
• Whenever a site list must be given, it is a space separated list of remote sites. If the list is preceeded by a bang (!), the list is complemented, that is, the sites that will be operated upon are all those not appearing in the list. As a special case, if the site list is completely empty, then all sites are selected.

4.4 Commands for remsync

Program commands to remsync may be given interactively by the user sitten at a terminal. They can come from the arguments of the remsync call at the shell level. Internally, the process command might obey many sub-commands found in a received synchronization package.

Program commands are given one per line. Lines beginning with a sharp (#) and white lines are ignored, they are meant to increase clarity or to introduce user comments. With only a few exceptions, commands are introduced by a keyword and often contains other keywords. In all cases, the keywords specific to remsync may be abbreviated to their first letter. When there are many keywords in succession, the space separating them may be omitted. So the following commands are all equivalent:

list remote
l remote
list r
l r
listremote
lr

while the following are not legal:

l rem
lisremote

Below, for clarity, keywords are written in full and separated by spaces. Commands often accept parameters, which are then separated by spaces. All available commands are given in the table. The first few commands do not pre-require the file `.remsync'. The last three commands are almost never used interactively, but rather automatically triggered while process'ing received synchronization packages.

?

Display a quick help summary of available commands.

! [ shell-command ]

If shell-command has been given, execute it right now as a shell command. When not given, rather start an interactive shell. Exiting from the shell will return to this program. The started shell is taken from the SHELL environment variable if set, else sh is used.

quit

Leave the program normally and return to the shell.

abort

Leave the program with a nonzero exit status and return to the shell. No attempt is made to save a logically modified `.remsync' file.

visit directory

Select another synchronized directory tree for any subsequent operation. directory is the top directory of the synchronized directory tree.

process [ file ]

list [ type ]

List all known statements about some information type. Allowable keywords for type are local, remote, scan, ignore and files. The keyword files asks for all empty statements (see later). If type is omitted, then list all known statements for all types, except those given by files.

[ create ] type value

Create a new statement introducing a value for a given type. Allowable keywords for type are remote, scan and ignore. The create keyword may be omitted.

For create ignore, when the pattern is preceeded by a bang (!), the condition is reversed. That is, only those files which do match the pattern will be kept for synchronization.

delete type value

Delete an existing statement supporting some value for a given type. Allowable keywords for type are remote, scan and ignore.

email remote value

Modify the electronic mail address associated with some remote site, giving it a new value. The special local keyword for remote may be used to modify the local electronic mail address.

home remote value

Modify the top directory of the synchronized directory tree associated with some remote site, giving it a new value. The special local keyword for remote may be used to modify the local top directory.

broadcast site_list

Send by electronic mail an update package to all sites from site_list, containing for each site all and only those files which are known to be different between the remote site and here.

version version

This command is not meant for interactive use. It establishes the remsync version needed to process the incoming commands.

from site_list

This command is not really meant for interactive use. The first site from the site_list is the remote site which originated the synchronization package. All the others are all the sites, including here, which were meant to be synchronized by the broadcast command that was issued at the originating remote site.

sum file checksum

This command is not really meant for interactive use. It declares the checksum value of a particular file at the originating remote site. Also, if at least one sum command is received, then it is guaranteed that the originating remote site sent one sum command for each and every file to be synchronized, so any found local file which was not subject of any sum command does not exist remotely.

if file checksum packaged

This command is not really meant for interactive use. It directs the remsync program to check if a local file has a given checksum. If the checksum agrees, then the local file will be replaced by the packaged file, as found in the received synchronization invoice.

4.5 How remsync works

How does remsync keep track of what is in sync, and what isn't? See section Format of the `.remsync' file, for a the documentation on the `.remsync' file format. I understand that a mere description of the format does not replace an explanation, but in the meantime, you might guess from the format how the program works.

All files are summarized by a checksum, computed by the sum program. There are a few variants of sum computing checksums in incompatible ways, under the control of options. remsync attempts to retrieve on each site a compatible way to do it, and complains if it cannot.

remsync does not compare dates or sizes. Experience shown that the best version of a file is not necessarily the one with the latest timestamp. The best version for a site is the current version on this site, as decided by its maintainer there, and this is this version that will be propagated.

Each site has an idea of the checksum of a file for all other sites. These checksums are not necessarily identical, for sites do not necessarily propagate to all others, and the propagation network maybe incomplete or asymmetrical in various ways.

Propagation is never done unattended. The user on a site has to call remsync broadcast to issue synchronization packages for other sites. If this is never done, the local modifications will never leave the site. The user also has to call remsync process to apply received synchronization packages. Applying a package does not automatically broadcast it further (maybe this could change?).

If a site A propagates some files to sites B and D, but not C, site B is informed that site D also received these files, and site D is informed that site B also received these files, so they will not propagate again the same files to one another. However, both site B and D are susceptible to propagate further the same files to site C.

It may happen that a site refuses to update a file, or modifies a file after having been received, or merges versions, or whatever. So, sites may have a wrong opinion of the file contents on other sites. These differences level down after a few exchanges, and it is very unlikely that a file would not be propagated when it should have.

This scheme works only when the various people handling the various files have confidence in one each other. If site B modifies a file after having received it from site A, the file will eventually be propagated back to site A. If the original file stayed undisturbed on site A, that is, if remsync proves that site B correctly knew the checksum of the original file, then the file will be replaced on site A without any user confirmation. So, the user on site A has to trust the changes made by the user on site B.

If the original file on site A had been modified after having been sent in a synchronization package, than it is the responsibility of the user on site A to correctly merge the local modifications with the modifications observed in the file as received from site B. This responsibility is real, since the merged file will later be propagated to the other sites in an authoritative way.

4.6 Related file formats

4.7 Format of the `.remsync' file

The `.remsync' file saves all the information a site needs for properly synchronizing a directory tree with remote sites. Even if it is meant to be editable using any ASCII editor, it has a very precise format and one should be very careful while modifying it directly, if ever. The `.remsync' file is better handled through the remsync program and commands.

The `.remsync' file is made up of statements, one per line. Each line begins with a statement keyword followed by a single TAB, then by one or more parameters. The keyword may be omitted, in this case, the keyword is said to be empty, and the line begins immediately with the TAB. After the TAB, if there are two parameters or more, they should all be separated with a single space. There should not be any space between the last parameter and the end of line (unless there are explicit empty parameters).

The following table gives the possible keywords. Their order of presentation in the table is also the order of appearance in the `.remsync' file.

remsync

This statement identifies the `.remsync' format. The only parameter states the file format version.

local

This statement should appear exactly once, and has exactly two parameters. The first parameter gives the electronic mail address the other sites should use for sending synchronization packages here. The second parameter gives the name of the local directory tree to synchronize, in absolute notation.

remote

This statement may appear zero, one or more times. Each occurrence connects the synchronized directory tree to another tree on a remote site. The first parameter gives one electronic mail address where to send remote synchronization packages. The second parameter gives the name of the corresponding directory tree for this remote electronic mail address, in absolute notation.

scan

This statement may appear zero, one or more times. When it does not appear at all, the whole local directory tree will always be scanned, searching for files to synchronize. When the statement appears at least once, the whole local directory tree will not be scanned, but only those files or directories appearing in one of these statements. Each scan statement has exactly one parameter, giving one file or directory to be studied. These are usually given relative to top directory of the local synchronization directory tree. Shell wildcards are acceptable.

ignore

This statement may appear zero, one or more times. Each occurrence has one parameter giving a regular expression, according to Perl syntax for regular expressions. These regexps are applied against each file resulting from the scan. If any of the ignore expression matches one of resulting file, the file is discarded and is not subject to remote synchronization.

After all the statements beginning by the previous keywords, the `.remsync' file usually contains many statements having the empty keyword. The empty keyword statement may appear zero, one or more times. Each occurrence list one file being remotely synchronized. The first parameter gives an explicit file name, usually given relative to the top directory of the local synchronized directory tree. Shell wildcards are not acceptable.

Besides the file name parameter, there are supplementary parameters to each empty keyword statement, each corresponding to one remote statement in the `.remsync' file. The second parameter corresponds to the first remote, the third parameter corresponds to the second remote, etc. If there are more remote statements than supplementary parameters, missing parameters are considered to be empty.

Each supplementary parameter usually gives the last known checksum value for this particular file, as computed on its corresponding remote site. The parameter contains a dash - while the remote checksum is unknown. The checksum value for the local copy of the file is never kept anywhere in the `.remsync' file. The special value `666' indicates a checksum from hell, used when the remote file is known to exist, but for which contradictory information has been received from various sources.

4.8 Format of synchronisation packages

Each synchronisation package is transmitted as a file named `.remsync.tar.gz', which has the format of a tar archive, further compressed with the gzip program. This archive always contains a file named `.remsync-work/orders', and zero or more files named `.remsync-work/1', `.remsync-work/2', etc. It contains no other files. Each numbered file is actually a full, non-modified file pertaining to the hierarchy of the project, as sent from the remote site.

The `.remsync-work/orders' file drives the processing of the received synchronization package. This ASCII file format quite closely resembles the `.remsync' format, which we do not explain again here. Only the keywords and their associated parameters are different, and there is no empty keyword. The following table gives the possible keywords, in the order where they normally appear.

format

title

here

remote

ignore

scan

All those keywords are used exactly the same way as within the `.remsync' file, and their format is not explained again here. They state the file format, project title, local and possibly many remote identifications and directories, zero or more ignores, zero or more scans; all of these exactly as known to the remote site who created the synchronization package. In particular, the here line states the originating site of the package rather than the receiving one; the receiving site should still be described by one of the remote lines.

visit

This statement appears exactly once, and has one numeric parameter. It specifies the zero-based index in the list of remote lines above. The index identifies the receiving site, that is, the site to which this package was sent.

copy

This statement appears exactly once, and has one or more numeric parameters. Each specifies a zero-based index in the list of remote lines above. All indices specify the set of all sites who where broadcasted simultaneously, at the time this synchronization package was issued. The index specified by the visit line should also be one of the indices of the copy lines. The order in which the indices are given is important, as it also establishes the order in which file signatures are listed on the check lines below.

check

This statement may appear zero, one or more times. Each occurrence describes one file known to the project at the originating site, and there is exactly one occurrence for each known file in the project. Each check line has exactly n+2 parameters, where n is the number of parameters of the copy command. The first parameter gives a file name, relative to the top directory. The second parameter gives the file signature for this file, as computed at the originating site. For each remote site presented in the copy command, and exactly in the same order, each supplementary parameter gives the originator's idea of the signature for the said file at this remote site. A dash (-) replaces the signature for a file known not to exist.

update

This statement may appear zero, one or more times. Each occurrence describes what to do with one of the `.remsync-work/n' files, distributed within the synchronization package. In fact, there should be exactly as many update lines that there are numbered files in the synchronization package. Usually, each update line immediately follows the corresponding check line, and has exactly three parameters. The first parameter gives a file name in the project, relative to the top level directory of the hierarchy. The second parameter gives a file signature which the said file should have at the receiving site, for it to be replaced safely, with no questions asked (this is the originator's idea of what the file signature was, on the receiving site, prior to its replacement). A dash (-) replaces this signature for a file known not to exist. The third parameter is the number n, which indicates the file `.remsync-work/n' in the synchronization package distribution which should replace the corresponding project file at the receiving site.

4.9 Other means to synchronization

One correspondent thinks that perhaps the news distribution mechanism could be pressed into service for this job. I could have started from C-news, say, instead of from scratch, and have progressively bent C-news to behave like I wanted.

My feeling is that the route was shorter as I did it, from scratch, that it would have been from C-news. Of course, I could have removed the heavy administrative details of C-news: the history and expire, the daemons, the cron entries, etc., then added the interactive features and specialized behaviors, but all this clean up would certainly have took energies. Right now, non counting the subsidiary scripts and shar/unshar sources, the heart of the result is a single (1200 lines) script written in Perl, which I find fairly more smaller and maintainable than a patched C-news distribution would have been.

4.10 Documentation for obsolete scripts

This is merely a place holder for previous documentation, waiting that I clean it up. You have no interest in reading further down.

4.10.1 mailsync

Usage: mailsync [ OPTION ] ... [ EMAIL_ADDRESS ] [ DIRECTORY ]
   or: mailsync [ OPTION ] ... SYNC_DIRECTORY

Option -i simply sends a ihave package, with no bulk files. Option -n inhibits any destructive operation and mailing.

In the first form of the call, find a synchronisation directory in DIRECTORY aimed towards some EMAIL_ADDRESS, then proceed with this synchronisation directory. EMAIL_ADDRESS may be the name of a file containing a distribution list. If EMAIL_ADDRESS is not specified, all the synchronisation directories at the top level in DIRECTORY are processed in turn. If DIRECTORY is not specified, the current directory is used.

In the second form of the call, proceed only with the given synchronisation directory SYNC_DIRECTORY.

For proceeding with a synchronisation directory, whatever the form of the call was, this script reads the ident files it contains to set the local user and directory and the remote user and directory. Then, selected files under the local directory which are modified in regard to the corresponding files in the remote directory are turned into a synchronisation package which is mailed to the remote user.

The list of selected files or directories to synchronize from the local directory are given in the list file in the synchronisation directory. If this list file is missing, all files under the local directory are synchronized.

What I usually do is to cd at the top of the directory tree to be synchronized, then to type mailsync without parameters. This will automatically prepare as many synchronisation packages as there are mirror systems, then email multipart shars to each of them. Note that the synchronisation package is not identical for each mirror system, because they do not usually have the same state of synchronisation.

mailsync will refuse to work if anything needs to be hand cleaned from a previous execution of mailsync or resync. Check for some remaining `_syncbulk' or `_synctemp' directory, or for a `_syncrm' script.

TODO:
- interrogate the user if `ident' file missing.
- automatically construct the local user address.
- create the synchronisation directory on the fly.
- avoid duplicating work as far as possible for multiple sends.
- have a quicker mode, depending on stamps, not on checksums.
- never send core, executables, backups, `.nsf*', `*/_synctemp/*', etc.

4.10.2 resync

Usage: resync [ OPTION ]... TAR_FILE
   or: resync [ OPTION ]... UNTARED_DIRECTORY

Given a tar file produced by mailsync at some remote end and already reconstructed on this end using unshar, or a directory containing the already untared invoice, apply the synchronization package locally.

Option -n inhibits destroying or creating files, but does everything else. It will in particular create a synchronization directory if necessary, produce the `_syncbulk' directory and the `_syncrm' script.

The synchronization directory for the package is automatically retrieved or, if not found, created and initialized. resync keeps telling you what it is doing.

There are a few cases when a resync should not complete without manual intervention. The common case is that several sites update the very same files differently since they were last resync'ed, and then mailsync to each other. The prerequisite checksum will then fail, and the files are then kept into the `_syncbulk' tree, which has a shape similar to the directory tree in which the files where supposed to go. For GNU Emacs users, a very handy package, called emerge, written by Dale Worley <drw@kutta.mit.edu>, helps reconciling two files interactiveley. The `_syncbulk' tree should be explicitely deleted after the hand synchronisation.

Another case of human intervention is when files are deleted at the mailsync'ing site. By choice, all deletions on the receiving side are accumulated in a `_syncrm' script, which is not executed automatically. Explicitely executed, `_syncrm' will remove any file in the receiving tree which does not exist anymore on the sender system. I often edit `_syncrm' before executing it, to remove the unwanted deletions (beware the double negation :-). The script removes itself.

All the temporary files, while resynchronizing, are held in `_synctemp', which is deleted afterwards; if something goes wrong, this directory should also be cleaned out by hand. resync will refuse to work if anything remains to be hand cleaned.

TODO:
- interrogates the user if missing receiving directory in `ident'.
- allow `remote.sum' to be empty or non-existent.

This document was generated by Bruce Korb on June, 3 2006 using texi2html 1.76.