gitolite/doc/mirroring.mkd
Sitaram Chamarty bbaacfaee7 (mostly) doc changes
- minor typo fixes, clarifications, etc.

  - keep sts.html url consistent, because many people link to
    http://sitaramc.github.com/gitolite/sts.html

  - create a common migration doc, so the old 'migr.html' does not 404
    when g3 docs become "main"

  - progit doc done

  - add gitosis convert script (FWIW)

  - a minor comment fix to Sugar.pm
2012-04-10 15:41:32 +05:30

7.7 KiB

mirroring using gitolite

WARNING existing gitolite mirroring users please note: there are [significant changes][g2i-mirroring] in syntax and usage compared to g2. If you're not the kind who reads documentation before doing serious system admin things, well... good luck!


TOC


Mirroring is simple: you have one "master" server and one or more "slave" servers. The slaves get updates only from the master; to the rest of the world they are at best read-only.

Gitolite extends this simple notion in the following ways:

  • different masters and sets of slaves for different repos

    This lets you do things like

    • use the server closest to most of its developers as the master for that repo
    • not mirror a repo at all to some servers
    • have repos that are purely local to a server (not mirrored at all)
    • negotiate mirroring with servers that are not even under your control
    • push to a slave on demand or via cron (helps deal with bandwidth or connectivity constraints).
  • pushes to a slave can be transparently forwarded to the real master

    Your developers need not worry about where a repo's master is -- they just write to their local mirror for all repos, even if their local mirror is only a slave for some.

caveats

  • mirroring will never create a repo on a slave; it has to exist and be prepared to receive updates from the master. (For example, [wild][] repos must be created on the slave as well, otherwise they will not propagate).

  • mirroring is only for git repos. Ancillary files like gl-creator and gl-perms in the repo directory are not mirrored; you must do that separately. Files in the admin directory (like log files) are also not mirrored.

  • if you ever do a [bypass push][bypass], mirroring will not work. Mirroring checks also will not work -- for example, you can push to a slave, which is not usually a good idea. So don't bypass gitolite if the repo is mirrored!

setting up mirroring

This is in two parts: the initial setup and the rc file, followed by the conf file settings and syntax.

the initial setup and the rc file

On each server:

  • install gitolite normally. Make clones of the server's 'gitolite-admin' repo on your workstation so you can admin them all from one place.

  • give the server a short, simple, "hostname" and set the HOSTNAME in the rc file to this name, for example 'mars'.

  • run ssh-keygen if needed and get an ssh key pair for the server. Copy the public key to a common area and name it after the host, but with 'server-' prefixed. So the pubkey for server 'mars' would be stored as 'server-mars.pub'.

  • copy all keys to all the admin repo clones on your workstation and and add them as usual. This is an O(N^2) operation ;-)

    You may have guessed that the prefix 'server-' is special, and distinguishes a human user from a mirroring peer.

  • create "host" aliases to refer to all other machines. See [here][ssh-ha] for what/how.

    The host alias for a host (in all other machines' ~/.ssh/config files) MUST be the same as the HOSTNAME in the referred host's ~/.gitolite.rc. Gitolite mirroring requires this consistency in naming; things will NOT work otherwise.

    Normally you should be able to build one common file and append it to all the servers' ~/.ssh/config files.

  • the following MUST work for each pair of servers that must talk to each other:

    # on server mars
    ssh phobos info
    # the response MUST start with "hello, server-mars..."
    

    Note the exact syntax used; variations like "ssh git@phobos.example.com info" are NOT sufficient. That is why you need the ssh host aliases.

    Check this command from everywhere to everywhere else, and make sure you get expected results. Do NOT proceed otherwise.

  • setup the gitolite.conf file on all the servers. If the slaves are to be exact copies of the master, you need to do the complete configuration only on the master; the slaves can have just this:

    repo gitolite-admin
        RW+     =   <some local admin>
    
        option mirror.master    =   mars
        option mirror.slaves    =   phobos
    

    because on the first push to the master it will update all the slaves anyway.

  • when that is all done and tested, enable mirroring by going through the rc file and uncommenting all the lines mentioning Mirroring.

conf file settings and syntax

Mirroring is defined by the following [options][]. You can have different settings for different repos, and of course some repos may not have any mirror options at all -- they are then purely local.

repo foo
    ...access rules...

    option mirror.master        =   mars
    option mirror.slaves        =   phobos deimos
    option mirror.redirectOK    =   all

The first line is easy, since a repo can have only one master.

The second is a space separated list of hosts that are all slaves. You can have several slave lists, as long as the config key starts with 'mirror.slaves' and is unique. For example.

    option mirror.slaves-1   =   phobos deimos
    option mirror.slaves-2   =   io europa
    option mirror.slaves-3   =   ganymede callisto

Do not repeat a key; then only the last line for that key will be effective.

redirected pushes

Please read carefully; there are security implications if you enable this for mirrors NOT under your control.

Normally, a master, (and only a master), pushes to a slave, and the slaves are "read-only" to the users. Gitolite allows a slave to receive pushes from a user and transparently redirect them to the master.

This simplifies things for users in complex setups, letting them use their local mirror for both fetch and push access to all repos.

Just remember that if you do this, authentication happens on the slave, but authorisation is on the master. The master is trusting the slave to authenticate the user correctly, and use the same authentication data (i.e., user alice on the slave should be guaranteed to be the same as user alice on the master).

The syntax for enabling this is one of these:

option mirror.redirectOK    =   all
option mirror.redirectOK    =   phobos deimos

The first syntax trusts all valid slaves to redirect user pushes, while the second one trusts only some slaves.

Note that you cannot redirect gitolite commands (like perms, etc).

#sync manually synchronising a slave repo

You can use the gitolite mirror push command on a master to manually synchronise any of its slaves. Try it with -h to get usage info.

Tip: if you want to do this to all the slaves, try this:

for s in `gitolite git-config -r reponame mirror.slave | cut -f3`
do
    gitolite mirror push $s reponame
done

This command can also be run remotely; run ssh git@host mirror -h for details.

#HOSTNAME appendix A: HOSTNAME substitution

Wherever gitolite sees the word %HOSTNAME, it will replace it with the HOSTNAME supplied in the rc file, if one was supplied. This lets you maintain configurations for all servers in one repo, yet havethem act differently on different servers, by saying something like:

subconf "%HOSTNAME/*.conf"

You can use it in other places also, for example:

RW+     VREF/NAME/subs/%HOSTNAME/       =   @%HOSTNAME-admins

(you still have to define @mars-admins, @phobos-admins, etc., but the actual VREF is now one line instead of one for each server!)

appendix B: efficiency versus paranoia

If you're paranoid enough to use mirrors, you should be paranoid enough to set this on each server, despite the possible CPU overhead:

git config --global receive.fsckObjects true