gerrit/Documentation/config-replication.txt

Gerrit Code Review - Git Replication
====================================

Gerrit can automatically push any changes it makes to its managed Git
repositories to another system.  Usually this would be configured to
provide mirroring of changes, for warm-standby backups, or a
load-balanced public mirror farm.

The replication runs on a short delay.  This gives Gerrit a small
time window to batch updates going to the same project, such as
when a user uploads multiple changes at once.

Typically replication should be done over SSH, with a passwordless
public/private key pair.  On a trusted network it is also possible to
use replication over the insecure (but much faster) git:// protocol,
by enabling the `receive-pack` service on the receiving system, but
this configuration is not recommended.

Enabling Replication
--------------------

If replicating over SSH (recommended), ensure the host key of the
remote system(s) is already in the Gerrit user's `~/.ssh/known_hosts`
file.  The easiest way to add the host key is to connect once by hand
with the command line:

====
  sudo su -c 'ssh mirror1.us.some.org echo' gerrit2
====

Next, create `'$site_path'/replication.config` as a Git-style config
file, and restart Gerrit.

Example `replication.config` to replicate in parallel to four
different hosts:

====
  [remote "host-one"]
    url = gerrit2@host-one.example.com:/some/path/${name}.git

  [remote "pubmirror"]
    url = mirror1.us.some.org:/pub/git/${name}.git
    url = mirror2.us.some.org:/pub/git/${name}.git
    url = mirror3.us.some.org:/pub/git/${name}.git
    push = +refs/heads/*:refs/heads/*
    push = +refs/tags/*:refs/tags/*
    threads = 3
    authGroup = Public Mirror Group
    authGroup = Second Public Mirror Group
====

To manually trigger replication at runtime, see
link:cmd-replicate.html[gerrit replicate].

[[replication_config]]File `replication.config`
-----------------------------------------------

The optional file `'$site_path'/replication.config` is a Git-style
config file that controls the replication settings for Gerrit.

The file is composed of one or more `remote` sections, each remote
section provides common configuration settings for one or more
destination URLs.

Each remote section uses its own thread pool.  If pushing to
multiple remotes, over differing types of network connections
(e.g. LAN and also public Internet), its a good idea to put them
into different remote sections, so that replication to the slower
connection does not starve out the faster local one.  The example
file above does this.

[[remote]]Section remote
~~~~~~~~~~~~~~~~~~~~~~~~

In the keys below, the <name> portion is unused by Gerrit, but must be
unique to distinguish the different sections if more than one remote
section appears in the file.

[[remote.name.url]]remote.<name>.url::
+
Address of the remote server to push to.  Multiple URLs may
be specified within a single remote block, listing different
destinations which share the same settings.  Assuming sufficient
threads in the thread pool, Gerrit pushes to all URLs in parallel,
using one thread per URL.
+
Within each URL value the magic placeholder `$\{name}` is replaced
with the Gerrit project name.  This is a Gerrit specific extension
to the otherwise standard Git URL syntax.
+
See link:http://www.kernel.org/pub/software/scm/git/docs/git-push.html#URLS[GIT URLS]
for details on Git URL syntax.

[[remote.name.receivepack]]remote.<name>.receivepack::
+
Path of the `git-receive-pack` executable on the remote system, if
using the SSH transport.
+
Defaults to `git-receive-pack`.

[[remote.name.uploadpack]]remote.<name>.uploadpack::
+
Path of the `git-upload-pack` executable on the remote system, if
using the SSH transport.
+
Defaults to `git-upload-pack`.

[[remote.name.push]]remote.<name>.push::
+
Standard Git refspec denoting what should be replicated.  Setting this
to `+refs/heads/\*:refs/heads/\*` would mirror only the active
branches, but not the change refs under `refs/changes/`, or the tags
under `refs/tags/`.
+
Multiple push keys can be supplied, to specify multiple patterns
to match against.  In the example file above, remote "pubmirror"
uses two push keys to match both `refs/heads/\*` and `refs/tags/*`,
but excludes all others, including `refs/changes/*`.
+
Defaults to `+refs/\*:refs/*` (all refs) if not specified.

[[remote.name.timeout]]remote.<name>.timeout::
+
Number of seconds to wait for a network read or write to complete
before giving up and declaring the remote side is not responding.
If 0, there is no timeout, and the push client waits indefinitely.
+
A timeout should be large enough to mostly transfer the objects to
the other side.  1 second may be too small for larger projects,
especially over a WAN link, while 10-30 seconds is a much more
reasonable timeout value.
+
Defaults to 0 seconds, wait indefinitely.

[[remote.name.replicationDelay]]remote.<name>.replicationDelay::
+
Number of seconds to wait before scheduling a remote push operation.
Setting the delay to 0 effectively disables the delay, causing the
push to start as soon as possible.
+
This is a Gerrit specific extension to the Git remote block.
+
By default, 15 seconds.

[[remote.name.replicationRetry]]remote.<name>.replicationRetry::
+
Number of minutes to wait before scheduling a remote push operation
previously failed due to an offline remote server.
+
If a remote push operation fails because a remote server was
offline, all push operations to the same destination URL are
blocked, and the remote push is continuously retried.
+
This is a Gerrit specific extension to the Git remote block.
+
By default, 1 minute.

[[remote.name.threads]]remote.<name>.threads::
+
Number of worker threads to dedicate to pushing to the repositories
described by this remote.  Each thread can push one project at a
time, to one destination URL.  Scheduling within the thread pool
is done on a per-project basis.  If a remote block describes 4
URLs, allocating 4 threads in the pool will permit some level of
parallel pushing.
+
By default, 1 thread.

[[remote.name.authGroup]]remote.<name>.authGroup::
+
Specifies the name of a group that the remote should use to access
the repositories. Multiple authGroups may be specified within a
single remote block to signify a wider access right. In the project
administration web interface the read access can be specified for
this group to control if a project should be replicated or not to the
remote.
+
By default, replicates without group control, i.e replicates
everything to all remotes.


[[ssh_config]]File `~/.ssh/config`
----------------------------------

If present, Gerrit reads and caches `~/.ssh/config` at startup, and
supports most SSH configuration options.  For example:

====
  Host host-one.example.com:
    IdentityFile ~/.ssh/id_hostone
    PreferredAuthentications publickey

  Host mirror*.us.some.org:
    User mirror-updater
    IdentityFile ~/.ssh/id_pubmirror
    PreferredAuthentications publickey
====

Supported options:

 * Host
 * Hostname
 * User
 * Port
 * IdentityFile
 * PreferredAuthentications
 * StrictHostKeyChecking

SSH authentication must be by passwordless public key, as there is
no facility to read passphases on startup or passwords during the
SSH connection setup, and SSH agents are not supported from Java.

Host keys for any destination SSH servers must appear in the user's
`~/.ssh/known_hosts` file, and must be added in advance, before
Gerrit starts.  If a host key is not listed, Gerrit will be unable to
connect to that destination, and replication to that URL will fail.

GERRIT
------
Part of link:index.html[Gerrit Code Review]