Replication Configuration

Enabling Replication

If replicating over SSH, 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

NOTE: make sure the local user's ssh keys format is PEM, here how to generate them:

  ssh-keygen -m PEM -t rsa -C "your_email@example.com"

Next, create $site_path/etc/replication.config as a Git-style config file, for example 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

Then reload the replication plugin to pick up the new configuration:

  ssh -p 29418 localhost gerrit plugin reload replication

To manually trigger replication at runtime, see SSH command start.

Configuring Cluster Replication

The replication plugin is designed to allow multiple primaries in a cluster to efficiently cooperate together via the replication event persistence subsystem. To enable this cooperation, the directory pointed to by the replication.eventsDirectory config key must reside on a shared filesystem, such as NFS. By default, simply pointing multiple primaries to the same eventsDirectory will enable some cooperation by preventing the same replication push from being duplicated by more than one primary.

To further improve cooperation across the cluster, the replication.distributionInterval config value can be set. With distribution enabled, the replication queues for all the nodes sharing the same eventsDirectory will reflect approximately the same outstanding replication work (i.e. tasks waiting in the queue). Replication pushes which are running will continue to only be visible in the queue of the node on which the push is actually happening. This feature helps administrators get a cluster wide view of outstanding replication tasks, while allowing replication tasks triggered by one primary to be fulfilled by another node which is less busy.

This enhanced replication work distribution allows the amount of replication work a cluster can handle to scale more evenly and linearly with the amount of primaries in the cluster. Adding more nodes to a cluster without distribution enabled will generally not allow the thread count per remote to be reduced without impacting service levels to those remotes. This is because without distribution, all events triggered by a node will only be fulfilled by the node which triggered the event, even if all the other nodes in the cluster are idle. This behavior implies that each node should be configured in a way that allows it alone to provide the level of service which each remote requires. However, with distribution enabled, it becomes possible to reduce the amount of replication threads configured per remote proportionally to the amount of nodes in the cluster, while maintaining the same approximate service level as before adding new nodes.

Threads per remote reduction without service impacts is possible with distribution, because when configuring a node it can be expected that other nodes will pick up some of the work it triggers. Then the node no longer needs to be configured as if it were the only node in the cluster. For example, if a remote requires 6 threads with one node to achieve acceptable service, it should only take 2 threads on 3 equivalently powered nodes to provide the same service level with distribution enabled. Scaling down such thread requirements per remote results in a reduced memory footprint per remote on each node in the cluster. This enables the nodes in the cluster to now scale to handle more remotes with the approximate same service level than without distribution. The amount of extra supported remotes then also scales approximately linearly with the extra nodes in a cluster.

File replication.config

The optional file $site_path/etc/replication.config is a Git-style config file that controls the replication settings for the replication plugin.

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.

In the keys below, the NAME portion is unused by this plugin, but must be unique to distinguish the different sections if more than one remote section appears in the file.

gerrit.replicateOnStartup : If true, replicates to all remotes on startup to ensure they are in-sync with this server. By default, false.

gerrit.autoReload : If true, automatically reloads replication destinations and settings after replication.config file is updated, without the need to restart the replication plugin. When the reload takes place, pending replication events based on old settings are discarded. By default, false.

gerrit.defaultForceUpdate : If true, the default push refspec will be set to use forced update to the remote when no refspec is given. By default, false.

gerrit.maxRefsToLog : Number of refs, that are pushed during replication, to be logged. For printing all refs to the logs, use a value of 0. By default, 0.

gerrit.maxRefsToShow : Number of refs, that are pushed during replication, to be shown in the show-queue output. To show all refs, use a value of 0. By default, 2, because whenever a new patchset is created there are two refs (change ref and meta ref) eg.

`(retry 1) push aaa.com:/git/test.git [refs/heads/b1 refs/heads/b2 (+2)]`

gerrit.pushBatchSize : Max number of refs that are pushed in a single push operation. If more than pushBatchSize are to be pushed then they are divided into batches and pushed sequentially one-by-one.

Can be overridden at remote-level by setting pushBatchSize.

By default, `0`, which means that there are no limitations on number of
refs to be transferred in a single push operation. Note that negative
values are treated as `0`.

Note that `pushBatchSize` is ignored when *Cluster Replication* is configured
- when `replication.distributionInterval` has value > 0.

gerrit.sshCommandTimeout : Timeout for SSH command execution. If 0, there is no timeout and the client waits indefinitely. By default, 0.

gerrit.sshConnectionTimeout : Timeout for SSH connections. If 0, there is no timeout and the client waits indefinitely. By default, 2 minutes.

replication.distributionInterval : Interval in seconds for running the replication distributor. When run, the replication distributor will add all persisted waiting tasks to the queue to ensure that externally loaded tasks are visible to the current process. If zero, turn off the replication distributor. By default, zero.

Turning this on is likely only useful when there are other processes
(such as other masters in the same cluster) writing to the same
persistence store. To ensure that updates are seen well before their
replicationDelay expires when the distributor is used, the recommended
value for this is approximately the smallest `remote.NAME.replicationDelay`
divided by 5.

replication.updateRefErrorMaxRetries : Number of times to retry a replication operation if an update ref error is detected.

If two or more replication operations (to the same GIT and Ref)
are scheduled at approximately the same time (and end up on different
replication threads), there is a large probability that the last
push to complete will fail with a remote "failed to update ref" error.
This error may also occur due to a transient issue like file system
being full which was previously returned as "failed to write" by git.

This option allows Gerrit to retry the replication push when the
"failed to update ref" error is detected. Also retry when the error
"failed to lock" is detected as that is the legacy string used by git.

A good value would be 3 retries or less, depending on how often
you see `updateRefError` collisions in your server logs. A too highly set
value risks keeping around the replication operations in the queue
for a long time, and the number of items in the queue will increase
with time.

Normally Gerrit will succeed with the replication during its first
retry, but in certain edge cases (e.g. a mirror introduces a ref
namespace with the same name as a branch on the master) the retry
will never succeed.

The issue can also be mitigated somewhat by increasing the
`replicationDelay`.

Default: 0 (disabled, i.e. never retry)

replication.lockErrorMaxRetries : Refer to the replication.updateRefErrorMaxRetries section.

If both `lockErrorMaxRetries` and `updateRefErrorMaxRetries` are
configured, then `updateRefErrorMaxRetries` takes precedence.

Default: 0 (disabled, i.e. never retry)

replication.maxRetries : Maximum number of times to retry a push operation that previously failed.

When a push operation reaches its maximum number of retries,
the replication event is discarded from the queue and the remote
destinations may remain out of sync.

Can be overridden at remote-level by setting `replicationMaxRetries`.

By default, pushes are retried indefinitely.

replication.eventsDirectory : Directory where replication events are persisted

When scheduling a replication, the replication event is persisted
under this directory. When the replication is done, the event is deleted.
If plugin is stopped before all scheduled replications are done, the
persisted events will not be deleted. When the plugin is started again,
it will trigger all replications found under this directory.

For replication to work, is is important that atomic renames be possible
from within any subdirectory of the `eventsDirectory` to within any other
subdirectory of the `eventsDirectory`. This generally means that the entire
contents of the `eventsDirectory` should live on the same filesystem.

When not set, defaults to the plugin's data directory.

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 `${name}` is
replaced with the Gerrit project name.  This is a Gerrit
specific extension to the otherwise standard Git URL syntax
and it must be included in each URL so that Gerrit can figure
out where each project needs to be replicated. `${name}` may
only be omitted if the remote refers to a single repository
(i.e.: Exactly one [remote.NAME.projects][3] and that name's
value is a single project match.).

See [git push][1] for details on Git URL syntax.

remote.NAME.adminUrl : Address of the alternative remote server only for repository creation. Multiple URLs may be specified within a single remote block, listing different destinations which share the same settings.

The `adminUrl` can be used as an ssh alternative to the `url`
option, but only related to repository creation.  If not
specified, the repository creation tries to follow the default
way through the url value specified.

It is useful when the `remote.NAME.url` protocols do not allow
repository creation although their usage is mandatory in the
local environment.  In that case, an alternative SSH url could
be specified to repository creation.

To enable replication to different Gerrit instance use
`gerrit+http://` or `gerrit+https://` as protocol name followed
by hostname of another Gerrit server eg.

`gerrit+http://replica2.my.org/`
<br>
`gerrit+https://replica3.my.org/`

In this case replication will use Gerrit's REST API
to create/remove projects and update repository HEAD references.

NOTE: In order to replicate project deletion, the
[delete-project](https://gerrit-review.googlesource.com/admin/projects/plugins/delete-project)
plugin must be installed on the other Gerrit.

*Backward compatibility notice*

Before Gerrit v2.13 it was possible to enable replication to different
Gerrit masters using `gerrit+ssh://`
as protocol name followed by hostname of another Gerrit server eg.

`gerrit+ssh://replica1.my.org/`

In that case replication would have used Gerrit's SSH API to
create/remove projects and update repository HEAD references.

The `gerrit+ssh` option is kept for backward compatibility, however
the use-case behind it is not valid anymore since the introduction of
Lucene indexes and the removal of ReviewDb, which would require
a lot more machinery to setup a master to master replication scenario.

The `gerrit+ssh` option is still possible but is limited to the
ability to replicate only regular Git repositories that do not
contain any code-review or NoteDb information.

Using `gerrit+ssh` for replicating all Gerrit repositories
would result in failures on the `All-Users.git` replication and
would not be able to replicate changes magic refs and indexes
across nodes.

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 : Path of the git-upload-pack executable on the remote system, if using the SSH transport.

Defaults to `git-upload-pack`.

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/.

Note that prefixing a source refspec with `+` causes the replication
to be done with a `git push --force` command.
Be aware that when you are pushing to remote repositories that may
have read/write access (e.g. GitHub) you may want to omit the `+`
to prevent the risk of overwriting branches that have been modified
on the remote.

Multiple push keys can be supplied, to specify multiple
patterns to match against.  In the [example above][2], 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/*` (push all refs) if not specified,
or `+refs/*:refs/*` (force push all refs) if not specified and
`gerrit.defaultForceUpdate` is true.

Note that the `refs/meta/config` branch is only replicated
when `replicatePermissions` is true, even if the push refspec
is 'all refs'.

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 : Time 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.rescheduleDelay : Delay when rescheduling a push operation due to an in-flight push running for the same project.

Cannot be set to a value lower than 3 seconds to avoid a tight loop
of schedule/run which could cause 1K+ retries per second.

A configured value lower than 3 seconds will be rounded to 3 seconds.

By default, 3 seconds.

remote.NAME.replicationRetry : Time 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 unless
the replicationMaxRetries value is set.

This is a Gerrit specific extension to the Git remote block.

By default, 1 minute.

remote.NAME.replicationMaxRetries : Maximum number of times to retry a push operation that previously failed.

When a push operation reaches its maximum number of retries
the replication event is discarded from the queue and the remote
destinations could be out of sync.

This is a Gerrit specific extension to the Git remote block.

By default, use `replication.maxRetries`.

remote.NAME.drainQueueAttempts : Maximum number of attempts to drain the replication event queue before stopping the plugin.

When stopping the plugin, the shutdown will be delayed trying to drain
the event queue.

The maximum delay is `drainQueueAttempts * replicationDelay` seconds.

When not set or set to 0, the queue is not drained and the pending
replication events are cancelled.

By default, do not drain replication events.

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 : 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.

*NOTE:* If an authGroup is provided, and you want a complete
mirror (for backup reasons or to run a Gerrit replica), at
least one of the provided authGroups must have "Access Database"
capability. Otherwise [db](../../../Documentation/note-db.html)
refs will not be replicated.

remote.NAME.createMissingRepositories : If true, a repository is automatically created on the remote site. If the remote site was not available at the moment when a new project was created, it will be created if during the replication of a ref it is found to be missing.

If false, repositories are never created automatically on this
remote.

By default, true, missing repositories are created.

remote.NAME.replicatePermissions : If true, permissions-only projects and the refs/meta/config branch will also be replicated to the remote site. These projects and branches may be needed to keep a backup or slave server current.

By default, true, replicating everything.

remote.NAME.replicateProjectDeletions : If true, project deletions will also be replicated to the remote site.

By default, false, do *not* replicate project deletions.

remote.NAME.replicateHiddenProjects : If true, hidden projects will be replicated to the remote site.

By default, false, do *not* replicate hidden projects.

remote.NAME.mirror : If true, replication will remove remote branches that are absent locally or invisible to the replication (for example read access denied via authGroup option).

By default, false, do not remove remote branches.

remote.NAME.remoteNameStyle : Provides possibilities to influence the name of the target repository, e.g. by replacing slashes in the ${name} placeholder, when the target remote repository is not served by Gerrit.

Github and Gitorious do not permit slashes "/" in repository
names and will change them to dashes "-" at repository creation
time.

If this setting is set to "dash", slashes will be replaced with
dashes in the remote repository name. If set to "underscore",
slashes will be replaced with underscores in the repository name.

Option `basenameOnly` makes `${name}` to be only the basename
(the part after the last slash) of the repository path on the
Gerrit server, e.g. `${name}` of `foo/bar/my-repo.git` would
be `my-repo`.

> **NOTE**: The use of repository name translation using `remoteNameStyle`
> may lead to dangerous situations if there are multiple repositories
> that may be mapped to the same target name. For instance when
> mapping `/foo/my-repo.git` to `my-repo` using "basenameOnly"
> would also map `/bar/my-repo.git` to the same `my-repo` leading
> to conflicts where commits can be lost between the two repositories
> replicating to the same target `my-repo`.

By default, `slash`, i.e. remote names will contain slashes as
they do in Gerrit.

remote.NAME.projects : Specifies which repositories should be replicated to the remote. It can be provided more than once, and supports three formats: regular expressions, wildcard matching, and single project matching. All three formats match case-sensitive.

Values starting with a caret `^` are treated as regular
expressions. `^foo/(bar|baz)` would match the projects
`foo/bar`, and `foo/baz`. Regular expressions have to fully
match the project name. So the above example would not match
`foo/bar2`, while `^foo/(bar|baz).*` would.

Projects may be excluded from replication by using a regular
expression with inverse match. `^(?:(?!PATTERN).)*$` will
exclude any project that matches.

Values that are not regular expressions and end in `*` are
treated as wildcard matches. Wildcards match projects whose
name agrees from the beginning until the trailing `*`. So
`foo/b*` would match the projects `foo/b`, `foo/bar`, and
`foo/baz`, but neither `foobar`, nor `bar/foo/baz`.

Values that are neither regular expressions nor wildcards are
treated as single project matches. So `foo/bar` matches only
the project `foo/bar`, but no other project.

By default, replicates without matching, i.e. replicates
everything to all remotes.

remote.NAME.slowLatencyThreshold : the time duration after which the replication of a project to this destination will be considered “slow”. A slow project replication will cause additional metrics to be exposed for further investigation. See metrics.md for further details.

default: 15 minutes

remote.NAME.pushBatchSize : Max number of refs that are pushed in a single push operation to this destination. If more than pushBatchSize are to be pushed then they are divided into batches and pushed sequentially one-by-one.

By default it falls back to `gerrit.pushBatchSize` value (which is `0` if
not set, which means that there are no limitations on number of refs to
be transferred in a single push operation). Note that negative values are
treated as `0`.

Note that `pushBatchSize` is ignored when *Cluster Replication* is configured
- when `replication.distributionInterval` has value > 0.

remote.NAME.replicateNoteDbMetaRefs : Whether to replicate the NoteDb meta refs (refs/changes/*/meta, refs/changes/*/robot-comments, refs/draft-comments/*, refs/starred-changes/*) or not. This setting is useful when the remote replica does not run a Gerrit instance and one wants to turn off replicating NoteDb meta refs to that remote.

By default, true.

Directory replication

The optional directory $site_path/etc/replication contains Git-style config files that controls the replication settings for the replication plugin. When present all remote sections from replication.config file are ignored.

Files are composed of one remote section. Multiple remote sections or any other section makes the file invalid and skipped by the replication plugin. File name defines remote section name. Each section provides common configuration settings for one or more destination URLs. For more details how to setup remote sections please refer to the replication.config section.

Configuration example:

Static configuration in $site_path/etc/replication.config:

[gerrit]
    autoReload = true
    replicateOnStartup = false
[replication]
    lockErrorMaxRetries = 5
    maxRetries = 5

Remote sections in $site_path/etc/replication directory:

  • File $site_path/etc/replication/host-one.config
[remote]
   url = gerrit2@host-one.example.com:/some/path/${name}.git
  • File $site_path/etc/replication/pubmirror.config
 [remote]
   url = mirror1.us.some.org:/pub/git/${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

Replication plugin resolves config files to the following configuration:

[gerrit]
    autoReload = true
    replicateOnStartup = false
[replication]
    lockErrorMaxRetries = 5
    maxRetries = 5

[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

File secure.config

The optional file $site_path/secure.config is a Git-style config file that provides secure values that should not be world-readable, such as passwords. Passwords for HTTP remotes can be obtained from this file.

remote.NAME.username : Username to use for HTTP authentication on this remote, if not given in the URL.

remote.NAME.password : Password to use for HTTP authentication on this remote.

File ~/.ssh/config

Gerrit reads and caches the ~/.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

IdentityFile and PreferredAuthentications must be defined for all the hosts. Here an example of the minimum ~/.ssh/config needed:

  Host *
    IdentityFile ~/.ssh/id_rsa
    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 passphrases 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.