The @PLUGIN@ plugin must be installed on all the instances. Each instance should be configured with the same gerrit.serverId. If there are existing changes in NoteDb made with another serverId, then this plugin might not be able to access them. Likewise, if the HA gerrit.serverIds differ, then changes conveyed by one instance will not be accessible by the other.
The following fields should be specified in $site_path/etc/@PLUGIN@.config files:
[main] sharedDirectory = /directory/accessible/from/both/instances [autoReindex] enabled = false [peerInfo] strategy = static [peerInfo "static"] url = first_target_instance_url url = second_target_instance_url [http] user = username password = password
[main] sharedDirectory = /directory/accessible/from/both/instances [autoReindex] enabled = false [peerInfo] strategy = jgroups [peerInfo "jgroups"] myUrl = local_instance_url [jgroups] clusterName = foo skipInterface = lo* skipInterface = eth2 protocolStack = protocolStack.xml [http] user = username password = password [healthcheck] enable = true
main.sharedDirectory : Path to a directory accessible from both instances. When given as a relative path, then it is resolved against the $SITE_PATH or Gerrit server. For example, if $SITE_PATH is “/gerrit/root” and sharedDirectory is given as “shared/dir” then the real path of the shared directory is “/gerrit/root/shared/dir”. When not specified, the default is “shared”.
autoReindex.enabled : Enable the tracking of the latest change indexed under data/high-availability for each of the indexes. At startup scans all the changes, accounts and groups and reindex the ones that have been updated by other nodes while the server was down. When not specified, the default is “false”, that means no automatic tracking and indexing at start.
autoReindex.delay : When autoReindex is enabled, indicates the delay aftere the plugin startup, before triggering the conditional reindexing of all changes, accounts and groups. Delay is expressed in Gerrit time values as in websession.cleanupInterval. When not specified, the default is “10 seconds”.
autoReindex.pollInterval : When autoReindex is enabled, indicates the interval between the conditional reindexing of all changes, accounts and groups. Delay is expressed in Gerrit time values as in websession.cleanupInterval. When not specified, polling of conditional reindexing is disabled.
peerInfo.strategy : Strategy to find other peers. Supported strategies are static or jgroups. Defaults to jgroups.
static strategy allows to staticly configure the peer gerrit instance using the configuration parameter peerInfo.static.url.jgroups strategy allows that a gerrit instance discovers the peer instance by using JGroups to send multicast messages. In this case the configuration parameters peerInfo.jgroups.* are used to control the sending of the multicast messages. During startup each instance will advertise its address over a JGroups multicast message. JGroups takes care to inform each cluster when a member joins or leaves the cluster.peerInfo.static.url : Specify the URL for the peer instance. If more than one peer instance is to be configured, add as many url entries as necessary.
peerInfo.jgroups.myUrl : The URL of this instance to be broadcast to other peers. If not specified, the URL is determined from the httpd.listenUrl in the gerrit.config. If httpd.listenUrl is configured with multiple values, is configured to work with a reverse proxy (i.e. uses proxy-http or proxy-https scheme), or is configured to listen on all local addresses (i.e. using hostname *), then the URL must be explicitly specified with myUrl.
jgroups.clusterName : The name of the high-availability cluster. When peers discover themselves dynamically this name is used to determine which instances should work together. Only those Gerrit interfaces which are configured for the same clusterName will communicate with each other. Defaults to “GerritHA”.
jgroups.skipInterface : A name or a wildcard of network interface(s) which should be skipped for JGroups communication. Peer discovery may fail if the host has multiple network interfaces and an inappropriate interface is chosen by JGroups. This option can be repeated many times in the jgroups section. Defaults to the list of: lo*, utun*, awdl* which are known to be inappropriate for JGroups communication.
jgroups.protocolStack : This optional parameter specifies the path of an xml file that contains the definition of JGroups protocol stack. If not specified the default protocol stack will be used. May be an absolute or relative path. If the path is relative it is resolved from the site's etc folder. For more information on protocol stack and its configuration file syntax please refer to JGroups documentation. See JGroups - Advanced topics.
NOTE: To work properly in certain environments, JGroups needs the System property java.net.preferIPv4Stack to be set to true. See JGroups - Trouble shooting.
http.user : Username to connect to the peer instance.
http.password : Password to connect to the peer instance.
@PLUGIN@ plugin uses REST API calls to keep the target instance in-sync. It is possible to customize the parameters of the underlying http client doing these calls by specifying the following fields:
http.connectionTimeout : Maximum interval of time in milliseconds the plugin waits for a connection to the target instance. When not specified, the default value is set to 5000ms.
http.socketTimeout : Maximum interval of time in milliseconds the plugin waits for a response from the target instance once the connection has been established. When not specified, the default value is set to 5000ms.
http.maxTries : Maximum number of times the plugin should attempt when calling a REST API in the target instance. Setting this value to 0 will disable retries. When not specified, the default value is 360. After this number of failed tries, an error is logged.
http.retryInterval : The interval of time in milliseconds between the subsequent auto-retries. When not specified, the default value is set to 10000ms.
cache.synchronize : Whether to synchronize cache evictions. Defaults to true.
cache.threadPoolSize : Maximum number of threads used to send cache evictions to the target instance. Defaults to 4.
cache.pattern : Pattern to match names of custom caches for which evictions should be forwarded (in addition to the core caches that are always forwarded). May be specified more than once to add multiple patterns. Defaults to an empty list, meaning only evictions of the core caches are forwarded.
event.synchronize : Whether to synchronize stream events. Defaults to true.
index.numStripedLocks : Number of striped locks to use during reindexing. Should be of the same order of magnitude of the open changes. I.e.: if one has 5000 changes, one might want to have at least 1000 striped locks. The value has to be tuned empirically by checking the number of failures in acquiring the locking. Checking the consider increasing the number of shards warnings should help. Defaults to 10.
index.synchronize : Whether to synchronize secondary indexes. Defaults to true.
index.synchronizeForced : Whether to synchronize forced index events. E.g. on-line reindex automatically triggered upon version upgrades. Defaults to true.
index.threadPoolSize : Maximum number of threads used to send index events to the target instance. Defaults to 4.
index.batchThreadPoolSize : Maximum number of threads used to send batch index events to the target instance and not associated to an interactive action performed by a user. Defaults equal index.threadPoolSize.
index.maxTries : Maximum number of times the plugin should attempt to reindex changes. Setting this value to 0 will disable retries. After this number of failed tries, an error is logged and the local index should be considered stale and needs to be investigated and manually reindexed. Defaults to 2.
index.retryInterval : The interval of time in milliseconds between the subsequent auto-retries. Defaults to 30000 (30 seconds).
NOTE: the default settings for http.socketTimeout and http.maxTries ensure that the plugin will keep retrying to forward a message for one hour.
websession.synchronize : Whether to synchronize web sessions. Defaults to true.
websession.cleanupInterval : Frequency for deleting expired web sessions. Values should use common time unit suffixes to express their setting:
1 week is treated as 7 days)1 month is treated as 30 days)1 year is treated as 365 days) If a time unit suffix is not specified, hours is assumed. Defaults to 24 hours.healthcheck.enable : Whether to enable the health check endpoint. Defaults to ‘true’.
ref-database.enabled : Enable the use of a global ref-database. Defaults to ‘false’.
ref-database.enforcementRules.<policy> : Level of consistency enforcement across sites on a project:refs basis. Supports two values for enforcing the policy on multiple projects or refs. If the project or ref is omitted, apply the policy to all projects or all refs.
The <policy> can be one of the following values: 1. REQUIRED - Throw an exception if a git ref-update is processed against a local ref not yet in sync with the global ref-database. The user transaction is cancelled. LOCK_FAILURE is reported upstream. 2. IGNORED - Do not validate against the global ref-database. *Example:* ``` [ref-database "enforcementRules"] IGNORED = AProject:/refs/heads/feature ``` Ignore the alignment with the global ref-db for AProject on refs/heads/feature. Defaults to no rule. All projects are REQUIRED to be consistent on all refs.
gerrit.instanceId : Optional identifier for this Gerrit instance. [docs].
Whilst this is not, specifically, a high-availability plugin configuration, it plays an important role on which events are forwarded to peers.
If instanceId is set, events produced by that Gerrit instance will contain its origin in the instanceId field, allowing to track where they are coming from.
The high-availability plugin will check the instanceId value to decide whether an event should be forwarded: events that originated from different Gerrit instances will not be forwarded.
When neither gerrit.instanceId nor event.instanceId are set, it is not possible to identify the origin of the event and thus the event is always forwarded.