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
In this case URLs of peers are not needed. All peers just need to join the same JGroups cluster defined by the jgroups.clusterName.
[main] transport = jgroups sharedDirectory = /directory/accessible/from/both/instances [autoReindex] enabled = false [jgroups] clusterName = foo skipInterface = lo* skipInterface = eth2 protocolStack = protocolStack.xml timeout = 5000 maxTries = 100
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”.
main.transport : Message transport layer. Could be: http or jgroups. When not specificed the default is http. When set to jgroups then all peerInfo.* sections are unnecessary and ignored.
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. Alternatively, this URL can also be specified using the environment variable GERRIT_URL. This is useful in environments like Kubernetes, where manual configuration of each Gerrit instance is not possible. If neither the configuration option nor the system property is 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.
jgroups.kubernetes : If true, a protocol stack optimized for Kubernetes will be used. Peers will be discovered by querying the Kubernetes API server for pods. The functionality is provided by the jgroups-kubernetes extension. To enable Gerrit to use the Kubernetes API, the pods require a ServiceAccount with permissions to list pods (example). Further, Gerrit requires a valid TLS certificate in its keystore, since the Kubernetes API server requires TLS. (Default: false)
jgroups.kubernetes.namespace : The namespace in which to query for pods. (Default: default)
jgroups.kubernetes.label : A label that will be used to select the pods in the format label=value. Can be set multiple times.
jgroups.timeout : Maximum interval of time in milliseconds the JGroups wait for a response Value is expressed in Gerrit time values as in websession.cleanupInterval. forwarding a message. When not specified, the default value is 5 seconds.
jgroups.maxTries : Maximum number of times JGroups should attempt to forward a messages. Must be at least 1. Setting this option to zero or negative will assume the default value. When not specified, the default value is 720 times.
jgroups.retryInterval : The interval of time in milliseconds between the subsequent auto-retries. Value is expressed in Gerrit time values as in websession.cleanupInterval. When not specified, the default value is set to 10 seconds.
jgroups.threadPoolSize : Maximum number of threads used to execute JGroups calls towards target instances.
NOTE: the default settings for jgroups.timeout and jgroups.maxTries ensure that JGroups will keep retrying to forward a message for one hour.
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. Value is expressed in Gerrit time values as in websession.cleanupInterval. When not specified, the default value is set to 5 seconds.
http.socketTimeout : Maximum interval of time in milliseconds the plugin waits for a response from the target instance once the connection has been established. Value is expressed in Gerrit time values as in websession.cleanupInterval. When not specified, the default value is set to 5 seconds.
http.maxTries : Maximum number of times the plugin should attempt when calling a REST API in the target instance. Must be at least 1. Setting this option to zero or negative will assume the default value; When not specified, the default value is 360. After this number of failed attempts, an error is logged.
http.retryInterval : The interval of time in milliseconds between the subsequent auto-retries. Value is expressed in Gerrit time values as in websession.cleanupInterval. When not specified, the default value is set to 10 seconds.
http.threadPoolSize : Maximum number of threads used to execute REST calls towards target instances.
cache.synchronize : Whether to synchronize cache evictions. Defaults to true.
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.allowedListeners : Class name or package name of the event listener that is always allowed to receive all events generated locally or from a remote end. Can be specified multiple times for allowing multiple listeners classes or packages. Defaults to an empty list.
event.synchronize : Whether to synchronize stream events. Defaults to true.
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 process index events in the receiving gerrit instance. Defaults to 4.
index.batchThreadPoolSize : Maximum number of threads used to process batch index events in the receiving gerrit instance and not associated to an interactive action performed by a user. Defaults to index.threadPoolSize.
index.maxTries : Maximum number of times the plugin should attempt to reindex changes. Must be at least 1. Setting this option to zero or negative will assume the default value; 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. Value is expressed in Gerrit time values as in websession.cleanupInterval. Defaults to 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.