RabbitMQ Server Configuration

RabbitMQ is used in distributed James in order to have a distributed MailQueue and distributed Event dispatching system. This configuration helps you configure components using RabbitMQ in case you want to setup a distributed James. And it is only applicable with Guice products.

Consult rabbitmq.properties in GIT to get some examples and hints.

RabbitMQ Configuration

uri
the amqp URI pointing to RabbitMQ server. If you use a vhost, specify it as well at the end of the URI. Details about amqp URI format is in RabbitMQ URI Specification
management.uri
the URI pointing to RabbitMQ Management Service. James need to retrieve some information about listing queues from this service in runtime. Details about URI format is in RabbitMQ Management URI
management.user
username used to access management service
management.password
password used to access management service
connection.pool.retries
Configure retries count to retrieve a connection. Exponential backoff is performed between each retries. Optional integer, defaults to 10
connection.pool.min.delay.ms
Configure initial duration (in ms) between two connection retries. Exponential backoff is performed between each retries. Optional integer, defaults to 100
channel.pool.retries
Configure retries count to retrieve a channel. Exponential backoff is performed between each retries. Optional integer, defaults to 3
channel.pool.max.delay.ms
Configure timeout duration (in ms) to obtain a rabbitmq channel. Defaults to 30 seconds. Optional integer, defaults to 30 seconds.
channel.pool.size
Configure the size of the channel pool. Optional integer, defaults to 3
driver.network.recovery.interval
Optional, non-negative integer, default to 100ms. The interval (in ms) that RabbitMQ driver will automatic recovery wait before attempting to reconnect. See https://www.rabbitmq.com/client-libraries/java-api-guide#connection-recovery
ssl.enabled
Is using ssl enabled Optional boolean, defaults to false
ssl.management.enabled
Is using ssl on management api enabled Optional boolean, defaults to false
ssl.validation.strategy
Configure the validation strategy used for rabbitmq connections. Possible values are default, ignore and override. Optional string, defaults to using systemwide ssl configuration
ssl.truststore
Points to the truststore (PKCS12) used for verifying rabbitmq connection. If configured then "ssl.truststore.password" must also be configured, Optional string, defaults to systemwide truststore. "ssl.validation.strategy: override" must be configured if you want to use this
ssl.truststore.password
Configure the truststore password. If configured then "ssl.truststore" must also be configured, Optional string, defaults to empty string. "ssl.validation.strategy: override" must be configured if you want to use this
ssl.hostname.verifier
Configure host name verification. Possible options are default and accept_any_hostname Optional string, defaults to subject alternative name host verifier
ssl.keystore
Points to the keystore(PKCS12) used for client certificate authentication. If configured then "ssl.keystore.password" must also be configured, Optional string, defaults to empty string
ssl.keystore.password
Configure the keystore password. If configured then "ssl.keystore" must also be configured, Optional string, defaults to empty string
quorum.queues.enable
Boolean. Whether to activate Quorum queue usage for all queues. Quorum queues enables high availability. False (default value) results in the usage of classic queues.
quorum.queues.replication.factor
Strictly positive integer. The replication factor to use when creating quorum queues.
quorum.queues.delivery.limit
Strictly positive integer. Value for x-delivery-limit queue parameter, default to none. Setting a delivery limit can prevent RabbitMQ outage if message processing fails. Read https://www.rabbitmq.com/docs/quorum-queues#poison-message-handling
hosts
Optional, default to the host specified as part of the URI. Allow creating cluster aware connections. A coma separated list of hosts, example: hosts=ip1:5672,ip2:5672
notification.queue.ttl
Optional integer, defaults is 3600000. This is used only on queues used to share notification patterns, are exclusive to a node. If omitted, it will not add the TTL configure when declaring queues. Configure queue ttl (in ms). References: https://www.rabbitmq.com/ttl.html#queue-ttl.
mailqueue.publish.confirm.enabled
Whether or not to enable publish confirms for the mail queue. Optional boolean, defaults to true.
event.bus.publish.confirm.enabled
Whether or not to enable publish confirms for the event bus. Optional boolean, defaults to true.
event.bus.notification.durability.enabled
Whether or not the queue backing notifications should be durable. Optional boolean, defaults to true.
event.bus.propagate.dispatch.error
Whether to propagate errors back to the callers when eventbus fails to dispatch group events to RabbitMQ (then store the failed events in the event dead letters). Optional boolean, defaults to true.
vhost
Optional string. This parameter is only a workaround to support invalid URIs containing character like '_'. You still need to specify the vhost in the uri parameter.

Tuning RabbitMQ for quorum queue use

While quorum queues are great at preserving your data and enabling High Availability, they demand more resources and a greater care than regular RabbitMQ queues.

See this section of RabbitMQ documentation regarding RabbitMQ quroum queue performance tunning.

  • Provide decent amount of RAM memory to RabbitMQ. 4GB is a good start.
  • Setting a delivery limit is advised as looping messages can cause extreme memory consumptions onto quorum queues.
  • Set up Raft for small messages:
raft.segment_max_entries = 32768

RabbitMQ MailQueue Configuration

RabbitMQ MailQueue Configuration

James mail queue is a component acting like a queue where it can enqueue and dequeue mails. Beside of the basic features, it also allows some extra operations like getting size, browsing all items in the mail queue... One of the mailqueue implementation is using RabbitMQ. As RabbitMQ doesn't offer enough features to implement efficiently all mailqueue operations, this implementation relies on Cassandra.

cassandra.view.enabled
Whether the Cassandra administrative view should be activated. Boolean value defaulting to true. Not necessarily needed for MDA deployments, mail queue management adds significant complexity.
mailqueue.view.sliceWindow
James divide the view into slices, each slice contains data for a given period, sliceWindow parameter controls this period. This dividing of periods allows faster browsing of the mail queue. Tips for choosing sliceWindow are explained in rabbitmq.properties
mailqueue.view.bucketCount
Mails in a mail queue are distributed across the underlying storage service. BucketCount describes how to be distributing mails to fit with your James setup Tips for choosing bucketCount are explained in rabbitmq.properties
mailqueue.view.updateBrowseStartPace
To browse, James needs a starting point and to continuously update that point in runtime. UpdateBrowseStartPace describes the probability to update the starting point. Tips for choosing updateBrowseStartPace are explained in rabbitmq.properties
mailqueue.size.metricsEnabled
By default, the metrics are disabled for the mail queue size. As computing the size of the mail queue is currently implemented on top of browse operation and thus have a linear complexity, sometimes it can get too big, making it impossible for the ES reporter to handle it correctly without crashing. It can be useful then to disable it. Tips for choosing metricsEnabled are explained in rabbitmq.properties

RabbitMQ Tasks Configuration

Tasks are WebAdmin triggered long running jobs. RabbitMQ is used to organise their execution in a work queue, with an exclusive consumer.

task.consumption.enabled
Whether to enable task consumption on this node. Disable with caution (this only makes sense in a distributed setup where other nodes consume tasks). Defaults to true.
task.queue.consumer.timeout
Task queue consumer timeout. Optional. Duration (support multiple time units cf `DurationParser`), defaults to 1 day. Required at least RabbitMQ version 3.12 to have effect. This is used to avoid the task queue consumer (which could run very long tasks) being disconnected by RabbitMQ after the default acknowledgement timeout 30 minutes. References: https://www.rabbitmq.com/consumers.html#acknowledgement-timeout.

Limitation: Sometimes, some tasks running on James can be very heavy and take a couple of hours to complete. If other tasks are being triggered meanwhile on WebAdmin, they go on the TaskManagerWorkQueue and James unack them, telling RabbitMQ it will consume them later. If they don't get consumed before the consumer timeout setup in RabbitMQ (default being 30 minutes), RabbitMQ closes the channel on an exception. It is thus advised to declare a longer timeout in rabbitmq.conf. More here.