Kubernetes Deployment

Deployment

HTM connectors can be configured to do retries and use the circuit breaker pattern to protect the HTM Http server (for more information on how to configure this, visit Connector).

Here is an example ConfigMap that can be used to deploy HTM to Kubernetes:

 akka {
   remote.artery.canonical.hostname = ${POD_IP}
 # Use Kubernetes API to discover the cluster
   discovery {
     kubernetes-api {
       pod-label-selector = "app=%s"
     }
   }

   management {
 # use the Kubernetes API to create the cluster
     cluster.bootstrap {
       contact-point-discovery {
         discovery-method          = kubernetes-api
         service-name              = ${AKKA_CLUSTER_BOOTSTRAP_SERVICE_NAME}
         required-contact-point-nr = 1
         required-contact-point-nr = ${?REQUIRED_CONTACT_POINT_NR}
       }
     }
 # available from Akka management >= 1.0.0
     health-checks {
       readiness-path  = "health/ready"
       liveness-path   = "health/alive"
     }
   }

   actor {
     provider = cluster
   }

   cluster {
     seed-nodes = []
     downing-provider-class = "akka.cluster.sbr.SplitBrainResolverProvider"
     split-brain-resolver {
       active-strategy = keep-majority
       stable-after = 20s
     }
     sharding {
       remember-entities = off
       handoff-timeout = 8s
       least-shard-allocation-strategy.rebalance-absolute-limit = 20
       rebalance-interval = 2s
       number-of-shards = 100
     }
   }
}

Considerations for CosmosDB on Azure

By default, HTM registers 3 event processors:

  • Task read side event processor

  • Task history read side event processor

  • Bulk execution event processor

These processors register their own event processor streams, which consumes more Request Units (RUs) on Cosmos DB.

To save some RU on Cosmos DB, DelegatingEventProcessor can be used. In this case only one event processor consumes events from journal and then delegate events to other registered event processors.

The disadvantage of DelegatingEventProcessor is if processing the event fails at any of delegated event processors, the same event will be replayed to all delegated event processors (even if event was successfully processed by others). When there are three independent event processors, then the failing event will be replayed only by the single event processor where the previous processing failed.

You can enable the Delegating Event Processor using the following configuration

ipf.htm.event-processor.delegating.enabled = true
Cosmos DB for MongoDB does not support creating unique indexes on non-empty collections. HTM detects ipf.mongodb.database-mode = cosmosdb and automatically creates its compound index without uniqueness to avoid startup errors.