Difference between revisions of "Monasca/Events"

Revision as of 10:27, 3 February 2015

Introduction

Real-time event stream processing in Monasca is work in progress that will allow events from external data sources to be sent to the Monasca API where they can be transformed, stored, queried, filtered, grouped and associated with notification methods. Streams can be defined by filtering and grouping events using fields in the event. Notification methods can be associated with streams that are invoked when fire and expire conditions occur, similar to how actions are associated with alarms.

An example use case is to send OpenStack "compute.instance.create.*" events (see https://wiki.openstack.org/wiki/NotificationEventExamples) to the API. A transform on the events could be defined that reduces the number of supplied fields in the source event to a more manageable number of fields as well as normalizing the data. An event stream can be creating by defining a filter to select all "compute.instance.create.*" events and group them by a set of fields in the event, such as "instance_id". When the "compute.instance.create.end" event occurs a fire criteria can be invoked that sends the event stream to the notification methods that have been registered. If the notification method is a web hook additional processing on the collected event stream can occur, such as calculating the duration between the "compute.instance.create.start" and "compute.instance.create.end" event from which new events and metrics can be created.

Events API

Events

• POST /v2.0/events: Publish an event.
• GET /v2.0/events/{event_id}: Get an event with the specific event ID.
• GET /v2.0/events: List events.

Transforms

• POST /v2.0/transforms - POST a transform
• GET /v2.0/transforms - List transforms
• GET /v2.0/transforms/{transform_id} - Get the specified transform
• DELETE /v2.0/transforms/{transform_id}

Stream Definition

• POST /v2.0/stream-definitions: Creates a stream definition with the following parameters in the JSON body
  • name (string(255), required) - A unique name for the stream.
  • description (string(255), optional) - A description of the stream.
  • select () - Fields of events to filter/match on. For example select all events where the field "event_type" like "compute.instance.create.*".
  • group_by () - Fields of events to group on. For all events that match the match_by criteria group them by the specified criteria. For example, group events by the field "instance_id" or "user_id".
  • expires (int, required) - Elapsed time in milliseconds from the start of a stream to when the expire actions are invoked if the fire actions haven't occurred yet.
  • fire_actions ([string(50)], optional) - Array of notification method IDs that are invoked when the pipeline fires.
  • expire_actions ([string(50)], optional) - Array of notification method IDs that are invoked when the pipeline expires.
• GET /v2.0/stream-definitions
• GET /v2.0/stream-definitions/{stream-definition-id}
• DELETE /v2.0/stream-definitions/{stream-definition-id}

Transformation Engine

Consumes events from Kafka, transforms them, and publishes to Kafka.

Event Engine

Consumes transformed events from Kafka, and uses the Winchester pipeline to process them.

Distiller

No changes required.

Winchester

• Add Support for multi-tenancy
• Dynamically update pipelines.
• Add and delete pipeline definitions at run-time. Currently, the Winchester pipelines needs to be created at start-up time.
  • Supply pipeline definitions in methods, not yaml files. Winchester currently reads the pipeline configuration information from yaml files at start-up time.
• Create pipeline handler that publishes notification events such that the Notification Engine can consume them.

Notification Engine

Needs to be able to consume general events from the Threshold Engine or Winchester Pipeline Handler.

Threshold Engine

Update to generate more general alarm state transition events.

MySQL

• Initialize Winchester schemas
• Initialize Monasca transforms and pipeline schemas.

Demo

Look into creating a demo of a pipeline handler using Iron.IO.