Posts Tagged ‘Disaster’

There can be many UI / API / DB level design patterns to follow but when / what to choose need to be carefully decided and following 5 important principals helped to decide on some of the recent decisions … (Thanks to our Chief Architect for providing these guidelines …)

  1. Keep original data
  2. Maintain SVOT – strengths, vulnerabilities, opportunities, threats
  3. Follow KISS – Keep It Simple Silly – easy maintenance / low cost / more reliability
  4. Don’t rely on PUSH
  5. Plan for disaster

Here are the examples of how above principles are helped us to decide between PUSH Vs PULL  &  Filtering out the data or Keeping the original msg …

Our team had created multiple microservices recently for achieving a big org level new initiative. For one of the end to end flow i.e. creating a new app instance for a trail user involves 3 microservices which follows #1, #4 from above principals.

  1. External Service – follows the event lifecycle model where it triggers the subscribed REST endpoint with user request msg when a new user request comes. Here event lifecycle can be synchronous/asynchronous and the service tries multiple times in case when REST endpoint is down. The service also expects REST endpoint to complete the event lifecycle.
  2. REST interface service – gets the msg from external service and pushes it into ZK.
    1. Follows #1 so that the “Original message” been “PUSHed” to  ZK so that it avoids unnecessary parsing/transformations on the msg, operation will be quick & future changes on the msg can be directly synched instead of parsing/editing the zk msg.
    2. Here #4 is not followed since the service been triggered by external service which intern pushes the original message to ZK and it wait for the core service to complete the provisioning and core service updates the status to ZK. Once it gets the provisioning status by following #4 i.e. PULL model  it triggers the external service to complete the event life cycle.
  3. Core service which I owned creates application instances in Docker Swarm for the trail customers based on data provided in Zookeeper (ZK)
    1. Zookeeper out of box provides watch capability where we can watch on a node which triggers an event when changes happen and based on that the core service can take care the provisioning, so it is PUSH model. But here based on #4, we followed the PULL model i.e. pulling ZK data in period interval to take care provisioning requests and it really helped us from single point failure with ZK (i.e. ZK connection failure or missing some of the events due to any exceptions).

 

Design Pattern References:

Advertisements