Lalitha Kolla

Authorization is the process of checking if a user can perform an action on a given resource. At Carium, we had big healthcare providers as our clients. Based on the provider's grouping of patients/coaches/admins and patient sharing preferences, access permissions can dynamically change. Authorization checks should be low latency as it is done for every back-end call. In a microservice architecture, the ability to add authorization with ease is important for developer productivity.

I designed and implemented a low latency, scalable solution, meeting all of the above requirements. Data is modeled so that a single SQL query can determine user permissions on a given resource. Redis cache was used to drive down latency further. Complex scenarios such as parallel permission modifications by providers and patients have been handled.

At Carium, the only way to access data from one microservice to another was by using REST API calls, which were getting expensive for frequently used data points like users, patient, and provider settings. I designed and implemented a framework used by microservices to publish/subscribe interesting data points. The publish/subscribe mechanism was based on Amazon SQS/SNS. Special care had been taken to ensure developers could integrate it with minimal configuration. This framework reduced traffic between microservices, reducing the latency for end-user API calls.

Blue Planet is an on-premises networking platform, based on Docker-based microservices. The requirement was to provide a highly available MySQL service.

After doing research and experimentation with different opensource options and HA designs, Galera with master/master replication proved to be the best fit. It was important to bring the cluster up and down on-demand, which was not easy with Galera. I integrated Galera with the Blue Planet platform, solved on-demand requirements by implementing state machines that control the start-up sequence of Galera pods. Later on, the design was extended to support geo-redundancy based on Galera master/slave replication.

Telecommunication providers always expand their networks by adding new nodes and replacing/upgrading existing nodes for higher capacity. The process of adding new nodes is referred to as Cut-In, and deletion is referred to as Cut-Out. It was a major business for Cyan, but was very manual to plan and figure out all the changes needed to be made, ranging from connecting physical fibers to changing software configurations. Each change generates many configuration combinations, which is further complicated by additional changes. Sales/support engineers used to spend weeks figuring out the right combination of changes for optimal network configuration.

I was the lead engineer and implemented easy-to-use components for Node Cut-In/Cut-Out. Changes and possible configurations were aggregated by layer and gave users the tools and options to fix the issues at one layer before proceeding further. Days of work have been reduced to hours with minimal traffic disruptions.