A BOSH deployment update involves the recreation of VMs while preserving the data on the persistence disk. But what about the in-memory data which was very much part of the deployment and was being used by the application. For example, the shared memory buffers used by databases such as PostgreSQL or transient messages used by messaging engines like RabbitMQ or the TCP connections for an application all of which are stored in memory. For a PaaS provider it becomes important to consider the in-memory data for an update use case so as to abide by SLAs and to avoid any performance impact on the customer used service instances.
In this talk, Ashish and Shashank will explain the problems we have observed during updates for services such as PostgreSQL, RabbitMQ and how we could mitigate them by preserving the in-memory state. The proposed solution can be easily plugged into BOSH drain scripts and provides a seamless experience in a live migration whereby preserving the end-user experience pre and post update. This talk will be followed by a short demo on a sample use case.
Saurav Mondal is working as developer and part of Service Fabrik in SAP, an incubation project of Cloud Foundry. His Interest mainly around cloud computing, distributed computing and machine learning etc.
