Cloud Publishing Microservice Architecture and Performance Analysis
This article share the insights into the architecture and performance numbers of the new cloud publishing microservice.
NOTEIssues with existing publishing workflows on cloud
DITA Publishing is a resource-intensive process dependent mainly on available system memory and CPU. The need for these resources increases further if publishers are publishing large maps with many topics or if multiple parallel publishing requests are triggered.
If you are not using the new service, then all the publishing happens on the same Kubernetes(k8) pod which is also running the AEM cloud server. A typical k8 pod has a limit on the amount of memory and CPU that it can use. If AEM Guides users are publishing large or parallel workloads, this limit can breach fast. K8 restarts pods which are trying to use more resources than the configured limit which can have serious impact on the AEM cloud instance itself.
This resource constraint was the main motivation to come up with a dedicated service which can allow us to run multiple concurrent and large publishing workloads on cloud.
Introduction to the new architecture
The service is using Adobe’s cutting edge cloud solutions like App Builder, IO Eventing, IMS to create a serverless offering. These services are itself based on the widely accepted industry standards like Kubernetes and docker.
Each request to the new publishing microservice is executed in an isolated docker container which runs only one publishing request at a time. Multiple new containers are automatically created in case new publishing requests are received. This single container per request configuration allows the microservice to deliver the best performance to the customers without introducing any security risks. These containers are discarded once the publishing is over thus freeing up any unused resources.
All these communications are secured by Adobe IMS using JWT-based authentication and authorization and are executed over HTTPS.
NOTEPerformance Analysis
This section showcases the performance numbers of the microservice. It compares the performance of the microservice with AEM Guides on-prem offering since the old cloud architecture had issues in concurrent publishing or in publishing very large maps.
If you are publishing a large map on on-prem, then you might have to tweak the Java heap parameters or else you can encounter out-of-memory errors. On cloud, the microservice is already profiled and has optimum Java heap and other configurations out of the box.
