Mete Atamel

In this post, I want to talk about an event-driven image processing pipeline that I built recently using Knative Eventing. Along the way, I’ll tell you about event sources, custom events and other components provided by Knative that simply development of event-driven architectures.

Requirements

Let’s first talk about the basic requirements I had for the image processing pipeline:

1. Users upload files to an input bucket and get processed images in an output bucket.
2. Uploaded images are filtered (eg. no adult or violent images) before sending through the pipeline.
3. Pipeline can contain any number of processing services that can be added or removed as needed. For the initial pipeline, I decided to go with 3 services: resizer, watermarker, and labeler. The resizer will resize large images. The watermarker will add a watermark to resized images and the labeler will extract information about images (labels) and save it.

Requirement #3 is especially important. I wanted to be able to add services to the pipeline as I need them or create multiple pipelines with different services chained together.

If you ever used Knative on Google Cloud, you must have heard of Knative-GCP project. As the name suggests, Knative-GCP project provides a number of sources such as CloudPubSubSource, CloudStorageSource, CloudSchedulerSource and more to help reading various Google Cloud sources into your Knative cluster.

I recently updated my Knative Tutorial to use the latest Knative Eventing release v0.14.2 and its corresponding Knative-GCP release v0.14.0. I ran into a weird authentication problem that I want to outline here.

Knative Eventing docs are a little confusing when it comes to different event delivery methods it supports. It talks about event brokers and triggers and it also talks about sources, services, channels, and subscriptions. What to use and when? It’s not clear. Let’s break it down.

Delivery methods

There are 3 distinct methods in Knative:

1. Simple delivery
2. Complex delivery with optional reply
3. Broker and Trigger delivery

Broker and Trigger delivery is what you should care about most of the time. However, the simple and complex delivery have been in Knative for a while and still good to know for what’s happening under the covers.

In part 3 of the blog series, I talked about how we transformed our Windows-only .NET Framework app to a containerized multi-platform .NET Core app.

This removed our dependency on Windows and enabled us to deploy to Linux-based platforms such as App Engine (Flex). On the other hand, the app still ran on VMs, it was billed per second even if nobody used it, deployments were slow and most importantly, it was a single monolith that was deployed and scaled as a single unit.

In part 1, I talked about the initial app and its challenges. In part 2, I talked about the lift & shift to the cloud with some unexpected benefits. In this part 3 of the series, I’ll talk about how we transformed our Windows-only .NET Framework app to a containerized multi-platform .NET Core app and the huge benefits we got along the way.

Why?

The initial Windows VM based cloud setup served us well with minimal issues about roughly 2 years (from early 2017 to early 2019). In early 2019, we wanted to revisit the architecture again. This was mainly driven by the advances in the tech scene namely:

In part 1 of app modernization series, I introduced a simple news aggregator and some of the challenges in its initial architecture. In part 2, I’ll talk about the journey to the cloud and some unexpected benefits and learnings along the way.

Why Cloud?

The initial backend had many issues that I outlined in part 1. After about 1 year, in late 2016, we decided to look into moving it to a more stable home. Our main goal was to improve resiliency of the app, as the IIS host kept crashing, but we didn’t want to rewrite or re-architecture the app in a major way. Around the same time, I started working at Google and was learning all about Google Cloud. We decided to give it a try and see what it took to move the app there.

We all have apps running some “legacy code” in some “legacy way”. The term “legacy” means different things in different projects but we know when we see it and we want to get the time to modernize those apps in some way.

I recently went through the latest phase of modernization of a legacy app. Even though it’s a relatively small app, it thought me a number of lessons that’s worth sharing.

It’s hard to keep with Knative releases with a release every 6 weeks. I finally managed to update my Knative Tutorial for the latest Knative v0.12.0. In this blog post, I want to outline some of the differences I’ve observed.

Knative Serving

Knative Serving has been pretty stable in the recent releases and Knative Serving v0.12.0 is no exception. I didn’t need to update my tutorial specifically for this release.

Knative Eventing

Knative Eventing v0.12.0 changed the default yaml for Knative Eventing bundles. Now, they are under eventing.yaml (previously, it was release.yaml) and this is the yaml you need to point to install eventing. This makes sense as it’s more consistent with Knative Serving and its serving.yaml.

The default Knative Installation instructions for Google Kubernete Engine (GKE) is problematic (see bug 2266). In this post, I want to outline what the problem is, tell you what I do, and also provide you the scripts that work for me until a proper solution is implemented either in gcloud or Knative.

The problem

The default Knative Installation instructions tell you to create a GKE cluster as follows:

gcloud beta container clusters create $CLUSTER_NAME \
 --addons=HorizontalPodAutoscaling,HttpLoadBalancing,Istio \
 --machine-type=n1-standard-4 \
 --cluster-version=latest --zone=$CLUSTER_ZONE \
 --enable-stackdriver-kubernetes --enable-ip-alias \
 --enable-autoscaling --min-nodes=1 --max-nodes=10 \
 --enable-autorepair \
 --scopes cloud-platform

Notice the Istio add-on. This command creates a Kubernetes cluster with Istio already installed. This is good because Istio is a dependency of Knative but keep reading.

In my previous post, I talked about Knative v0.9.0 and some of the eventing changes in the latest release. I’ve been playing with Knative v0.9.0 since then to read Google Cloud Pub/Sub messages using PullSubscription and I ran into a rather fundamental issue that baffled me for a while. I’d like to outline the problem and the solution here, just in case it’s useful to others.

Knative Services as eventing sinks

In my PullSubscription, I could define Kubernetes Services as event sinks as follows: