Zesty.iohas completed the transition of its Webengine service to the Google Cloud Run platform 

Webengine is Zesty.io's service for rendering web request responses. These responses take the form of HTML, JSON, CSS, JavaScript, and many more file formats. This service has been operating on the Google AppEngine(GAE) platform for many years. While GAE provided a major improvement over our previous Kubernetes cluster. By providing on-demand resource scaling. It had one major drawback for our compute loads. 

Cold Starts

Whenever the system autoscaled it had to start new instances and put them into rotation. With GAE this startup latency took roughly 2 - 5 minutes. During which time existing resources could become constrained. Resulting in not enough downstream resources and our load balancer returning 503 Service Unavailable or 504 Gateway Timeout responses. This scenario was particularly problematic during malicious attacks.

Burst Volume Attacks

This attack takes the form of a large volume of traffic sent for short timeframes. Typically repeated at some interval.  For example, making requests at a rate of 2,000 per second for 5 minutes. Then repeating that an hour later.

It is a targeted type of attack. Meaning it is done against a particular domain with the intent of taking it offline.

In this type of attack the bad actor is operating with the understanding that a sophisticated platform will have auto scaling capabilities. Meaning that the platform is designed to automatically provision additional resources as needed. Any solution which has auto scaling also needs to ensure those same resources are taken offline once they are no longer needed. Otherwise it would create serious cost control issues.

Thus the bad actor bursts the attack for short periods of time and repeats. With the intent to introduce instability then back off once the platform has autoscaled but then strike again once the platform releases previously provisioned resources. Causing continuous points of interruption over a length of time.

Google Cloud Run

With the transition to Cloud Run complete we have seen our container startup latency drop to an average of 1 second. This has resulted in significant stability improvement. Making 503 and 504 responses almost non-existent. While we are very excited about these changes and the positive impact they will have on our customers' traffic, we are not done. We still have many improvements we are looking forward to in 2026.