CDNs

[blackfalcon]

The conversation surrounding the development of a custom CDN for Auro versus utilizing a third-party service has been recurrent.

The problem

Auro initially offered a CDN solution to facilitate the adoption of web components in environments deemed challenging for the technology. While encountering few issues with low usage and non-critical interfaces, our reliance on unpkg.com proved problematic once it became unreliable. Subsequently, we transitioned to cdn.jsdelivr.net, which offered greater reliability. However, occasional issues persist, and our inability to provide support, given our lack of ownership or financial investment in this resource, remains a concern.

The question arises: why not establish our own CDN?

It is important to note that creating our own CDN incurs costs, including Azure instance expenses and the time required for development and maintenance. Additionally, an Azure CDN lacks default features present in cdn.jsdelivr.net, such as version support.

Pros

cdn.jsdelivr.net is free and seamlessly integrates with npm, allowing for the instant creation of pointers to code published in a node package.

Cons

We lack control over this resource, and in the event of an outage, access to support is unavailable.

The recommendation

Regarding CDNs in general, we lack a centralized CDN easily accessible to all teams, with no teams willing to support and maintain such a resource. Consequently, the consensus among stakeholders in CDN discussions is that teams should create their own CDNs as needed, based on prior evidence supporting this approach.

For instance, the Search team has developed its own CDN instance for hosting icons, thereby avoiding reliance on JSDelivr as suggested in the Auro documentation. Similarly, the CMS maintains a CDN for content, and individual teams deploy Azure buckets as necessary for their requirements. The Content team has opted against repeating past mistakes by refraining from opening the CMS as a general CDN for teams.

Auro's responsibility in this matter is to update documentation, suggesting any reference to a CDN-supported part of the ecosystem as a mere suggestion, accompanied by a "use at your own risk" warning. Furthermore, this update should offer instructions for integrating resources into a private CDN solution should critical UI support be necessary.

The expected outcomes

1. Teams will possess greater autonomy in managing their CDN needs, leading to expedited issue resolution and reduced dependence on third-party services.
2. Enhanced reliability and availability of critical interfaces through the utilization of private CDN solutions tailored to specific team requirements.
3. Auro's documentation will undergo updates to explicitly recommend the use of private CDN solutions, along with warnings about the risks associated with third-party services.
4. Improved control over CDN resources, empowering teams to efficiently support and maintain their own CDN instances.
5. Decreased downtime and reliance on external support channels, thereby bolstering overall system stability and performance.

[blackfalcon]

Some additional context.

Since we have moved to jsDeliver, there has been no downtime reported. https://www.isitupdown.com/jsdelivr and this is because of the large failover network they deliver https://www.jsdelivr.com/network

Even though this is a free CDN service, this appears to be a solidly supported network with a typical 99.9% uptime.