fnproject 生产环境使用

Posted 2020-10-09 rongfengliang-荣锋亮

此为官方的参考说明

 

Running Fn in Production

The QuickStart guide is intended to quickly get started and kick the tires. To run in production and be ready to scale, you need to use more production ready components.

• Put the Fn API behind a load balancer and launch run several instances of them (the more the merrier).
• Run a database that can scale.
• Asynchronous functions requires a message queue (preferably one that scales).

Here\'s a rough diagram of what a production deployment looks like:

Load Balancer

Any load balancer will work, put every instance of Fn that you run behind the load balancer.

Database

We\'ve done our best to keep the database usage to a minimum. There are no writes during the request/response cycle which where most of the load will be.

The database is pluggable and we currently support a few options that can be found here. We welcome pull requests for more!

Message Queue

The message queue is an important part of asynchronous functions, essentially buffering requests for processing when resources are available. The reliability and scale of the message queue will play an important part in how well Fn runs, in particular if you use a lot of asynchronous function calls.

The message queue is pluggable and we currently support a few options that can be found here. We welcome pull requests for more!

Logging, Metrics and Monitoring

Logging is a particularly important part of Fn. It not only emits logs, but metrics are also emitted to the logs. Ops teams can then decide how they want to use the logs and metrics without us prescribing a particular technology. For instance, you can logspout-statsd to capture metrics from the logs and forward them to statsd.

More about Metrics

Scaling

There are metrics emitted to the logs that can be used to notify you when to scale. The most important being the wait_timemetrics for both the synchronous and asynchronous functions. If wait_time increases, you\'ll want to start more Fn instances.