作为代理，RESTBase本身不执行任何重要的内容处理。 相反，它在需要时请求来自后端服务的内容转换，并且通常（取决于配置）将其存储回来供以后检索。 对于高容量静态端点，大多数请求将直接从存储中满足。
Its storage backends expose a RESTful storage API similar to Amazon DynamoDB and Google DataStore. The primary implementation uses Apache Cassandra. Notable features include automatically maintained secondary indexes and some lightweight transaction support. A SQLite backend has been developed and is the default backend in the package.
RESTBase automatically emits statsd metrics about all storage and backend requests. This provides a good baseline level of performance and error instrumentation in a micro-service architecture.
Our first use case is speeding up VisualEditor by reducing HTML size, and eliminating Varnish cache misses. RESTBase stores Parsoid metadata separately from the HTML of the page, which reduces the size of the latter by about 40%. RESTBase provides only this HTML to VE, which reduces network transfer and processing latency significantly. In the longer term, we are aiming to bring down the size of the HTML to that of current PHP parser output to make it suitable for regular page views. This has the potential to make switching to visual editing instantaneous and free of any scrolling.
If parse time is not a pressing concern for your wiki (for example it does not have complex templates or large transclusion counts), then accessing Parsoid directly may make more sense than introducing a dependency on RESTBase.
Another use case we are strongly interested in is providing a section-level editing API for micro-contributions and extremely fast VisualEditor saves, even faster than wikitext.
- Browse the source code
- Follow architectural discussions about RESTBase by joining & watching the RESTBase-architecture project
- Report issues on Phabricator, project 'RESTBase'
- RESTBase/Alternative architectural options considered
- RESTBase/Table storage backend options
- Original requests for comments: Storage service (talk) and Content API (talk)