http-range-fetcher

http-range-fetcher

Cache/manager for HTTP byte-range requests that merges requests together and caches results. Designed for applications that request lots of small byte ranges over HTTP that are often adjacent to each other.

Uses cross-fetch in the backend, so it works either in node or webpack/browserify. Respects HTTP caching semantics, with the exception of setting a default minimum TTL of 1 second on requests that are not really supposed to be cached (e.g. Cache-Control: no-cache). You can turn that behavior off by setting minimumTTL to 0 though.

Install

$ npm install --save http-range-fetcher

Usage

const { HttpRangeFetcher } = require('http-range-fetcher')

const cache = new HttpRangeFetcher({})
cache.getRange('http://foo.bar/baz.bam', 20, 10)
.then( response => {
  assert(response.buffer.length === 10)
  assert(response.headers['content-range'] === '20-29/23422')
  // response objects contain `headers` and `buffer`.  the `headers` object
  // contains the original headers that came from the server in response to the
  // aggregated call, except the Content-Range header has been overwritten
  // to match the requested range, and it adds a X-Resource-Length header that
  // conveniently gives the total length of the remote resource so you don't
  // have to parse the Content-Range header.
  assert(response.headers['x-resource-length'] === '23422')
})

// these will be aggregated behind the scenes
// as a single request for a big chunk of the remote file,
// which will be cached to satisfy subsequent requests
Promise.all([
    cache.getRange('http://foo.bar/baz.bam', 20, 10),
    cache.getRange('http://foo.bar/baz.bam', 30, 10),
    cache.getRange('http://foo.bar/baz.bam', 40, 10),
    cache.getRange('http://foo.bar/baz.bam', 50, 10),
    cache.getRange('http://foo.bar/baz.bam', 60, 10),
    cache.getRange('http://foo.bar/baz.bam', 70, 10),
])
.then(fetchResults => {
    fetchResults.forEach(res => assert(res.buffer.length === 10))
})

API

Table of Contents

• HttpRangeFetcher
  • getRange
  • stat
  • reset

HttpRangeFetcher

smart cache that fetches chunks of remote files. caches chunks in an LRU cache, and aggregates upstream fetches

Parameters

• $0 Object
  • $0.fetch (optional, default crossFetchBinaryRange) callback with signature (key, start, end) => Promise({ headers, buffer })
  • $0.size (optional, default 10000000) size in bytes of cache to keep
  • $0.chunkSize (optional, default 32768) size in bytes of cached chunks
  • $0.aggregationTime (optional, default 100) time in ms over which to pool requests before dispatching them
  • $0.minimumTTL (optional, default 1000) time in ms a non-cacheable response will still be cached
  • $0.maxFetchSize (optional, default chunkSize * 4) maximum size of an aggregated request
  • $0.maxExtraFetch (optional, default chunkSize) max number of additional bytes to fetch when aggregating requests that don't actually overlap

getRange

Fetch a range of a remote resource.

Parameters

• key string the resource's unique identifier, this would usually be a URL. This is passed along to the fetch callback.
• position number? offset in the file at which to start fetching (optional, default 0)
• length number? number of bytes to fetch, defaults to the remainder of the file
• options object? request options (optional, default {})
  • options.signal AbortSignal? object that can be used to abort the fetch. See AbortController on MDN for details

Returns Promise for a response object containing { headers, buffer }

stat

Fetches the first few bytes of the remote file (if necessary) and uses the returned headers to populate a fs-like stat object.

Currently, this attempts to set size, mtime, and mtimeMs, if the information is available from HTTP headers.

Parameters

• key string

Returns Promise for a stats object like { size, mtime, mtimeMs }

reset

Throw away all cached data, resetting the cache.

Academic Use

This package was written with funding from the NHGRI as part of the JBrowse project. If you use it in an academic project that you publish, please cite the most recent JBrowse paper, which will be linked from jbrowse.org.

License

MIT © Robert Buels