I’ve been working on a simple streaming parser for plain text maps. While writing the browser example (I originally wrote it for Node), I began to wonder if I could use the Power of Streaming to use it to render some really huge maps.
I gave it a map file with about 16 million entities that were each to render to three SVG entities. After fiddling with each of the streams involved to make them respect backpressure, I was able to keep the browser responsive for a while while it rendered these entities. However, at some point, the page would inevitably become unresponsive and the tab would crash.
Who to blame?
<g> elements, each containing a
Chrome 34.0.1847.131 quits on it (“Aw, Snap!” page) on a Mac running Mavericks with a 2.8 GHz Core i7 and 16 GB RAM. You can try it yourself.
(However, Chrome will successfully load a page with 480,000
Streaming into a lagoon
I think there’s a lesson here about streaming as well. Streams are about processing data a manageable chunk at a time, but if you can’t dispose of those chunks after you’re done with them, those chunks pool up, and the chunk-at-a-time benefit of streams is negated.
In the situation I set up — piping a huge text file to a parser stream, then piping the parsed tokens to a renderer — streams of text were transformed into streams of token objects which were then transformed into SVG elements. But those SVG elements did not “pass through.” They piled up in the DOM, eating up memory.
Not every object benefits from being hit with a hammer, and some situations benefit less than others from having a streaming pattern applied.