Awesome

maplibre-preload

A tiny (5.53 kB gziped) zero-configuration plugin for preloading tiles and smoothen the experience when using targeted movements in MapLibreGL JS.

Why?

It started here, a conversation about the need of precaching tiles when the user start a movement, and as one the reference mentioned there was MapWorkBox, a little PoC I built for testing preemptive tiles caching using Service Workers.

So, the idea is to smooth the rendering of the animation frames of the camera and final scenario when using targeted movements map methods (panTo, zoomTo, jumpTo, easeTo and flyTo). The standard applications request new tiles as they change the camera view during the animation, and maybe batch-preloading the tiles of the final scenarion, the final animation transition might look way better.

How

This little pluging make use of inline webworkers to preload the tiles, so it is not run in the main thread. As of today, this plugin offers the next features:

• Full final scenario preload
• Partial inbetween animation preload

It adds precached map methods to the old ones:

• cachedPanTo
• cachedZoomTo
• cachedJumpTo
• cachedEaseTo
• cachedFlyTo

With the same signature than the original methods, but the run flag (boolean, default false) to trigger the original methods or just preload the tiles for future use.

Changelog

• v0.0.2
  • [Fix] Return the map object in the cached__To methods to keep the original output
  • [Fix] Use the Bresenham algorithm to preload only the tiles in the start -> end path instead of all the tiles in the bounding box defined by those two points
  • [Feature] Add a run flag to allow preloading with/without running the actual ___To map methods. Intended to enable the preload of the movement when we can expect the animation later (v.g.: preload when hovering a button, flyTo when clicking)
  • [Feature] Hide log messages behind a debug flag (boolean, default false)
• v0.0.1
  • Initial release

To Do

• Add bearing and pitch logic.
• Improve the management of zoom levels while the animation, as it is not a constant zmin value

How to use

npm install
npm run build

And you will find several versions of the library at /dist (more info at Microbundle)

• maplibre-preload.cjs: CommonJS bundle
• maplibre-preload.module.js: ESM bundle
• maplibre-preload.modern.js: Designed to work in all modern browsers, generally smaller and faster to execute than the plain ESM bundle.

You can find a simple example here, run it in a local server and check the dev console to watch how the map hits the cache for all the cached tiles, giving the render engine an estimated extra time of

0.9 * single_tile_loading_time * num_of_tiles_in_final_scenario / 6

(as per my simple benchmarks)

Caveats and final thoughts

Workers comply with the global max number of connections per host name (typically 6 while using HTTP1.1, with HTTP2... will see), so preloading will always interfere with standard requestes made by panTo, zoomTo, jumpTo, easeTo or flyTo. So, those expected requests are queued while the prefetch requests are resolved, eventually leading to rendering glitches as the animation goes on but there is no data to render. That can be easily analyzed checking the waterfall in the network panel of the dev console.

With optimal network conditions, the above issue is almost unnoticeable because, while the animation takes ~ 5000ms, it takes ~ 400ms to preload all the tiles, and then, retrieveing each tile from cache takes like ~ 5ms per batch of 6.

This being said, prefetching needs an extra-fine tuning to select the right amount of tiles to be preloaded, maybe just the final scenario. Or maybe it just doesn't make sense at all. I'd love to hear about use cases where the pros outweigh the cons.