Use “cache-feature: true” for better performance with attachments

[sommerluk]

Fixes #3756

Changes proposed in this pull request:

• Add cache-feature: true for all layers that use attachments to get better performance. We have quite a lot of them…

Test rendering with links to the example places: Should trigger no rendering change.

This PR would need performance testing, but I am not able to do that.

Another question: What about adding cache-feature: true to all layers? If this would be performance-neutral, it might be worth to do so, to prevent adding new attachments while forgetting about adding cache-feature: true?

[pnorman]

Another question: What about adding cache-feature: true to all layers? If this would be performance-neutral, it might be worth to do so, to prevent adding new attachments while forgetting about adding cache-feature: true?

I would expect it to boost RAM usage if done to all layers, but I have no idea how much.

I'm loading up a DB to run low zoom tests on, but if they're inconclusive, then I'll need to do high-zoom testing too. My current test setup use mapproxy, so because it creates the map object each render that flaw might overwhelm any performance differences, and I might need to go to renderd

[sommerluk]

For comparison: https://github.com/sommerluk/openstreetmap-carto/tree/cache02 provides code that uses cache-features for all layers.

[pnorman]

The parent of this branch took 158m31.916s while the branch took 148m17.483s. That's a 7% gain for z0-z10 pre-rendering, and we have relatively few attachments there. I expect higher gains at higher zooms, but don't feel the need to quantify exact gains.

I'm against adding cache-features to all layers regardless of attachment status without quantifying memory changes.

[sommerluk]

Thanks for testing!