Optimize indexes for specific map styles #10215

New issue

Open

opened 2025-02-05 15:09:57 +00:00 by pastk · 0 comments

pastk commented

2025-02-05 15:09:57 +00:00

Owner

But in fact OM has just one index used for all styles. Which means it has to include all features at the minimum possible zoom level across all styles.
That makes it somewhat less fast/efficient for the general style and even more so for the vehicle style (it includes much fewer features).

The same could apply to e.g. search index.

Its possible to have a standalone index specifically for each map style.
It'd increase maps sizes. E.g. I've checked a random map and a visibility index there takes 5%. 3 indexes (general, vehicle, outdoor) would inflate this map by 10%.
It should make index scans faster though. Benchmarking is needed to estimate potential gains and to avoid premature optimization.

Another possibility is to have one base visibility index (features at same min zoom present in all styles) and to store only differences in per-style indexes.
This would decrease maps size penalty.
But OM would have to scan two indexes (base + difference). This might kill all potential performance benefits.

E.g. visibility index needed for the outdoor style is generally bigger than for general one, because the outdoor one has the same set of features but some of them displayed at lower zoom levels. Also there are some features (like minor power lines) that are not needed for the general style at all. But in fact OM has just one index used for all styles. Which means it has to include all features at the minimum possible zoom level across all styles. That makes it somewhat less fast/efficient for the general style and even more so for the vehicle style (it includes much fewer features). The same could apply to e.g. search index. Its possible to have a standalone index specifically for each map style. It'd increase maps sizes. E.g. I've checked a random map and a visibility index there takes 5%. 3 indexes (general, vehicle, outdoor) would inflate this map by 10%. It should make index scans faster though. Benchmarking is needed to estimate potential gains and to avoid premature optimization. Another possibility is to have one base visibility index (features at same min zoom present in all styles) and to store only differences in per-style indexes. This would decrease maps size penalty. But OM would have to scan two indexes (base + difference). This might kill all potential performance benefits.

Rows
Columns