Tiles cached at wrong zoom level

[juba]

Hi,

We are using bbox to create a new version of the lifemap website (version still in development, not public yet). It is working very smoothly, except for one issue : sometimes certain tiles are cached at the wrong zoom level, leading to display inconsistencies.

Here is an example screenshot: in this case the central tiles have been cached at a different (wrong) zoom level than the surrounding ones.

When this problem arises, clearing the bbox tile cache solves the problem.

I'm sorry to not have a reproducible example to give, but maybe other people did encounter the same issue ? Or someone may have an idea about how to prevent this kind of problem ?

Many thanks in any case...

[pka]

You're using a file based cache, right? Are there any special settings like https://www.bbox.earth/docs/tile-server/reference/#cachelimit ?

[juba]

Yes, we use a file based cache, I don't think we have any special settings as our cache config is limited to:

[[tilestore]]
name = "tilecache"
[tilestore.files]
base_dir = "/opt/data/tilecache"

[FabianRechsteiner]

I've encountered a similar issue where tiles are displayed incorrectly, especially at large scales, on www.vectormap.ch/basemapav.

[[tilestore]]
name = "tilecache"
[tilestore.files]
base_dir = "/tmp/tilecache"

[pka]

My first guess for the initial example was that during the zoom process a part doesn't get overwritten by the new zoom level possibly because of a timeout or an error. Speaking against that is that deleting the cache fixes it.

The second example is difficult to understand. On the right side it looks more like a tile border problem on the same zoom level. Is there a place where I can see the error? Bahnhofstrasse and Gartenstrasse are hard to find...

[FabianRechsteiner]

I can no longer find Gartenstrasse myself. Generally, you can zoom in anywhere, and you'll find tile errors if you pan around at a large scale.

[juba]

I may have a live example on our development website, if it can be useful.

• go here: https://lifemap.cnrs.fr/tree?tool=search
• search for "archosauria"
• click on the corresponding search result

You should then arrive to a place where there is a tile discrepancy:

[pka]

This looks more like shifted by a few pixels, but with the correct zoom levels. When I zoom from the start page, I often see offsets between lines at tile borders. This looks like a numerical precision problem in tile generation?

[juba]

What I find strange is that these errors only appear for certain tiles at certain zoom levels, and they disappear (but may reappear elsewhere) if I clear the cache.

[pka]

By panning around in http://www.vectormap.ch/basemapav I found only tile errors like

This looks also like the the zoom level is correct, but that some layers in the tiles are missing. Are there any database error messages?

[pka]

What I find strange is that these errors only appear for certain tiles at certain zoom levels, and they disappear (but may reappear elsewhere) if I clear the cache.

Could you try once without cache?

[juba]

I've disabled server side caching on:

https://lifemap.cnrs.fr/tree

I've tried navigating the map for a few minutes and for the moment I didn't encounter any discrepancies.

[FabianRechsteiner]

Caching has been disabled, and the old cache files have been deleted. However, errors still appear in the tiles. The issue does not seem to be related to caching but rather to the tile generation process. I haven’t found any errors in the logs.

[pka]

@juba I've no idea how caching can interfere here. Could you either send me the --loglevel debug output (with caching) or do a screen sharing session with me?

@FabianRechsteiner We should move your problems into an other issue. The last image looks like a buffer_size problem with thick lines?

[FabianRechsteiner]

I moved my buffer_size problem to the new issue #90 (comment)

[juba]

Sorry for my late answer...

I've tried to adjust some settings regarding to buffer_size in bbox and RenderBuffer in OpenLayers, setting them both to the tile size, and I haven't met any discrepancy for a while. So I can't be sure the issue has been solved, but at least I can't reproduce it for the moment.