As of now, the 1st example works for me in both the thumbnail view as well as fullscreen.

I can't find the 2nd example anywhere in the history of https://de.wikipedia.org/wiki/Benutzerin:Johanna_Strodt_(WMDE)/Spielwiese/Kartentests.

I was playing around with this URL:

• https://maps.wikimedia.org/geoline?getgeojson=1&ids=Q316012 currently returns the expected GeoJSON form me.
• https://maps.wikimedia.org/geoline?getgeojson=1&ids=Q316012&foo=1 does the same.
• https://maps.wikimedia.org/geoline?getgeojson=1&ids=Q316012&foo=4 returns a 404 and says "Cannot GET /geoline". This doesn't make sense.
• When I continue increasing the number I randomly get 404s for some. I don't see a pattern. Trying to reload a failing URL never fixes it. But increasing the number typically does.
• I also have seen empty results like {"type":"FeatureCollection","features":[]} but currently can't reproduce this any more.

In other words: No wonder we have so many broken examples where either the thumbnail or the fullscreen map (or both) don't show the shape. Unfortunately I have no idea how to proceed from here.

In T312913#8093961, @thiemowmde wrote:

As of now, the 1st example works for me in both the thumbnail view as well as fullscreen.

On my device, the problem currently exists:

I can't find the 2nd example anywhere in the history of https://de.wikipedia.org/wiki/Benutzerin:Johanna_Strodt_(WMDE)/Spielwiese/Kartentests.

It was hidden in the history, I have now added it back to the current version. My apologies for sending you searching.
For me, the line is currently not shown at all:

In T312913#8097787, @thiemowmde wrote:

• I also have seen empty results like {"type":"FeatureCollection","features":[]} but currently can't reproduce this any more.

@thiemowmde maybe T313844 is of interest as well?

I can confirm that the last example is because of the same "random failure". When I add a dummy parameter to the end of https://maps.wikimedia.org/geoline?getgeojson=1&ids=Q35333853 I sometimes get the data, sometimes empty GeoJSON as shown above, and sometimes a 404.

T313844 appears to be something else.

I assume there's some caching and replication involved in the entire setup (Is there any documentation of the architecture availble? Couldn't find it on first glimpse). Could it be related to a single rogue server in a load balanced farm behind some cache line? Because that's what it looks like to me. It works most of the time, sometime it breaks, but after a while, it is fine again.

Architecture diagrams are in https://phabricator.wikimedia.org/diffusion/MSKA/browse/master/docs/diagrams/output/ if you are curious.

Another example:
https://maps.wikimedia.org/geoshape?getgeojson=1&ids=Q106908292 -- empty
https://maps.wikimedia.org/geoshape?getgeojson=1&ids=Q106908292&foo=3 -- works
https://maps.wikimedia.org/geoshape?getgeojson=1&ids=Q106908292&d=1 -- works
https://maps.wikimedia.org/geoshape?getgeojson=1&ids=Q106908292&x=1 -- 404
From: https://de.wikipedia.org/wiki/Hohes_Holz_mit_Ketzheide_und_Gew%C3%A4ssern -- Currently, the shape is visible in the PNG, not in fullscreen.

I just did a sample:

13.1% of my requests end in a 404
47.6% of my requests return an empty geojson object {"type":"FeatureCollection","features":[]}
39.4% of my requests return the correct geojson object containing all the points from OSM.

So over 60% of the time, the rendering of the geojson data from OSM on the maps fails. This means that the chance is 1/6 that both the PNG and the full screen map show the OSM content.

We found T307671: High rate of 5XX errors from maps.wikimedia.org since 2022-05-05 ~03:20 where this small but significant change was made: https://gerrit.wikimedia.org/r/789598. It appears like this was merely silencing the log spam without actually fixing anything. What we get now are the empty {"type":"FeatureCollection","features":[]} described in this ticket here.

For reference, the code that creates this little snippet of GeoJSON is here: https://phabricator.wikimedia.org/diffusion/MSKA/browse/master/lib/snapshot/mapdataLoader.js$99. This is the same file touched in T307671. I suspect the error might be in this file.