Parts of Positron's Lake Michigan’s turns white depending on zooming

[ramiroaznar]

A user has detected a strange behaviour within the Positron basemap. In zoom levels 7 and 8 parts of Lake Michigan turns white.

[pnorman]

This is probably a case of the great lakes are not rendering, but only on some servers

• The Great Lakes are coming from water multipolygons now as opposed to coastline data in OSM
• The water queries are complex and confusing
• The global false background is land, so it's a case of missing water instead of excess land
• The water is coming from #land_negative where is_lake=1

(SELECT
    id::bigint,
    the_geom_webmercator,
    name,
    type,
    is_lake,
    ne_scalerank,
    area
  FROM water_zoomed('!scale_denominator!', !bbox!) AS _)
UNION ALL
(SELECT
    cartodb_id::bigint,
    the_geom_webmercator,
    '' AS name,
    'ocean' AS type,
    1 AS is_lake,
    0 AS ne_scalerank,
    999999999 AS area
  FROM land_negative_zoomed('!scale_denominator!', !bbox!) AS _)

• land_negative_zoomed only kicks in at z9
• on z7 water_zoomed is

(SELECT
    cartodb_id::bigint AS id,
    the_geom_webmercator,
    name::text,
    ''river'' AS type,
    0 AS is_lake,
    scalerank::integer,
    ST_Area(the_geom_webmercator)::bigint
  FROM ne_10m_rivers_lake_centerlines
  WHERE the_geom_webmercator && $1)
UNION 
(SELECT
    cartodb_id::bigint AS id,
    the_geom_webmercator,
    name,
    ''lake'' AS type,
    1 AS is_lake,
    scalerank::integer,
    ST_Area(the_geom_webmercator)::bigint
  FROM ne_10m_lakes
  WHERE the_geom_webmercator && $1)
  ORDER BY is_lake ASC

• on z8 water_zoomed is

SELECT id::bigint AS id, the_geom_webmercator, name::text, type::text,
    (CASE 
      WHEN type IN (''water'',''bay'',''riverbank'',''reservoir'') 
        AND ST_GeometryType(the_geom_webmercator) IN (''ST_Polygon'',''ST_MultiPolygon'') THEN 1 
      ELSE 0 
    END) as is_lake,
    0 as ne_scalerank,
    area::bigint
  FROM water_areas_z10
  WHERE the_geom_webmercator && $1
  ORDER BY is_lake ASC

• water_areas_z10 is a mat view, and is effectively

SELECT
    tags->'name' AS name,
    ST_Area(the_geom)::bigint AS area,
    COALESCE(tags -> 'natural', tags -> 'waterway', tags -> 'landuse') AS type,
    generalize(generalize(the_geom ,13),10) AS the_geom_webmercator
  FROM planet
  WHERE ST_Area(the_geom)::bigint > 25000 OR COALESCE(tags -> 'natural', tags -> 'waterway', tags -> 'landuse') = 'riverbank'
    AND (tags -> 'natural' IN ('water','bay')
      OR tags -> 'waterway' IN ('riverbank','canal','river')
      OR tags -> 'landuse' IN ('reservoir'))

• On z8 it looks like a problem with one of the three mat view refreshes or planet, and on z7 it looks like a problem with ne_10m_lakes

[pnorman]

z8 was a problem with a transient problem in OSM data sometime in the past which has been fixed, and errors in the code calling refresh.js were stopping it from correctly refreshing the materialized views on some servers. It happened on some tiles but not others because different servers were behaving differently.

I've manually refreshed and @zenitraM is fixing the script issue.

The problems on z7 seem to have gone away, please comment if you see them again after the next hour or so.