Release 0.16.2

Use the mode to aggregate category columns in overviews

Makefile

@@ -1,7 +1,7 @@
 # cartodb/Makefile
 
 EXTENSION = cartodb
-EXTVERSION = 0.16.0
+EXTVERSION = 0.16.2
 
 SED = sed
 
@@ -67,6 +67,8 @@ UPGRADABLE = \
   0.15.0 \
   0.15.1 \
   0.16.0 \
+  0.16.1 \
+  0.16.2 \
   $(EXTVERSION)dev \
   $(EXTVERSION)next \
   $(END)

NEWS.md

@@ -1,3 +1,22 @@
+0.16.2 (2016-04-27)
+-------------------
+
+* Use the mode to aggregate category columns in overviews
+  [#246](https://github.com/CartoDB/cartodb-postgresql/pull/246)
+
+0.16.1 (2016-04-25)
+-------------------
+
+* Optimize column information functions performance
+  [#238](https://github.com/CartoDB/cartodb-postgresql/pull/238)
+
+* Adjust overview points to pixel CDB_EqualIntervalBins
+  [#242](https://github.com/CartoDB/cartodb-postgresql/pull/242)
+
+* Compute webmercator resolution using full numeric precision
+  [#243](https://github.com/CartoDB/cartodb-postgresql/pull/243)
+
+
 0.16.0 (2016-04-15)
 -------------------
 * Adds table for storing camshaft analysis nodes

doc/CDB_Overviews.md

@@ -2,18 +2,25 @@ Overviews are tables that represent a *reduced* version of a dataset intended
 for efficient rendering at certain zoom levels while preserving the
 general visual appearance of the complete dataset.
 
-The *reduction* consists in a fewer number of records
+The *reduction* consists in havig a fewer number of records
 (while each overview record may represent an aggregation of multiple records)
 and/or simplified record geometries.
 
-Overviews are created through the `CDB_CreateOverviews`.
+Overviews are created through the `CDB_CreateOverviews` function.
 The statement timeout may need to be adjusted before using this function,
 as overview creation for large tables is a time-consuming operation.
 
 The `CDB_Overviews` function can be used determine what overview tables
 exist for a given dataset table and which zoom levels correspond to it.
 
-The `CDB_DropOverviews` remove a dataset's existing overviews.
+The `CDB_DropOverviews` function removes a dataset's existing overviews.
+
+To know if overview tables exist for some base table, and to obtain
+a list of which overview tables are approrpiate for which zoom levels,
+the `CDB_Overviews` functions can be used.
+
+The zoom level we're referring here to are those used
+by the tiler: http://wiki.openstreetmap.org/wiki/Zoom_levels
 
 ### CDB_CreateOverviews
 
@@ -51,10 +58,14 @@ CDB_CreateOverviews(table_name, ref_z_strategy, reduction_strategy)
 #### Tolerance / level of detail
 
 The level of detail to be representable by each overview layer can
-be specified as a tolerance in pixels (if different from the default of 2 pixels)
+be specified as a tolerance in pixels (if different from the default of 1 pixel)
 with the function `CDB_CreateOverviewsWithToleranceInPixels`
 which has as a second additional argument the desired tolerance.
 
+This tolerance defines the maximum deviation in pixels of the overviews
+geometries with respect to the original geometries when overview tables
+are used for their intendend zoom level.
+
 ### CDB_Overviews
 
 Obtain overview metadata for a given table (existing overviews).
@@ -79,7 +90,7 @@ SELECT CDB_Overviews(CDB_QueryTablesText('SELECT * FROM table1, table2'));
 The result of `CDB_Overviews` has three columns:
 
 | base_table | z | overview_table |
-|------------+---+----------------|
+| ---------- | - | -------------- |
 | table1     | 1 | table1_ov1     |
 | table1     | 2 | table1_ov2     |
 | table1     | 4 | table1_ov4     |

doc/cartodbfy-requirements.rst

@@ -33,7 +33,7 @@ Additionally, a CartoDB table can contain other columns.
 
 See the `CartoDB User Table documentation`_
 
-.. _CartoDB User Table documentation: https://github.com/CartoDB/cartodb-postgresql/blob/master/doc/CartoDB-user-table.md 
+.. _CartoDB User Table documentation: https://github.com/CartoDB/cartodb-postgresql/blob/master/doc/CartoDB-user-table.rst 
 for further information.
 
 High level requirements

scripts-available/CDB_ColumnNames.sql

@@ -2,15 +2,13 @@
 CREATE OR REPLACE FUNCTION CDB_ColumnNames(REGCLASS)
 RETURNS SETOF information_schema.sql_identifier
 AS $$
-
-    SELECT c.column_name
-      FROM information_schema.columns c, pg_class _tn, pg_namespace _sn
-      WHERE table_name = _tn.relname
-        AND table_schema = _sn.nspname
-        AND _tn.oid = $1::oid
-        AND _sn.oid = _tn.relnamespace
-      ORDER BY ordinal_position;
-
+  SELECT
+    a.attname::information_schema.sql_identifier column_name
+    FROM pg_class c
+         LEFT JOIN pg_attribute a ON a.attrelid = c.oid
+    WHERE c.oid = $1::oid
+    AND a.attstattarget < 0 -- exclude system columns
+   ORDER BY a.attnum;
 $$ LANGUAGE SQL;
 
 -- This is to migrate from pre-0.2.0 version

scripts-available/CDB_ColumnType.sql

@@ -2,15 +2,13 @@
 CREATE OR REPLACE FUNCTION CDB_ColumnType(REGCLASS, TEXT)
 RETURNS information_schema.character_data
 AS $$
-
-    SELECT c.data_type
-      FROM information_schema.columns c, pg_class _tn, pg_namespace _sn
-      WHERE table_name = _tn.relname
-        AND table_schema = _sn.nspname
-        AND column_name = $2
-        AND _tn.oid = $1::oid
-        AND _sn.oid = _tn.relnamespace;
-         
+  SELECT
+    format_type(a.atttypid, NULL)::information_schema.character_data data_type
+  FROM pg_class c
+       LEFT JOIN pg_attribute a ON a.attrelid = c.oid
+  WHERE c.oid = $1::oid
+  AND a.attname = $2
+  AND a.attstattarget < 0; -- exclude system columns
 $$ LANGUAGE SQL;
 
 -- This is to migrate from pre-0.2.0 version

scripts-available/CDB_Overviews.sql

@@ -88,6 +88,26 @@ AS $$
   END;
 $$ LANGUAGE PLPGSQL IMMUTABLE;
 
+CREATE OR REPLACE FUNCTION _CDB_OverviewBaseTable(overview_table REGCLASS)
+RETURNS REGCLASS
+AS $$
+  DECLARE
+    table_name TEXT;
+    schema_name TEXT;
+    base_name TEXT;
+    base_table REGCLASS;
+  BEGIN
+    SELECT * FROM _cdb_split_table_name(overview_table) INTO schema_name, table_name;
+    base_name := _CDB_OverviewBaseTableName(table_name);
+    IF base_name != table_name THEN
+      base_table := Format('%I.%I', schema_name, base_name)::regclass;
+    ELSE
+      base_table := overview_table;
+    END IF;
+    RETURN base_table;
+  END;
+$$ LANGUAGE PLPGSQL IMMUTABLE;
+
 -- Schema and relation names of a table given its reloid
 -- Scope: private.
 -- Parameters
@@ -520,18 +540,65 @@ CREATE OR REPLACE FUNCTION _cdb_unlimited_text_column(reloid REGCLASS, col_name
 RETURNS BOOLEAN
 AS $$
   SELECT EXISTS (
-    SELECT  *
-      FROM information_schema.columns c, pg_class _tn, pg_namespace _sn
-      WHERE table_name = _tn.relname
-        AND table_schema = _sn.nspname
-        AND c.column_name = col_name
-        AND _tn.oid = reloid
-        AND _sn.oid = _tn.relnamespace
-        AND character_maximum_length IS NULL
-        AND c.data_type IN ('text', 'character varying', 'character')
+    SELECT a.attname
+    FROM pg_class c
+         LEFT JOIN pg_attribute a ON a.attrelid = c.oid
+         LEFT JOIN pg_type t ON t.oid = a.atttypid
+    WHERE c.oid = reloid
+      AND a.attname = col_name
+      AND format_type(a.atttypid, NULL) IN ('text', 'character varying', 'character')
+      AND format_type(a.atttypid, NULL) = format_type(a.atttypid, a.atttypmod)
   );
 $$ LANGUAGE SQL STABLE;
 
+CREATE OR REPLACE FUNCTION _cdb_categorical_column(reloid REGCLASS, col_name TEXT)
+RETURNS BOOLEAN
+AS $$
+DECLARE
+    schema_name TEXT;
+    table_name TEXT;
+    available BOOLEAN;
+    categorical BOOLEAN;
+BEGIN
+    SELECT * FROM _cdb_split_table_name(reloid) INTO schema_name, table_name;
+    SELECT n_distinct IS NOT NULL
+    FROM pg_stats
+    WHERE pg_stats.schemaname = schema_name
+      AND pg_stats.tablename = table_name
+      AND pg_stats.attname = col_name
+    INTO available;
+    IF available IS NULL OR NOT available THEN
+      EXECUTE Format('ANALYZE %s;', reloid);
+    END IF;
+    SELECT n_distinct > 0 AND n_distinct <= 20
+    FROM pg_stats
+    WHERE pg_stats.schemaname = schema_name
+      AND pg_stats.tablename = table_name
+      AND pg_stats.attname = col_name
+    INTO categorical;
+    RETURN categorical;
+END;
+$$ LANGUAGE PLPGSQL VOLATILE;
+
+CREATE OR REPLACE FUNCTION _cdb_mode_of_array(anyarray)
+  RETURNS anyelement AS
+$$
+    SELECT a
+    FROM unnest($1) a
+    GROUP BY 1
+    ORDER BY COUNT(1) DESC, 1
+    LIMIT 1;
+$$
+LANGUAGE SQL IMMUTABLE;
+
+-- Tell Postgres how to use our aggregate
+CREATE AGGREGATE _cdb_mode(anyelement) (
+  SFUNC=array_append,
+  STYPE=anyarray,
+  FINALFUNC=_cdb_mode_of_array,
+  INITCOND='{}'
+);
+
 -- SQL Aggregation expression for a datase attribute
 -- Scope: private.
 -- Parameters
@@ -549,6 +616,7 @@ DECLARE
   has_counter_column BOOLEAN;
   feature_count TEXT;
   total_feature_count TEXT;
+  base_table REGCLASS;
 BEGIN
   IF table_alias <> '' THEN
     qualified_column := Format('%I.%I', table_alias, column_name);
@@ -569,20 +637,30 @@ BEGIN
     total_feature_count := 'count(*)';
   END IF;
 
+  base_table := _CDB_OverviewBaseTable(reloid);
+
   CASE column_type
   WHEN 'double precision', 'real', 'integer', 'bigint', 'numeric' THEN
     IF column_name = '_feature_count' THEN
       RETURN 'SUM(_feature_count)';
     ELSE
-      RETURN Format('SUM(%s*%s)/%s::' || column_type, qualified_column, feature_count, total_feature_count);
+      IF column_type = 'integer' AND _cdb_categorical_column(base_table, column_name) THEN
+        RETURN Format('CDB_Math_Mode(%s)::', qualified_column) || column_type;
+      ELSE
+        RETURN Format('SUM(%s*%s)/%s::' || column_type, qualified_column, feature_count, total_feature_count);
+      END IF;
     END IF;
   WHEN 'text', 'character varying', 'character' THEN
-    IF _cdb_unlimited_text_column(reloid, column_name) THEN
-      -- TODO: this should not be applied to columns containing largish text;
-      -- it is intended only to short names/identifiers
-      RETURN  'CASE WHEN count(distinct ' || qualified_column || ') = 1 THEN MIN(' || qualified_column || ') WHEN ' || total_feature_count || ' < 5 THEN string_agg(distinct ' || qualified_column || ','' / '') ELSE ''*'' END::' || column_type;
+    IF _cdb_categorical_column(base_table, column_name) THEN
+      RETURN Format('_cdb_mode(%s)::', qualified_column) || column_type;
     ELSE
-      RETURN 'CASE count(*) WHEN 1 THEN MIN(' || qualified_column || ') ELSE NULL END::' || column_type;
+      IF _cdb_unlimited_text_column(base_table, column_name) THEN
+        -- TODO: this should not be applied to columns containing largish text;
+        -- it is intended only to short names/identifiers
+        RETURN  'CASE WHEN count(distinct ' || qualified_column || ') = 1 THEN MIN(' || qualified_column || ') WHEN ' || total_feature_count || ' < 5 THEN string_agg(distinct ' || qualified_column || ','' / '') ELSE ''*'' END::' || column_type;
+      ELSE
+        RETURN 'CASE count(*) WHEN 1 THEN MIN(' || qualified_column || ') ELSE NULL END::' || column_type;
+      END IF;
     END IF;
   WHEN 'boolean' THEN
     RETURN 'CASE count(*) WHEN 1 THEN BOOL_AND(' || qualified_column || ') ELSE NULL END::' || column_type;
@@ -652,7 +730,13 @@ AS $$
     overview_rel TEXT;
     reduction FLOAT8;
     base_name TEXT;
+    pixel_m FLOAT8;
     grid_m FLOAT8;
+    offset_m FLOAT8;
+    offset_x TEXT;
+    offset_y TEXT;
+    cell_x TEXT;
+    cell_y TEXT;
     aggr_attributes TEXT;
     attributes TEXT;
     columns TEXT;
@@ -678,8 +762,10 @@ AS $$
 
     SELECT * FROM _cdb_split_table_name(reloid) INTO schema_name, table_name;
 
-    -- compute grid cell size using the overview_z dimension...
-    SELECT CDB_XYZ_Resolution(overview_z)*grid_px INTO grid_m;
+    -- pixel_m: size of a pixel in webmercator units (meters)
+    SELECT CDB_XYZ_Resolution(overview_z) INTO pixel_m;
+    -- grid size in meters
+    grid_m = grid_px * pixel_m;
 
     attributes := _CDB_Aggregable_Attributes_Expression(reloid);
     aggr_attributes := _CDB_Aggregated_Attributes_Expression(reloid);
@@ -690,7 +776,21 @@ AS $$
       aggr_attributes := aggr_attributes || ', ';
     END IF;
 
-    point_geom = Format('ST_SetSRID(ST_MakePoint(gx*%1$s + %2$s, gy*%1$s + %2$s), 3857)', grid_m, grid_m/2);
+    -- Center of each cell:
+    cell_x := Format('gx*%1$s + %2$s', grid_m, grid_m/2);
+    cell_y := Format('gy*%1$s + %2$s', grid_m, grid_m/2);
+
+    -- Displacement to the nearest pixel center:
+    IF MOD(grid_px::numeric, 1.0::numeric) = 0 THEN
+      offset_m := pixel_m/2 - MOD((grid_m/2)::numeric, pixel_m::numeric)::float8;
+      offset_x := Format('%s', offset_m);
+      offset_y := Format('%s', offset_m);
+    ELSE
+      offset_x := Format('%2$s/2 - MOD((%1$s)::numeric, (%2$s)::numeric)::float8', cell_x, pixel_m);
+      offset_y := Format('%2$s/2 - MOD((%1$s)::numeric, (%2$s)::numeric)::float8', cell_y, pixel_m);
+    END IF;
+
+    point_geom := Format('ST_SetSRID(ST_MakePoint(%1$s + %3$s, %2$s + %4$s), 3857)', cell_x, cell_y, offset_x, offset_y);
 
     -- compute the resulting columns in the same order as in the base table
     WITH cols AS (

scripts-available/CDB_XYZ.sql

@@ -6,7 +6,7 @@ CREATE OR REPLACE FUNCTION CDB_XYZ_Resolution(z INTEGER)
 RETURNS FLOAT8
 AS $$
   -- circumference divided by 256 is z0 resolution, then divide by 2^z
-  SELECT 40075017.0 / 256 / power(2, z);
+  SELECT 6378137.0*2.0*pi() / 256.0 / power(2.0, z);
 $$ LANGUAGE SQL IMMUTABLE STRICT;
 -- }