Merge remote-tracking branch 'origin/master' into new_cartodbfy

Conflicts: test/CDB_QuotaTest.sql
2015-08-17 15:27:33 +02:00
parent a61a92a8f7 219d876973
commit 2b48f90374
45 changed files with 869 additions and 57 deletions
@@ -0,0 +1,122 @@
+--
+-- CDB_DistType classifies the histograms of a column into
+-- one of the basic types listed by Galtung: http://druedin.com/2012/12/08/galtungs-ajus-system/
+-- 
+-- Future improvements:
+--    variable number of bins (7 is baked in right now)
+--    catch the number of items to ensure that the sample is large enough
+--
+-- Refs:
+--    1. width_bucket/histograms: http://tapoueh.org/blog/2014/02/21-PostgreSQL-histogram
+--    2. R implementation: https://github.com/cran/agrmt
+
+CREATE OR REPLACE FUNCTION CDB_DistType ( in_array NUMERIC[] ) RETURNS text as $$
+DECLARE
+    element_count INT4;
+    minv numeric;
+    maxv numeric;
+    bins numeric[];
+    freqs numeric[];
+    ajus INT[];
+    freq INT4;
+    signature text;
+    i INT := 1;
+BEGIN
+    SELECT min(e), max(e), count(e) INTO minv, maxv, element_count FROM ( SELECT unnest(in_array) e ) x;
+
+    IF abs(maxv - minv) < 1e-7 THEN -- if max and min are nearly equal, call if 'F' (make relative to maxv?)
+        signature = 'F';
+    ELSE
+        -- Calculate bins and count in bins
+        EXECUTE 'WITH stats as (
+            SELECT min(e) as minv,
+                   max(e) as maxv,
+                   count(e) as total
+            FROM (SELECT unnest($1) e) x
+            WHERE e is not null
+        ),
+        hist as (
+            SELECT width_bucket(e, s.minv, s.maxv, 7) bucket,
+                   count(*) freq
+            FROM (SELECT unnest($1) e) x, stats s
+            WHERE e is not null
+            GROUP BY 1
+            ORDER BY 1
+        )
+        SELECT array_agg(round(100.0 * hist.freq::numeric / stats.total::numeric,1)) freqs,
+               array_agg(hist.bucket) buckets
+        FROM hist, stats'
+        INTO freqs, bins
+        USING in_array;
+
+        LOOP
+            IF i < 7 THEN
+                ajus[i] = CASE WHEN freqs[i] > freqs[i+1] THEN -1
+                               WHEN abs(freqs[i] - freqs[i+1]) <= 0.05 THEN 0
+                               ELSE 1 END;
+            ELSE
+                EXIT;
+            END IF;
+            i := i + 1;
+        END LOOP;
+
+        signature = _CDB_DistTypeClassify(ajus);
+    END IF;
+
+    RETURN signature;
+END;
+$$ language plpgsql IMMUTABLE;
+
+-- Classify data into AJUSFL
+
+CREATE OR REPLACE FUNCTION _CDB_DistTypeClassify ( in_array INT[] ) RETURNS text as $$
+DECLARE
+    element_count INT4;
+    maxv numeric;
+    minv numeric;
+    uniques INT[];
+    type text;
+BEGIN
+    SELECT max(e), min(e) INTO maxv, minv FROM ( SELECT unnest(in_array) e ) x;
+
+    IF (maxv = 0 AND minv = 0) THEN
+        type = 'F';
+    ELSIF maxv < 1 THEN
+        type = 'L';
+    ELSIF minv > -1 THEN
+        type = 'J';
+    ELSE
+        -- Get distinct elements ordered by original position
+        EXECUTE 'WITH b AS (
+            SELECT a
+            FROM (SELECT unnest($1) a) x
+        ),
+        c AS (
+            SELECT a, row_number() OVER () r
+            FROM b
+        ),
+        d AS (
+            SELECT DISTINCT a
+            FROM c
+        ),
+        e AS (
+            SELECT a FROM d ORDER BY (
+                SELECT r FROM c WHERE d.a = c.a ORDER BY r ASC LIMIT 1
+            ) ASC)
+        SELECT array_agg(a) FROM e'
+        INTO uniques
+        USING in_array;
+
+        -- Decide if it's an A, U, or other
+        IF (uniques = ARRAY[1,-1] OR uniques = ARRAY[1,0,-1] OR uniques = ARRAY[1,-1,0] OR uniques = ARRAY[0,1,-1]) THEN
+            type = 'A';
+        ELSIF (uniques = ARRAY[-1,1] OR uniques = ARRAY[-1,0,1] OR uniques = ARRAY[-1,1,0] OR uniques = ARRAY[0,-1,1]) THEN
+            type = 'U';
+        ELSE
+            type = 'S';
+        END IF;
+    END IF;
+
+    RETURN type;
+END;
+$$ language plpgsql IMMUTABLE;
@@ -0,0 +1,46 @@
+--
+-- CDB_DistinctMeasure 
+--     calculates the fraction of rows in the 10 most common distinct categories
+--     returns true if the number of rows in these 10 categories is >= 0.9 * total number of rows
+-- 
+-- 
+
+CREATE OR REPLACE FUNCTION CDB_DistinctMeasure ( in_array text[], threshold numeric DEFAULT null ) RETURNS numeric as $$
+DECLARE
+    element_count INT4;
+    maxval numeric;
+    passes numeric;
+BEGIN
+    SELECT count(e) INTO element_count FROM ( SELECT unnest(in_array) e ) x;
+
+    -- count number of occurrences per bin
+    -- calculate the normalized cumulative sum
+    -- return the max value: which corresponds nth entry 
+    -- for n <= 10 depending on # of distinct values
+    EXECUTE 'WITH a As (
+              SELECT
+                count(*) cnt
+              FROM
+                (SELECT * FROM unnest($2) e ) x
+              WHERE e is not null
+              GROUP BY e
+              ORDER BY cnt DESC
+            ),
+            b As (
+              SELECT
+                sum(cnt) OVER (ORDER BY cnt DESC) / $1 As cumsum
+              FROM a
+              LIMIT 10
+            )
+            SELECT max(cumsum) maxval FROM b'
+            INTO maxval
+            USING element_count, in_array;
+    IF threshold is null THEN
+        passes = maxval;
+    ELSE
+        passes = CASE WHEN (maxval >= threshold) THEN 1 ELSE 0 END;
+    END IF;
+
+    RETURN passes;
+END;
+$$ language plpgsql IMMUTABLE;
@@ -0,0 +1,37 @@
+--
+-- Calculate the equal interval bins for a given column
+--
+-- @param in_array A numeric array of numbers to determine the best
+--                   to determine the bin boundary
+--
+-- @param breaks The number of bins you want to find.
+--  
+--
+-- Returns: upper edges of bins
+-- 
+--
+
+CREATE OR REPLACE FUNCTION CDB_EqualIntervalBins ( in_array NUMERIC[], breaks INT ) RETURNS NUMERIC[] as $$
+DECLARE 
+    diff numeric;
+    min_val numeric;
+    max_val numeric;
+    tmp_val numeric;
+    i INT := 1;
+    reply numeric[];
+BEGIN
+    SELECT min(e), max(e) INTO min_val, max_val FROM ( SELECT unnest(in_array) e ) x WHERE e IS NOT NULL;
+    diff = (max_val - min_val) / breaks::numeric;
+    LOOP
+        IF i < breaks THEN
+            tmp_val = min_val + i::numeric * diff;
+            reply = array_append(reply, tmp_val);
+            i := i+1;
+        ELSE
+            reply = array_append(reply, max_val);
+            EXIT;
+        END IF;
+    END LOOP;
+    RETURN reply;
+END;
+$$ language plpgsql IMMUTABLE;
@@ -15,18 +15,29 @@ DECLARE
    i INT := 1; 
    reply numeric[]; 
 BEGIN 
-    -- get our unique values
-    SELECT array_agg(e) INTO in_array FROM (SELECT unnest(in_array) e GROUP BY e ORDER BY e ASC) x;
-    -- get the total size of our row
-    element_count := array_upper(in_array, 1) - array_lower(in_array, 1); 
+    -- sort our values
+    SELECT array_agg(e) INTO in_array FROM (SELECT unnest(in_array) e ORDER BY e ASC) x;
+    -- get the total size of our data
+    element_count := array_length(in_array, 1); 
    break_size :=  element_count::numeric / breaks;
    -- slice our bread
    LOOP  
-        IF i > breaks THEN  EXIT;  END IF;  
-        SELECT e INTO tmp_val FROM ( SELECT unnest(in_array) e LIMIT 1 OFFSET round(break_size * i)) x;
+        IF i < breaks THEN
+            IF break_size * i % 1 > 0 THEN
+                SELECT e INTO tmp_val FROM ( SELECT unnest(in_array) e LIMIT 1 OFFSET ceil(break_size * i) - 1) x;
+            ELSE
+                SELECT avg(e) INTO tmp_val FROM ( SELECT unnest(in_array) e LIMIT 2 OFFSET ceil(break_size * i) - 1 ) x;
+            END IF;
+        ELSIF i = breaks THEN
+            -- select the last value
+            SELECT max(e) INTO tmp_val FROM ( SELECT unnest(in_array) e ) x;
+        ELSE
+            EXIT;
+        END IF;
+
        reply = array_append(reply, tmp_val);
-        i := i+1; 
-    END LOOP; 
-    RETURN reply; 
+        i := i+1;
+    END LOOP;
+    RETURN reply;
 END; 
-$$ language plpgsql IMMUTABLE;
+$$ language plpgsql IMMUTABLE;
@@ -2,12 +2,12 @@
 --
 -- Requires PostgreSQL 9.x+
 --
-CREATE OR REPLACE FUNCTION CDB_QueryTables(query text)
-RETURNS name[]
+CREATE OR REPLACE FUNCTION CDB_QueryTablesText(query text)
+RETURNS text[]
 AS $$
 DECLARE
  exp XML;
-  tables NAME[];
+  tables text[];
  rec RECORD;
  rec2 RECORD;
 BEGIN
@@ -41,11 +41,11 @@ BEGIN
          xpath('//x:Relation-Name/text()', exp, ARRAY[ARRAY['x', 'http://www.postgresql.org/2009/explain']]) as x,
          xpath('//x:Relation-Name/../x:Schema/text()', exp, ARRAY[ARRAY['x', 'http://www.postgresql.org/2009/explain']]) as s
      )
-      SELECT unnest(x)::name as p, unnest(s)::name as sc from inp
+      SELECT unnest(x) as p, unnest(s) as sc from inp
    LOOP
      -- RAISE DEBUG 'tab: %', rec2.p;
      -- RAISE DEBUG 'sc: %', rec2.sc;
-      tables := array_append(tables, (rec2.sc || '.' || rec2.p)::name);
+      tables := array_append(tables, (rec2.sc || '.' || rec2.p));
    END LOOP;

    -- RAISE DEBUG 'Tables: %', tables;
@@ -65,3 +65,14 @@ BEGIN
  return tables;
 END
 $$ LANGUAGE 'plpgsql' VOLATILE STRICT;
+
+
+-- Keep CDB_QueryTables with same signature for backwards compatibility.
+-- It should probably be removed in the future.
+CREATE OR REPLACE FUNCTION CDB_QueryTables(query text)
+RETURNS name[]
+AS $$
+BEGIN
+  RETURN CDB_QueryTablesText(query)::name[];
+END
+$$ LANGUAGE 'plpgsql' VOLATILE STRICT;
@@ -1,3 +1,22 @@
+CREATE OR REPLACE FUNCTION cartodb._CDB_total_relation_size(_schema_name TEXT, _table_name TEXT)
+RETURNS bigint AS
+$$
+BEGIN
+  IF EXISTS (
+      SELECT 1 FROM information_schema.tables
+      WHERE table_catalog = current_database()
+        AND table_schema = _schema_name
+        AND table_name = _table_name
+  )
+  THEN
+    RETURN pg_total_relation_size(format('"%s"."%s"', _schema_name, _table_name));
+  ELSE
+    RETURN 0;
+  END IF;
+END;
+$$
+LANGUAGE 'plpgsql' VOLATILE;
+
 -- Return the estimated size of user data. Used for quota checking.
 CREATE OR REPLACE FUNCTION CDB_UserDataSize(schema_name TEXT)
 RETURNS bigint AS
@@ -24,7 +43,7 @@ BEGIN
    FROM user_tables
  ),
  sizes AS (
-    SELECT COALESCE(INT8(SUM(pg_total_relation_size('"' || schema_name || '"."' || table_name || '"')))) table_size,
+    SELECT COALESCE(INT8(SUM(cartodb._CDB_total_relation_size(schema_name, table_name)))) table_size,
      CASE
        WHEN is_overview THEN 0
 	WHEN is_raster THEN 1
@@ -5,35 +5,22 @@
 --
 -- Currently accepted permissions are: 'public', 'private' or 'all'
 --
+DROP FUNCTION IF EXISTS cdb_usertables(text);
 CREATE OR REPLACE FUNCTION CDB_UserTables(perm text DEFAULT 'all')
-RETURNS SETOF information_schema.sql_identifier
+RETURNS SETOF name
 AS $$
-  WITH usertables AS ( 
-    -- TODO: query CDB_TableMetadata for this ?
-    -- See http://github.com/CartoDB/cartodb/issues/254#issuecomment-26044777
-    SELECT table_name as t
-    FROM information_schema.tables
-    WHERE
-         table_type='BASE TABLE'
-     AND table_schema='public'
-     AND table_name NOT IN (
-      'cdb_tablemetadata',
-      'spatial_ref_sys'
-     )
-  ), perms AS (
-    SELECT t, has_table_privilege('public', 'public'||'.'||t, 'SELECT') as p
-    FROM usertables
-  )
-  SELECT t FROM perms
-  WHERE (
-    p = CASE WHEN $1 = 'private' THEN false
-                 WHEN $1 = 'public' THEN true
-                 ELSE not p -- none
-            END
-    OR $1 = 'all'
-  )
-  AND has_table_privilege('public'||'.'||t, 'SELECT')
-   ;
+
+SELECT c.relname 
+FROM pg_class c 
+JOIN pg_namespace n ON n.oid = c.relnamespace
+WHERE c.relkind = 'r' 
+AND c.relname NOT IN ('cdb_tablemetadata', 'spatial_ref_sys')
+AND n.nspname NOT IN ('pg_catalog', 'information_schema', 'topology')
+AND CASE WHEN perm = 'public' THEN has_table_privilege('publicuser', c.oid, 'SELECT')
+         WHEN perm = 'private' THEN has_table_privilege(current_user, c.oid, 'SELECT') AND NOT has_table_privilege('publicuser', c.oid, 'SELECT')
+         WHEN perm = 'all' THEN has_table_privilege(current_user, c.oid, 'SELECT') OR has_table_privilege('publicuser', c.oid, 'SELECT')
+         ELSE false END;
+
 $$ LANGUAGE 'sql';

 -- This is to migrate from pre-0.2.0 version