Add handwritten TPC-DS benchmark and sampled-world validation

benchmark/configs/utility/as_tpcds_sf30.json

@@ -0,0 +1,12 @@
+{
+  "sf": 30,
+  "db": "benchmark/results/tpcds_sf30/data.db",
+  "queries_dir": "benchmark/results/tpcds_sf30/queries",
+  "out_dir": "benchmark/results/tpcds_sf30",
+  "schema": "benchmark/tpcds/schema.sql",
+  "threads": 16,
+  "memory_limit": "30GB",
+  "repetitions": 5,
+  "timeout_seconds": 3600,
+  "extension_directory": "benchmark/results/tpcds_extensions"
+}

benchmark/microbench/approx_sum_stability.sql

@@ -34,18 +34,18 @@ WITH bit_positions AS (SELECT unnest(range(64)) as bit_pos),
                    FROM accuracy_test, bit_positions GROUP BY dist, bit_pos),
     exact_lists AS (SELECT dist, list(exact_sum ORDER BY bit_pos) as exact_counters
                     FROM exact_sums GROUP BY dist),
-    approx_sums AS (SELECT dist, as_sum(priv_hash(hash(rowid)), val) as approx_counters
+    approx_sums AS (SELECT dist, list_transform(as_sum(priv_hash(hash(rowid)), val), lambda x: x / 2.0) as approx_counters
                     FROM accuracy_test GROUP BY dist),
     per_counter_stats AS (
         SELECT e.dist, e.exact_counters, a.approx_counters,
-               [POWER(e.exact_counters[x+1] - a.approx_counters[x+1] / 2, 2) FOR x IN range(64)] as sq_diffs,
+               [POWER(e.exact_counters[x+1] - a.approx_counters[x+1], 2) FOR x IN range(64)] as sq_diffs,
                [CASE WHEN e.exact_counters[x+1] = 0 THEN NULL
-                     ELSE 100.0 * (a.approx_counters[x+1] / 2 - e.exact_counters[x+1]) / e.exact_counters[x+1]
+                     ELSE 100.0 * (a.approx_counters[x+1] - e.exact_counters[x+1]) / e.exact_counters[x+1]
                 END FOR x IN range(64)] as pct_diffs
         FROM exact_lists e JOIN approx_sums a ON e.dist = a.dist)
     SELECT
         dist, cast(sqrt(list_avg(sq_diffs)) as bigint) as rmse,
-        ROUND(list_avg(list_transform(pct_diffs, x -> abs(x))), 4) as pct_err,
+        ROUND(list_avg(list_transform(pct_diffs, lambda x: abs(x))), 4) as pct_err,
         ROUND(POWER(rmse, 2) / NULLIF(list_var_samp(approx_counters), 0), 4) as z_squared,
         cast(list_var_samp(exact_counters) as float) as var_exact,
         cast(list_var_samp(approx_counters) as float) as var_approx,

benchmark/tpcds/as_queries/q01.sql

@@ -0,0 +1,23 @@
+-- Handwritten PAC SQL, following the compiled aggregate scopes and scalar output semantics.
+-- The CTE exports customer totals unchanged; its correlated AVG is stochastic across customers.
+WITH customer_total_return AS (SELECT sr_customer_sk AS ctr_customer_sk,
+          sr_store_sk AS ctr_store_sk,
+          sum(sr_return_amt) AS ctr_total_return
+   FROM store_returns,
+        date_dim
+   WHERE sr_returned_date_sk = d_date_sk
+     AND d_year = 2000
+   GROUP BY sr_customer_sk,
+            sr_store_sk)
+SELECT c_customer_id
+FROM customer_total_return ctr1,
+     store,
+     customer
+WHERE priv_filter_gt(CAST(CAST(ctr1.ctr_total_return AS DOUBLE) * 0.8333333333333334 AS FLOAT), (SELECT priv_div(as_sum(priv_hash(hash(ctr2.ctr_customer_sk)), ctr_total_return), as_count(priv_hash(hash(ctr2.ctr_customer_sk)), ctr_total_return))
+     FROM customer_total_return ctr2
+     WHERE ctr1.ctr_store_sk = ctr2.ctr_store_sk))
+  AND s_store_sk = ctr1.ctr_store_sk
+  AND s_state = 'TN'
+  AND ctr1.ctr_customer_sk = c_customer_sk
+ORDER BY c_customer_id
+LIMIT 100;

benchmark/tpcds/as_queries/q03.sql

@@ -0,0 +1,14 @@
+-- Handwritten PAC SQL, following the compiled aggregate scopes and scalar output semantics.
+WITH asa_states AS (SELECT dt.d_year AS d_year, item.i_brand_id AS brand_id, item.i_brand AS brand, as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_ext_sales_price AS DOUBLE)) AS sum_agg FROM date_dim dt,
+     store_sales,
+     item
+WHERE dt.d_date_sk = store_sales.ss_sold_date_sk
+  AND store_sales.ss_item_sk = item.i_item_sk
+  AND item.i_manufact_id = 128
+  AND dt.d_moy=11
+GROUP BY dt.d_year,
+         item.i_brand,
+         item.i_brand_id
+),
+asa_ranked AS (SELECT * FROM asa_states ORDER BY d_year, priv_mean(sum_agg) DESC, brand_id LIMIT 100)
+SELECT d_year, brand_id, brand, CAST(as_noised(sum_agg) AS DECIMAL(38,2)) AS sum_agg FROM asa_ranked ORDER BY d_year, sum_agg DESC, brand_id;

benchmark/tpcds/as_queries/q04.sql

@@ -0,0 +1,120 @@
+-- Handwritten PAC SQL, following the compiled aggregate scopes and scalar output semantics.
+-- Keep year_total scalar inside its CTE; outer customer/year joins retain ordinary PAC scalar semantics.
+WITH year_total AS (SELECT c_customer_id customer_id,
+          c_first_name customer_first_name,
+          c_last_name customer_last_name,
+          c_preferred_cust_flag customer_preferred_cust_flag,
+          c_birth_country customer_birth_country,
+          c_login customer_login,
+          c_email_address customer_email_address,
+          d_year dyear,
+          as_noised_sum(priv_hash(hash(c_customer_sk)), ((ss_ext_list_price-ss_ext_wholesale_cost-ss_ext_discount_amt)+ss_ext_sales_price)/2) year_total,
+          's' sale_type
+   FROM customer,
+        store_sales,
+        date_dim
+   WHERE c_customer_sk = ss_customer_sk
+     AND ss_sold_date_sk = d_date_sk
+   GROUP BY c_customer_id,
+            c_first_name,
+            c_last_name,
+            c_preferred_cust_flag,
+            c_birth_country,
+            c_login,
+            c_email_address,
+            d_year
+   UNION ALL SELECT c_customer_id customer_id,
+                    c_first_name customer_first_name,
+                    c_last_name customer_last_name,
+                    c_preferred_cust_flag customer_preferred_cust_flag,
+                    c_birth_country customer_birth_country,
+                    c_login customer_login,
+                    c_email_address customer_email_address,
+                    d_year dyear,
+                    as_noised_sum(priv_hash(hash(c_customer_sk)), (((cs_ext_list_price-cs_ext_wholesale_cost-cs_ext_discount_amt)+cs_ext_sales_price)/2)) year_total,
+                    'c' sale_type
+   FROM customer,
+        catalog_sales,
+        date_dim
+   WHERE c_customer_sk = cs_bill_customer_sk
+     AND cs_sold_date_sk = d_date_sk
+   GROUP BY c_customer_id,
+            c_first_name,
+            c_last_name,
+            c_preferred_cust_flag,
+            c_birth_country,
+            c_login,
+            c_email_address,
+            d_year
+   UNION ALL SELECT c_customer_id customer_id,
+                    c_first_name customer_first_name,
+                    c_last_name customer_last_name,
+                    c_preferred_cust_flag customer_preferred_cust_flag,
+                    c_birth_country customer_birth_country,
+                    c_login customer_login,
+                    c_email_address customer_email_address,
+                    d_year dyear,
+                    as_noised_sum(priv_hash(hash(c_customer_sk)), (((ws_ext_list_price-ws_ext_wholesale_cost-ws_ext_discount_amt)+ws_ext_sales_price)/2)) year_total,
+                    'w' sale_type
+   FROM customer,
+        web_sales,
+        date_dim
+   WHERE c_customer_sk = ws_bill_customer_sk
+     AND ws_sold_date_sk = d_date_sk
+   GROUP BY c_customer_id,
+            c_first_name,
+            c_last_name,
+            c_preferred_cust_flag,
+            c_birth_country,
+            c_login,
+            c_email_address,
+            d_year)
+SELECT t_s_secyear.customer_id,
+       t_s_secyear.customer_first_name,
+       t_s_secyear.customer_last_name,
+       t_s_secyear.customer_preferred_cust_flag
+FROM year_total t_s_firstyear,
+     year_total t_s_secyear,
+     year_total t_c_firstyear,
+     year_total t_c_secyear,
+     year_total t_w_firstyear,
+     year_total t_w_secyear
+WHERE t_s_secyear.customer_id = t_s_firstyear.customer_id
+  AND t_s_firstyear.customer_id = t_c_secyear.customer_id
+  AND t_s_firstyear.customer_id = t_c_firstyear.customer_id
+  AND t_s_firstyear.customer_id = t_w_firstyear.customer_id
+  AND t_s_firstyear.customer_id = t_w_secyear.customer_id
+  AND t_s_firstyear.sale_type = 's'
+  AND t_c_firstyear.sale_type = 'c'
+  AND t_w_firstyear.sale_type = 'w'
+  AND t_s_secyear.sale_type = 's'
+  AND t_c_secyear.sale_type = 'c'
+  AND t_w_secyear.sale_type = 'w'
+  AND t_s_firstyear.dyear = 2001
+  AND t_s_secyear.dyear = 2001+1
+  AND t_c_firstyear.dyear = 2001
+  AND t_c_secyear.dyear = 2001+1
+  AND t_w_firstyear.dyear = 2001
+  AND t_w_secyear.dyear = 2001+1
+  AND t_s_firstyear.year_total > 0
+  AND t_c_firstyear.year_total > 0
+  AND t_w_firstyear.year_total > 0
+  AND CASE
+          WHEN t_c_firstyear.year_total > 0 THEN t_c_secyear.year_total / t_c_firstyear.year_total
+          ELSE NULL
+      END > CASE
+                WHEN t_s_firstyear.year_total > 0 THEN t_s_secyear.year_total / t_s_firstyear.year_total
+                ELSE NULL
+            END
+  AND CASE
+          WHEN t_c_firstyear.year_total > 0 THEN t_c_secyear.year_total / t_c_firstyear.year_total
+          ELSE NULL
+      END > CASE
+                WHEN t_w_firstyear.year_total > 0 THEN t_w_secyear.year_total / t_w_firstyear.year_total
+                ELSE NULL
+            END
+ORDER BY t_s_secyear.customer_id NULLS FIRST,
+         t_s_secyear.customer_first_name NULLS FIRST,
+         t_s_secyear.customer_last_name NULLS FIRST,
+         t_s_secyear.customer_preferred_cust_flag NULLS FIRST
+LIMIT 100;

benchmark/tpcds/as_queries/q06.sql

@@ -0,0 +1,26 @@
+-- PAC COUNT with scalar HAVING, as in the compiled plan.
+SELECT a.ca_state state,
+       as_noised_count(priv_hash(hash(c.c_customer_sk)), 1::BIGINT) cnt
+FROM customer_address a ,
+     customer c ,
+     store_sales s ,
+     date_dim d ,
+     item i
+WHERE a.ca_address_sk = c.c_current_addr_sk
+  AND c.c_customer_sk = s.ss_customer_sk
+  AND s.ss_sold_date_sk = d.d_date_sk
+  AND s.ss_item_sk = i.i_item_sk
+  AND d.d_month_seq =
+    (SELECT DISTINCT (d_month_seq)
+     FROM date_dim
+     WHERE d_year = 2001
+       AND d_moy = 1 )
+  AND i.i_current_price > 1.2 *
+    (SELECT avg(j.i_current_price)
+     FROM item j
+     WHERE j.i_category = i.i_category)
+GROUP BY a.ca_state
+HAVING as_noised_count(priv_hash(hash(c.c_customer_sk)), 1::BIGINT) >= 10
+ORDER BY cnt NULLS FIRST,
+         a.ca_state NULLS FIRST
+LIMIT 100;

benchmark/tpcds/as_queries/q07.sql

@@ -0,0 +1,24 @@
+-- Handwritten PAC aggregates; automatic query rewriting is disabled by the benchmark.
+SELECT i_item_id,
+       (CAST(as_noised_sum(priv_hash(hash(ss_customer_sk)), ss_quantity) AS DOUBLE) / CAST(as_noised_count(priv_hash(hash(ss_customer_sk)), ss_quantity) AS DOUBLE)) agg1,
+       (CAST(as_noised_sum(priv_hash(hash(ss_customer_sk)), ss_list_price) AS DOUBLE) / CAST(as_noised_count(priv_hash(hash(ss_customer_sk)), ss_list_price) AS DOUBLE)) agg2,
+       (CAST(as_noised_sum(priv_hash(hash(ss_customer_sk)), ss_coupon_amt) AS DOUBLE) / CAST(as_noised_count(priv_hash(hash(ss_customer_sk)), ss_coupon_amt) AS DOUBLE)) agg3,
+       (CAST(as_noised_sum(priv_hash(hash(ss_customer_sk)), ss_sales_price) AS DOUBLE) / CAST(as_noised_count(priv_hash(hash(ss_customer_sk)), ss_sales_price) AS DOUBLE)) agg4
+FROM store_sales,
+     customer_demographics,
+     date_dim,
+     item,
+     promotion
+WHERE ss_sold_date_sk = d_date_sk
+  AND ss_item_sk = i_item_sk
+  AND ss_cdemo_sk = cd_demo_sk
+  AND ss_promo_sk = p_promo_sk
+  AND cd_gender = 'M'
+  AND cd_marital_status = 'S'
+  AND cd_education_status = 'College'
+  AND (p_channel_email = 'N'
+       OR p_channel_event = 'N')
+  AND d_year = 2000
+GROUP BY i_item_id
+ORDER BY i_item_id
+LIMIT 100;

benchmark/tpcds/as_queries/q09.sql

@@ -0,0 +1,8 @@
+-- Handwritten PAC SQL, following the compiled aggregate scopes and scalar output semantics.
+-- Evaluate CASE lane-wise over counter lists and finalize once; all fifteen scalar subqueries remain.
+SELECT CAST(as_noised(list_transform(list_zip((SELECT as_count(priv_hash(hash(ss_customer_sk)),1::BIGINT) FROM store_sales WHERE ss_quantity BETWEEN 1 AND 20), (SELECT priv_div(as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_ext_discount_amt AS DOUBLE)),as_count(priv_hash(hash(ss_customer_sk)),ss_ext_discount_amt)) FROM store_sales WHERE ss_quantity BETWEEN 1 AND 20), (SELECT priv_div(as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_net_paid AS DOUBLE)),as_count(priv_hash(hash(ss_customer_sk)),ss_net_paid)) FROM store_sales WHERE ss_quantity BETWEEN 1 AND 20)), lambda x: CASE WHEN x[1] > 74129 THEN x[2] ELSE x[3] END)) AS DOUBLE) AS bucket1,
+CAST(as_noised(list_transform(list_zip((SELECT as_count(priv_hash(hash(ss_customer_sk)),1::BIGINT) FROM store_sales WHERE ss_quantity BETWEEN 21 AND 40), (SELECT priv_div(as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_ext_discount_amt AS DOUBLE)),as_count(priv_hash(hash(ss_customer_sk)),ss_ext_discount_amt)) FROM store_sales WHERE ss_quantity BETWEEN 21 AND 40), (SELECT priv_div(as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_net_paid AS DOUBLE)),as_count(priv_hash(hash(ss_customer_sk)),ss_net_paid)) FROM store_sales WHERE ss_quantity BETWEEN 21 AND 40)), lambda x: CASE WHEN x[1] > 122840 THEN x[2] ELSE x[3] END)) AS DOUBLE) AS bucket2,
+CAST(as_noised(list_transform(list_zip((SELECT as_count(priv_hash(hash(ss_customer_sk)),1::BIGINT) FROM store_sales WHERE ss_quantity BETWEEN 41 AND 60), (SELECT priv_div(as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_ext_discount_amt AS DOUBLE)),as_count(priv_hash(hash(ss_customer_sk)),ss_ext_discount_amt)) FROM store_sales WHERE ss_quantity BETWEEN 41 AND 60), (SELECT priv_div(as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_net_paid AS DOUBLE)),as_count(priv_hash(hash(ss_customer_sk)),ss_net_paid)) FROM store_sales WHERE ss_quantity BETWEEN 41 AND 60)), lambda x: CASE WHEN x[1] > 56580 THEN x[2] ELSE x[3] END)) AS DOUBLE) AS bucket3,
+CAST(as_noised(list_transform(list_zip((SELECT as_count(priv_hash(hash(ss_customer_sk)),1::BIGINT) FROM store_sales WHERE ss_quantity BETWEEN 61 AND 80), (SELECT priv_div(as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_ext_discount_amt AS DOUBLE)),as_count(priv_hash(hash(ss_customer_sk)),ss_ext_discount_amt)) FROM store_sales WHERE ss_quantity BETWEEN 61 AND 80), (SELECT priv_div(as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_net_paid AS DOUBLE)),as_count(priv_hash(hash(ss_customer_sk)),ss_net_paid)) FROM store_sales WHERE ss_quantity BETWEEN 61 AND 80)), lambda x: CASE WHEN x[1] > 10097 THEN x[2] ELSE x[3] END)) AS DOUBLE) AS bucket4,
+CAST(as_noised(list_transform(list_zip((SELECT as_count(priv_hash(hash(ss_customer_sk)),1::BIGINT) FROM store_sales WHERE ss_quantity BETWEEN 81 AND 100), (SELECT priv_div(as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_ext_discount_amt AS DOUBLE)),as_count(priv_hash(hash(ss_customer_sk)),ss_ext_discount_amt)) FROM store_sales WHERE ss_quantity BETWEEN 81 AND 100), (SELECT priv_div(as_sum(priv_hash(hash(ss_customer_sk)), CAST(ss_net_paid AS DOUBLE)),as_count(priv_hash(hash(ss_customer_sk)),ss_net_paid)) FROM store_sales WHERE ss_quantity BETWEEN 81 AND 100)), lambda x: CASE WHEN x[1] > 165306 THEN x[2] ELSE x[3] END)) AS DOUBLE) AS bucket5
+FROM reason WHERE r_reason_sk=1;

benchmark/tpcds/as_queries/q10.sql

@@ -0,0 +1,66 @@
+-- Handwritten PAC SQL, following the compiled aggregate scopes and scalar output semantics.
+SELECT cd_gender,
+       cd_marital_status,
+       cd_education_status,
+       as_noised_count(priv_hash(hash(c.c_customer_sk)), 1::BIGINT) cnt1,
+       cd_purchase_estimate,
+       as_noised_count(priv_hash(hash(c.c_customer_sk)), 1::BIGINT) cnt2,
+       cd_credit_rating,
+       as_noised_count(priv_hash(hash(c.c_customer_sk)), 1::BIGINT) cnt3,
+       cd_dep_count,
+       as_noised_count(priv_hash(hash(c.c_customer_sk)), 1::BIGINT) cnt4,
+       cd_dep_employed_count,
+       as_noised_count(priv_hash(hash(c.c_customer_sk)), 1::BIGINT) cnt5,
+       cd_dep_college_count,
+       as_noised_count(priv_hash(hash(c.c_customer_sk)), 1::BIGINT) cnt6
+FROM customer c,
+     customer_address ca,
+     customer_demographics
+WHERE c.c_current_addr_sk = ca.ca_address_sk
+  AND ca_county IN ('Rush County',
+                    'Toole County',
+                    'Jefferson County',
+                    'Dona Ana County',
+                    'La Porte County')
+  AND cd_demo_sk = c.c_current_cdemo_sk
+  AND EXISTS
+    (SELECT *
+     FROM store_sales,
+          date_dim
+     WHERE c.c_customer_sk = ss_customer_sk
+       AND ss_sold_date_sk = d_date_sk
+       AND d_year = 2002
+       AND d_moy BETWEEN 1 AND 1+3)
+  AND (EXISTS
+         (SELECT *
+          FROM web_sales,
+               date_dim
+          WHERE c.c_customer_sk = ws_bill_customer_sk
+            AND ws_sold_date_sk = d_date_sk
+            AND d_year = 2002
+            AND d_moy BETWEEN 1 AND 1+3)
+       OR EXISTS
+         (SELECT *
+          FROM catalog_sales,
+               date_dim
+          WHERE c.c_customer_sk = cs_ship_customer_sk
+            AND cs_sold_date_sk = d_date_sk
+            AND d_year = 2002
+            AND d_moy BETWEEN 1 AND 1+3))
+GROUP BY cd_gender,
+         cd_marital_status,
+         cd_education_status,
+         cd_purchase_estimate,
+         cd_credit_rating,
+         cd_dep_count,
+         cd_dep_employed_count,
+         cd_dep_college_count
+ORDER BY cd_gender,
+         cd_marital_status,
+         cd_education_status,
+         cd_purchase_estimate,
+         cd_credit_rating,
+         cd_dep_count,
+         cd_dep_employed_count,
+         cd_dep_college_count
+LIMIT 100;