From 9ae7b9a024be6a97d48a554a76bca08425ca7b5c Mon Sep 17 00:00:00 2001
From: Caetano Sauer <caetanosauer@gmail.com>
Date: Thu, 13 Aug 2026 11:25:33 +0200
Subject: [PATCH] Hyper: load native benchmark data from Parquet

Switch the native Hyper benchmark from the 81 GB CSV input to the 15 GB
single-file Parquet dataset. This reduces the measured load time from 327.0
seconds to 187.1 seconds (43%) while producing the same 99,997,497 rows and
the same logical results for all 43 ClickBench queries.
---
 hyper/benchmark.sh |   2 +-
 hyper/create.sql   | 112 +++++++++++++++++++++++++++++++++++++++++++++
 hyper/load         |  12 +++--
 3 files changed, 120 insertions(+), 6 deletions(-)

diff --git a/hyper/benchmark.sh b/hyper/benchmark.sh
index f61f185e49..a33527ee30 100755
--- a/hyper/benchmark.sh
+++ b/hyper/benchmark.sh
@@ -1,3 +1,3 @@
 #!/bin/bash
-export BENCH_DOWNLOAD_SCRIPT="download-hits-csv"
+export BENCH_DOWNLOAD_SCRIPT="download-hits-parquet-single"
 exec ../lib/benchmark-common.sh
diff --git a/hyper/create.sql b/hyper/create.sql
index 3b376b0624..267a35b710 100644
--- a/hyper/create.sql
+++ b/hyper/create.sql
@@ -1,3 +1,7 @@
+create temp external table hits_parquet
+for 'hits.parquet'
+with (format => 'parquet', binary_as_text => true, immutable => true);
+
 create table hits (
     watchid               bigint       not null,
     javaenable            smallint     not null,
@@ -106,3 +110,111 @@ create table hits (
     clid                  integer      not null,
     assumed primary key (counterid, eventdate, userid, eventtime, watchid)
 );
+
+INSERT INTO hits SELECT
+"WatchID",
+"JavaEnable",
+"Title",
+"GoodEvent",
+to_timestamp("EventTime") AS "EventTime",
+(DATE '1970-01-01' + "EventDate"::integer) AS "EventDate",
+"CounterID",
+"ClientIP",
+"RegionID",
+"UserID",
+"CounterClass",
+"OS",
+"UserAgent",
+"URL",
+"Referer",
+"IsRefresh",
+"RefererCategoryID",
+"RefererRegionID",
+"URLCategoryID",
+"URLRegionID",
+"ResolutionWidth",
+"ResolutionHeight",
+"ResolutionDepth",
+"FlashMajor",
+"FlashMinor",
+"FlashMinor2",
+"NetMajor",
+"NetMinor",
+"UserAgentMajor",
+"UserAgentMinor",
+"CookieEnable",
+"JavascriptEnable",
+"IsMobile",
+"MobilePhone",
+"MobilePhoneModel",
+"Params",
+"IPNetworkID",
+"TraficSourceID",
+"SearchEngineID",
+"SearchPhrase",
+"AdvEngineID",
+"IsArtifical",
+"WindowClientWidth",
+"WindowClientHeight",
+"ClientTimeZone",
+to_timestamp("ClientEventTime") AS "ClientEventTime",
+"SilverlightVersion1",
+"SilverlightVersion2",
+"SilverlightVersion3",
+"SilverlightVersion4",
+"PageCharset",
+"CodeVersion",
+"IsLink",
+"IsDownload",
+"IsNotBounce",
+"FUniqID",
+"OriginalURL",
+"HID",
+"IsOldCounter",
+"IsEvent",
+"IsParameter",
+"DontCountHits",
+"WithHash",
+"HitColor",
+to_timestamp("LocalEventTime") AS "LocalEventTime",
+"Age",
+"Sex",
+"Income",
+"Interests",
+"Robotness",
+"RemoteIP",
+"WindowName",
+"OpenerName",
+"HistoryLength",
+"BrowserLanguage",
+"BrowserCountry",
+"SocialNetwork",
+"SocialAction",
+"HTTPError",
+"SendTiming",
+"DNSTiming",
+"ConnectTiming",
+"ResponseStartTiming",
+"ResponseEndTiming",
+"FetchTiming",
+"SocialSourceNetworkID",
+"SocialSourcePage",
+"ParamPrice",
+"ParamOrderID",
+"ParamCurrency",
+"ParamCurrencyID",
+"OpenstatServiceName",
+"OpenstatCampaignID",
+"OpenstatAdID",
+"OpenstatSourceID",
+"UTMSource",
+"UTMMedium",
+"UTMCampaign",
+"UTMContent",
+"UTMTerm",
+"FromTag",
+"HasGCLID",
+"RefererHash",
+"URLHash",
+"CLID"
+FROM hits_parquet;
diff --git a/hyper/load b/hyper/load
index 9c34e7d839..b3cd219b19 100755
--- a/hyper/load
+++ b/hyper/load
@@ -1,8 +1,8 @@
 #!/bin/bash
-# Create hits.hyper and COPY hits.csv into it, using the PERSISTENT Hyper
+# Create hits.hyper and load hits.parquet into it, using the PERSISTENT Hyper
 # server started by ./start (descriptor in server.endpoint). Loading through
 # the already-running server avoids briefly running two hyperd instances
-# (each of which would try to claim up to 80% of RAM) during the heavy COPY.
+# (each of which would try to claim up to 80% of RAM) during the heavy load.
 set -e
 
 # shellcheck disable=SC1091
@@ -19,9 +19,11 @@ with open("server.endpoint") as f:
 
 endpoint = Endpoint(connection_descriptor=descriptor, user_agent="clickbench")
 with Connection(endpoint, 'hits.hyper', CreateMode.CREATE_AND_REPLACE) as connection:
-    connection.execute_command(open("create.sql").read())
-    connection.execute_command("copy hits from 'hits.csv' with (format csv)")
+    # The Hyper API accepts one statement per execute_command call.
+    for command in open("create.sql").read().split(";"):
+        if command.strip():
+            connection.execute_command(command)
 PY
 
-rm -f hits.csv
+rm -f hits.parquet
 sync