Firebolt: pull the engine image, add a cleanup script, new results

firebolt/benchmark.sh

@@ -1,6 +1,7 @@
 #!/bin/bash
 # Thin shim — actual flow is in lib/benchmark-common.sh.
-export BENCH_DOWNLOAD_SCRIPT="download-hits-parquet-single"
+export BENCH_DOWNLOAD_SCRIPT="download-hits-parquet-partitioned"
 export BENCH_DURABLE=yes
 export BENCH_RESTARTABLE=no
+./cleanup  # a previous run may have left its container and data behind
 exec ../lib/benchmark-common.sh

firebolt/cleanup

@@ -0,0 +1,10 @@
+#!/bin/bash
+set -eu
+
+# Remove everything a run leaves behind: the container, the engine's data
+# directory and the staged parquet source.
+sudo docker rm -f firebolt-core >/dev/null 2>&1 || true
+# The engine owns its data directory as uid 3473, hence sudo.
+sudo rm -rf fb-volume
+rm -rf data
+rm -f hits_*.parquet

firebolt/create.sql

@@ -111,7 +111,7 @@ CREATE EXTERNAL TABLE hits_external
     "CLID" INTEGER NOT NULL
 )
 URL = 'file:///firebolt-core/clickbench'
-OBJECT_PATTERN = 'hits.parquet'
+OBJECT_PATTERN = 'hits_*.parquet'
 TYPE = PARQUET;
 
 CREATE TABLE hits

firebolt/data-size

@@ -1,6 +1,7 @@
 #!/bin/bash
 set -e
 
-# Firebolt-core writes its database state under /firebolt-core/volume
-# inside the container, which we bind-mount to ./fb-volume on the host.
-du -bcs fb-volume 2>/dev/null | awk '/total$/ { print $1 }'
+# The engine writes its database state under /var/lib/firebolt inside the
+# container, which we bind-mount to ./fb-volume on the host. It owns that
+# directory as uid 3473 mode 2770, so du without sudo reports 0.
+sudo du -bcs fb-volume | awk '/total$/ { print $1 }'

firebolt/install

@@ -1,6 +1,9 @@
 #!/bin/bash
 set -eu
 
-sudo apt-get update -y
-sudo apt-get install -y docker.io jq
-sudo docker pull ghcr.io/firebolt-db/firebolt-core:preview-rc
+# Installing docker.io over a docker that came from elsewhere breaks it.
+if ! command -v docker >/dev/null || ! command -v jq >/dev/null; then
+    sudo apt-get update -y
+    sudo apt-get install -y docker.io jq
+fi
+sudo docker pull ghcr.io/firebolt-db/engine:dev

firebolt/load

@@ -1,12 +1,13 @@
 #!/bin/bash
 set -eu
 
-# Stage hits.parquet where the container can see it (./data is
+# Stage the parquet source where the container can see it (./data is
 # bind-mounted as /firebolt-core/clickbench).
 mkdir -p data
-if [ -f hits.parquet ]; then
-    mv -f hits.parquet data/hits.parquet
-fi
+for f in hits_*.parquet; do
+    [ -e "$f" ] || break
+    mv -f "$f" data/
+done
 
 run_sql() {
     local url="$1" data="$2" body_file http_code
@@ -31,13 +32,13 @@ run_sql() {
     fi
 }
 
-# create.sql CREATEs hits_external pointing at the parquet file, then
+# create.sql CREATEs hits_external pointing at the parquet files, then
 # INSERTs into the managed `hits` table — the ingested-to-Firebolt
 # variant of the benchmark.
 run_sql 'http://localhost:3473/?enable_multi_query_requests=true' \
     'DROP DATABASE IF EXISTS clickbench;CREATE DATABASE clickbench;'
 run_sql 'http://localhost:3473/?database=clickbench&enable_multi_query_requests=true' \
     @create.sql
 
-rm -f data/hits.parquet
+rm -f data/hits_*.parquet
 sync

firebolt/results/20260819/c8g.metal-48xl.json

@@ -0,0 +1,59 @@
+{
+    "system": "Firebolt",
+    "date": "2026-08-19",
+    "machine": "c8g.metal-48xl",
+    "cluster_size": 1,
+    "proprietary": "yes",
+    "hardware": "cpu",
+    "tuned": "no",
+    "tags": ["C++","column-oriented","PostgreSQL compatible","ClickHouse derivative"],
+    "load_time": 170,
+    "data_size": 15829362967,
+    "concurrent_qps": 42.822,
+    "concurrent_error_ratio": 0,
+    "result": [
+        [0.008, 0.002, 0.002],
+        [0.4, 0.168, 0.114],
+        [0.175, 0.033, 0.025],
+        [0.163, 0.015, 0.015],
+        [0.052, 0.046, 0.049],
+        [1.258, 0.066, 0.063],
+        [0.071, 0.013, 0.013],
+        [0.051, 0.106, 0.112],
+        [0.562, 0.088, 0.103],
+        [0.109, 0.08, 0.09],
+        [0.069, 0.041, 0.039],
+        [0.192, 0.065, 0.021],
+        [0.048, 0.043, 0.041],
+        [0.079, 0.1, 0.066],
+        [0.081, 0.049, 0.049],
+        [0.046, 0.046, 0.045],
+        [0.095, 0.085, 0.085],
+        [0.071, 0.072, 0.07],
+        [4.894, 0.146, 0.103],
+        [0.019, 0.007, 0.007],
+        [11.688, 0.026, 0.013],
+        [0.016, 0.011, 0.011],
+        [10.167, 0.034, 0.021],
+        [0.217, 0.066, 0.066],
+        [0.019, 0.018, 0.017],
+        [0.016, 0.015, 0.016],
+        [0.018, 0.018, 0.017],
+        [0.072, 0.063, 0.058],
+        [9.543, 0.181, 0.121],
+        [0.029, 0.026, 0.026],
+        [0.654, 0.038, 0.04],
+        [4.92, 0.078, 0.048],
+        [0.205, 0.184, 0.157],
+        [0.176, 0.16, 0.159],
+        [0.157, 0.148, 0.154],
+        [0.052, 0.056, 0.05],
+        [0.049, 0.025, 0.02],
+        [0.026, 0.018, 0.016],
+        [0.031, 0.011, 0.01],
+        [0.099, 0.046, 0.043],
+        [0.046, 0.012, 0.011],
+        [0.042, 0.014, 0.01],
+        [0.016, 0.012, 0.014]
+]
+}

firebolt/start

@@ -8,12 +8,10 @@ if curl -sS --max-time 5 'http://localhost:3473/' \
 fi
 
 mkdir -p data fb-volume
-# firebolt-core runs as UID/GID 1111 inside the container and refuses
-# to start if its data dir is not writeable by that uid (the engine
-# self-checks and aborts with "directory ... is not readable or
-# writeable by the Firebolt Core process"). Set the host-side
-# ownership accordingly so the bind-mounted dir is usable.
-sudo chown 1111:1111 fb-volume
+# The engine runs as UID/GID 3473 inside the container and refuses to
+# start if its data dir is not writeable by that uid. World-writeable is
+# what the official installer at get.firebolt.io does.
+sudo chmod 0777 fb-volume
 
 # If the container exists (stopped from a prior agent pre-snapshot
 # cycle), just start it back — the data lives on the bind-mounted
@@ -22,20 +20,20 @@ sudo chown 1111:1111 fb-volume
 if sudo docker ps -a --format '{{.Names}}' | grep -qx firebolt-core; then
     sudo docker start firebolt-core >/dev/null
 else
-    # `firebolt-core` is the public self-hosted image. Container needs
+    # `engine` is the public self-hosted image. Container needs
     # memlock 8 GiB and seccomp unconfined per upstream's run docs.
     # /firebolt-core/clickbench: parquet source (read at load time).
-    # /firebolt-core/volume:     engine data directory (must persist
+    # /var/lib/firebolt:         engine data directory (must persist
     #                            across the agent's pre-snapshot
     #                            stop+start cycle or the snapshot
     #                            ships an empty DB).
     sudo docker run -dit --name firebolt-core \
         --ulimit memlock=8589934592:8589934592 \
         --security-opt seccomp=unconfined \
         -p 127.0.0.1:3473:3473 \
-        -v "$(pwd)/fb-volume:/firebolt-core/volume" \
+        -v "$(pwd)/fb-volume:/var/lib/firebolt" \
         -v "$(pwd)/data:/firebolt-core/clickbench" \
-        ghcr.io/firebolt-db/firebolt-core:preview-rc >/dev/null
+        "${FB_IMAGE:-ghcr.io/firebolt-db/engine:dev}" >/dev/null
 fi
 
 # Wait for the cluster to be "actually" ready. firebolt-core's HTTP