Merge pull request #903 from ClickHouse/clickbench-fixes

Fixes for various systems

Author: alexey-milovidov

arc/install

@@ -26,3 +26,41 @@ if [ ! -f "$DEB_FILE" ]; then
 fi
 
 sudo dpkg -i "$DEB_FILE" || sudo apt-get install -f -y
+
+# Arc 25.12+ links against GLIBC_2.38 (Debian/Ubuntu 24.04+ build
+# environment); on a 22.04 host (glibc 2.35) the daemon dies with
+# "libm.so.6: version `GLIBC_2.38' not found". Pull a newer libc6
+# into a sysroot and wrap /usr/bin/arc so it boots via that loader,
+# instead of bumping the whole base image to noble.
+HAVE_GLIBC=$(ldd --version | head -1 | awk '{print $NF}')
+if dpkg --compare-versions "$HAVE_GLIBC" lt 2.38; then
+    NOBLE_DIR=/opt/glibc-noble
+    if [ ! -x "$NOBLE_DIR/lib64/ld-linux-x86-64.so.2" ]; then
+        # Noble's libc6, statically pinned. Source URL: launchpad mirror.
+        # Plain `pool.ubuntu.com` redirects via HTTPS and works with
+        # curl --location.
+        LIBC_URL="https://launchpad.net/ubuntu/+archive/primary/+files/libc6_2.39-0ubuntu8.6_amd64.deb"
+        sudo mkdir -p "$NOBLE_DIR"
+        tmp=$(mktemp --suffix=.deb)
+        curl -fsSL "$LIBC_URL" -o "$tmp"
+        sudo dpkg-deb -x "$tmp" "$NOBLE_DIR"
+        rm -f "$tmp"
+    fi
+    if [ ! -f /usr/bin/arc.bin ]; then
+        sudo mv /usr/bin/arc /usr/bin/arc.bin
+        # dpkg-deb -x preserves the package's internal paths under
+        # $NOBLE_DIR, i.e. NOBLE_DIR/usr/lib64/ld-linux-x86-64.so.2
+        # (NOT NOBLE_DIR/lib64/...).
+        sudo tee /usr/bin/arc >/dev/null <<EOF
+#!/bin/bash
+# Launch arc.bin via noble's ld-linux + libc, leaving the rest of the
+# system unchanged. --library-path scopes the override to this
+# invocation only.
+exec ${NOBLE_DIR}/usr/lib64/ld-linux-x86-64.so.2 \\
+    --library-path ${NOBLE_DIR}/usr/lib/x86_64-linux-gnu:${NOBLE_DIR}/usr/lib:/usr/lib/x86_64-linux-gnu:/lib/x86_64-linux-gnu \\
+    /usr/bin/arc.bin "\$@"
+EOF
+        sudo chmod +x /usr/bin/arc
+    fi
+    sudo systemctl daemon-reload
+fi

arc/load

@@ -8,13 +8,6 @@ TARGET_FILE="$TARGET_DIR/hits.parquet"
 
 sudo mkdir -p "$TARGET_DIR"
 
-if [ -f "$TARGET_FILE" ] && \
-   [ "$(stat -c%s hits.parquet)" -eq "$(stat -c%s "$TARGET_FILE")" ]; then
-    : # already loaded
-else
-    sudo cp hits.parquet "$TARGET_FILE"
-fi
-
-# Free up local space.
-rm -f hits.parquet
+# Symlink rather than copy — hits.parquet is 14 GB and we read it once.
+sudo ln -sfn "$PWD/hits.parquet" "$TARGET_FILE"
 sync

arc/start

@@ -14,21 +14,38 @@ fi
 sudo systemctl start arc
 
 # Wait for the HTTP endpoint to come up before we try to read the token.
-for _ in $(seq 1 30); do
+for _ in $(seq 1 60); do
     if curl -sf "$ARC_URL/health" >/dev/null 2>&1; then
         break
     fi
     sleep 1
 done
 
 # On first start, Arc prints its admin token to its journal; capture it.
+# The log line has drifted between releases ("Initial admin API token:",
+# "Admin API token:", "API token:", ...) and journald can lag behind
+# /health, so we retry with a broader regex over ~60 s.
 if [ ! -f arc_token.txt ] || \
    ! curl -sf "$ARC_URL/health" -H "x-api-key: $(cat arc_token.txt)" >/dev/null 2>&1; then
-    TOKEN=$(sudo journalctl -u arc --no-pager \
-        | grep -oP '(?:Initial admin API token|Admin API token): \K[^\s]+' \
-        | head -1)
+    TOKEN=""
+    for _ in $(seq 1 60); do
+        sudo journalctl --sync >/dev/null 2>&1 || true
+        JOURNAL=$(sudo journalctl -u arc --no-pager 2>/dev/null || true)
+        TOKEN=$(printf '%s\n' "$JOURNAL" \
+            | grep -oP '(?:[Ii]nitial[[:space:]]+)?[Aa]dmin[[:space:]]+(?:API[[:space:]]+)?[Tt]oken[[:space:]]*[:=][[:space:]]*\K[^[:space:],]+' \
+            | head -1)
+        if [ -z "$TOKEN" ]; then
+            TOKEN=$(printf '%s\n' "$JOURNAL" \
+                | grep -oP '(?:API[[:space:]]+)?[Tt]oken[[:space:]]*[:=][[:space:]]*\K[A-Za-z0-9_.\-]{16,}' \
+                | head -1)
+        fi
+        if [ -n "$TOKEN" ]; then break; fi
+        sleep 1
+    done
     if [ -z "$TOKEN" ]; then
         echo "Error: Could not extract Arc admin API token from journal" >&2
+        echo "---journal tail---" >&2
+        sudo journalctl -u arc --no-pager 2>&1 | tail -50 >&2
         exit 1
     fi
     echo "$TOKEN" > arc_token.txt

byconity/stop

@@ -1,3 +1,7 @@
 #!/bin/bash
+# stop without --volumes — anonymous volumes hold the HDFS namenode
+# data and the bench database. The playground snapshots the VM after
+# ./stop and restores it on every query, so we need the data to outlive
+# the stop.
 
-docker compose down --volumes || true
+docker compose stop || true

cedardb-parquet/start

@@ -5,15 +5,33 @@ if PGPASSWORD=test psql -h localhost -U postgres -c 'SELECT 1' >/dev/null 2>&1;
     exit 0
 fi
 
+# After a VM snapshot+restore, dockerd's in-memory networking/cgroup state
+# is out of sync with the (also-restored) kernel-side resources, and the
+# next `docker run` either fails or starts a container that can't be
+# reached on its mapped port. Restarting dockerd reconciles it.
+sudo systemctl restart docker
+for _ in $(seq 1 30); do
+    sudo docker info >/dev/null 2>&1 && break
+    sleep 1
+done
+
 sudo docker stop cedardb >/dev/null 2>&1 || true
 sudo docker rm cedardb >/dev/null 2>&1 || true
 
-sudo docker run -d --rm -p 5432:5432 \
-    -v "$(pwd)/data:/data" \
-    -v "$(pwd)/db:/var/lib/cedardb/data" \
-    -e CEDAR_PASSWORD=test \
-    --name cedardb cedardb/cedardb:latest >/dev/null
+if ! sudo docker run -d --rm -p 5432:5432 \
+        -v "$(pwd)/data:/data" \
+        -v "$(pwd)/db:/var/lib/cedardb/data" \
+        -e CEDAR_PASSWORD=test \
+        --name cedardb cedardb/cedardb:latest; then
+    echo "docker run failed; ps -a:" >&2
+    sudo docker ps -a >&2 || true
+    exit 1
+fi
 
-until pg_isready -h localhost --dbname postgres -U postgres >/dev/null 2>&1; do
+for _ in $(seq 1 60); do
+    pg_isready -h localhost --dbname postgres -U postgres >/dev/null 2>&1 && exit 0
     sleep 1
 done
+echo "cedardb did not become ready in 60 s; container logs:" >&2
+sudo docker logs cedardb 2>&1 | tail -40 >&2 || true
+exit 1

cedardb/start

@@ -5,16 +5,35 @@ if PGPASSWORD=test psql -h localhost -U postgres -c 'SELECT 1' >/dev/null 2>&1;
     exit 0
 fi
 
+# After a VM snapshot+restore, dockerd's in-memory networking/cgroup state
+# is out of sync with the (also-restored) kernel-side resources, and the
+# next `docker run` either fails or starts a container that can't be
+# reached on its mapped port. Restarting dockerd reconciles it. No-op on
+# initial provision (the daemon was just started).
+sudo systemctl restart docker
+for _ in $(seq 1 30); do
+    sudo docker info >/dev/null 2>&1 && break
+    sleep 1
+done
+
 # `docker run --rm` cleans up container on exit; we run detached.
 sudo docker stop cedardb >/dev/null 2>&1 || true
 sudo docker rm cedardb >/dev/null 2>&1 || true
 
-sudo docker run -d --rm -p 5432:5432 \
-    -v "$(pwd)/data:/data" \
-    -v "$(pwd)/db:/var/lib/cedardb/data" \
-    -e CEDAR_PASSWORD=test \
-    --name cedardb cedardb/cedardb:latest >/dev/null
+if ! sudo docker run -d --rm -p 5432:5432 \
+        -v "$(pwd)/data:/data" \
+        -v "$(pwd)/db:/var/lib/cedardb/data" \
+        -e CEDAR_PASSWORD=test \
+        --name cedardb cedardb/cedardb:latest; then
+    echo "docker run failed; ps -a:" >&2
+    sudo docker ps -a >&2 || true
+    exit 1
+fi
 
-until pg_isready -h localhost --dbname postgres -U postgres >/dev/null 2>&1; do
+for _ in $(seq 1 60); do
+    pg_isready -h localhost --dbname postgres -U postgres >/dev/null 2>&1 && exit 0
     sleep 1
 done
+echo "cedardb did not become ready in 60 s; container logs:" >&2
+sudo docker logs cedardb 2>&1 | tail -40 >&2 || true
+exit 1

chdb-dataframe/.preserve-state

@@ -1,6 +1,6 @@
 #!/bin/bash
-# Reads a SQL query from stdin, dispatches to the running chdb server.
-# Stdout: server response JSON (small).
+# Reads a query from stdin, dispatches to the running in-VM server.
+# Stdout: result (rendered table or scalar from the server).
 # Stderr: query runtime in fractional seconds on the last line.
 # Exit non-zero on error.
 set -e
@@ -19,5 +19,10 @@ if [ "$status" != "200" ]; then
     exit 1
 fi
 
-echo "$body"
-echo "$body" | python3 -c 'import json,sys; print(json.load(sys.stdin)["elapsed"])' >&2
+# Pull `result` for stdout and `elapsed` for stderr (host timing protocol).
+python3 - "$body" <<'PY'
+import json, sys
+d = json.loads(sys.argv[1])
+print(d.get("result", ""))
+sys.stderr.write(str(d["elapsed"]) + "\n")
+PY

chdb-dataframe/query

@@ -57,9 +57,11 @@ async def query(request: Request):
     if not sql:
         raise HTTPException(status_code=400, detail="empty query")
     start = timeit.default_timer()
-    conn.query(sql, "Null")
+    # Pretty so the playground UI shows the actual result table, not
+    # just the timing.
+    res = conn.query(sql, "Pretty")
     elapsed = round(timeit.default_timer() - start, 3)
-    return {"elapsed": elapsed}
+    return {"elapsed": elapsed, "result": str(res)}
 
 
 @app.get("/data-size")