#!/bin/bash

# Variables (env var names match what gizmosql_client recognizes natively)

export GIZMOSQL_HOST=localhost

export GIZMOSQL_PORT=31337

export GIZMOSQL_USER=clickbench

export GIZMOSQL_PASSWORD=clickbench

# Fixed PID-file path so start/stop/load/query all resolve to the same file

# even though each one sources util.sh in its own subshell.

PID_FILE="${PWD}/gizmosql_server.pid"

# Wait for the server to become reachable. Used by stop after kill, and by

# load before reusing the database.

wait_for_gizmosql() {

while ! nc -z "${GIZMOSQL_HOST}" "${GIZMOSQL_PORT}" 2>/dev/null; do

sleep 1

done

}

# Start the GizmoSQL server. Bounded wait + crash detection: the old code

# spun in `while ! nc -z` forever, so if the spawned server crashed during

# startup (DuckDB lock conflict, missing perms, ...) the bench hung

# permanently — pstree would show ./start stuck on `sleep 1`. Retry the

# spawn a few times if it dies fast (the common case is a stale DuckDB

# fcntl lock from a not-quite-finished sibling thread of the previously

# killed server; sleeping briefly and respawning lets the kernel finish

# releasing the lock).

start_gizmosql() {

local attempt

for attempt in 1 2 3 4 5; do

nohup gizmosql_server \

--username ${GIZMOSQL_USER} \

--database-filename clickbench.db \

--storage-version latest \

--print-queries >> gizmosql_server.log 2>&1 &

local pid=$!

echo "$pid" > "${PID_FILE}"

echo "Waiting for gizmosql_server to start (attempt ${attempt})..."

local i

for i in $(seq 1 60); do

if nc -z ${GIZMOSQL_HOST} ${GIZMOSQL_PORT} 2>/dev/null; then

echo "gizmosql_server is ready (PID: ${pid})"

return 0

fi

if ! kill -0 "$pid" 2>/dev/null; then

echo "gizmosql_server (PID: ${pid}) exited before opening port" >&2

break

fi

sleep 1

done

# Either dead-on-arrival or didn't bind in 60 s. Clean up before

# retrying.

kill "$pid" 2>/dev/null || true

rm -f "${PID_FILE}"

sleep 2

done

echo "gizmosql_server failed to start after 5 attempts" >&2

return 1

}

# Stop the GizmoSQL server. The old code did `kill $pid; wait $pid`, but

# `wait` on a non-child returns immediately, so the function exited

# before the process was actually gone. The next start_gizmosql then

# raced the not-fully-released DuckDB fcntl lock and the new server

# crashed with "Could not set lock on file ...: Conflicting lock is held

# in . (PID <old>)" — visible in gizmosql_server.log around the time

# the bench hangs.

stop_gizmosql() {

if [ -f "${PID_FILE}" ]; then

local pid=$(cat "${PID_FILE}")

if kill -0 "$pid" 2>/dev/null; then

echo "Stopping gizmosql_server (PID: ${pid})..."

kill "$pid"

# Poll until the process is actually gone (kill -0 fails).

local i

for i in $(seq 1 60); do

kill -0 "$pid" 2>/dev/null || break

sleep 1

done

# Still alive after 60 s — escalate.

if kill -0 "$pid" 2>/dev/null; then

kill -KILL "$pid" 2>/dev/null || true

sleep 1

fi

fi

rm -f "${PID_FILE}"

fi

}