Files
spring-ai/batch-llm/scripts/crash-restart.sh
T

33 lines
1.6 KiB
Bash
Executable File

#!/usr/bin/env bash
# Two real JVMs. JVM 1 is killed with SIGKILL once it has committed KILL_AT chunks; JVM 2 starts the
# same job (same run.id, same H2 file) and finishes it. Nothing is shut down gracefully.
set -euo pipefail
source "$(dirname "$0")/common.sh"
[ -f "$JAR" ] || mvn -B -q -T1 package -DskipTests
CHUNKS=$((TICKETS / CHUNK))
KILL_AT=$(( (CHUNKS * 4 / 10 / 50) * 50 )); [ "$KILL_AT" -ge 50 ] || KILL_AT=50
D=target/crash
rm -rf $D && mkdir -p $D output
OPTS=(--batch.llm.tickets="$TICKETS" --batch.llm.run-id=crash-demo
--spring.datasource.url="jdbc:h2:file:./$D/batchdb;DB_CLOSE_ON_EXIT=FALSE"
--sim.provider.meter-file=$D/provider-meter.bin)
java -jar "$JAR" "${OPTS[@]}" > $D/jvm1.log 2>&1 &
PID=$!
echo "JVM 1 started, pid $PID; waiting for $KILL_AT committed chunks"
while kill -0 "$PID" 2>/dev/null; do
if grep -q "committed-chunks-this-jvm=$KILL_AT " $D/jvm1.log; then
kill -9 "$PID"; echo "kill -9 sent to JVM 1 (pid $PID)"; break
fi
sleep 0.2
done
wait "$PID" 2>/dev/null || true
echo "JVM 1 exit code: $? (137 = SIGKILL)"
echo "--- last lines JVM 1 printed:"; clean_log < $D/jvm1.log | tail -3
echo "--- JVM 2 (same run.id, same database file):"
java -jar "$JAR" "${OPTS[@]}" 2>&1 | clean_log | grep -v '^PROGRESS' | tee $D/jvm2.log
{ echo "# Crash and restart, $TICKETS tickets: JVM 1 killed with kill -9 after $KILL_AT committed chunks, JVM 2 resumes"; echo;
echo "## JVM 1 (progress lines, then SIGKILL)"; clean_log < $D/jvm1.log | tail -4; echo "(process killed by SIGKILL; no shutdown hook ran)"; echo;
echo "## JVM 2"; cat $D/jvm2.log; } > output/09-crash-restart.txt