[#1] 构建镜像;USER root; 使用supervisor守护进程;端到端本地测试

This commit is contained in:
root
2025-09-15 09:59:49 +00:00
parent 653a0c04a6
commit 31a6ab9803
25 changed files with 559 additions and 209 deletions

View File

@@ -0,0 +1,78 @@
version: "3.8"
services:
es:
build:
context: ../elasticsearch/build
dockerfile: Dockerfile
image: argus-elasticsearch:latest
environment:
- discovery.type=single-node
- xpack.security.enabled=false
- ES_JAVA_OPTS=-Xms512m -Xmx512m
volumes:
- ./private/argus/log/elasticsearch:/private/argus/log/elasticsearch
ports: ["9200:9200"]
healthcheck:
test: ["CMD-SHELL", "curl -fs http://localhost:9200 >/dev/null || exit 1"]
interval: 10s
timeout: 5s
retries: 30
kibana:
build:
context: ../kibana/build
dockerfile: Dockerfile
image: argus-kibana:latest
environment:
- ELASTICSEARCH_HOSTS=http://es:9200
volumes:
- ./private/argus/log/kibana:/private/argus/log/kibana
ports: ["5601:5601"]
depends_on:
es:
condition: service_healthy
fluent-bit-host01:
image: ubuntu:22.04
environment:
- CLUSTER=local
- RACK=dev
- HOSTNAME=host01
- ES_HOST=es
- ES_PORT=9200
volumes:
- ../fluent-bit/start-fluent-bit.sh:/private/start-fluent-bit.sh:ro
- ../fluent-bit/fluent-bit-bundle.tar.gz:/private/fluent-bit-bundle.tar.gz:ro
ports: ["2020:2020"]
depends_on:
es:
condition: service_healthy
command: /private/start-fluent-bit.sh
healthcheck:
test: ["CMD-SHELL", "curl -fs http://localhost:2020/api/v2/metrics >/dev/null || exit 1"]
interval: 15s
timeout: 10s
retries: 30
fluent-bit-host02:
image: ubuntu:22.04
environment:
- CLUSTER=local
- RACK=dev
- HOSTNAME=host02
- ES_HOST=es
- ES_PORT=9200
volumes:
- ../fluent-bit/start-fluent-bit.sh:/private/start-fluent-bit.sh:ro
- ../fluent-bit/fluent-bit-bundle.tar.gz:/private/fluent-bit-bundle.tar.gz:ro
ports: ["2021:2020"]
depends_on:
es:
condition: service_healthy
command: /private/start-fluent-bit.sh
healthcheck:
test: ["CMD-SHELL", "curl -fs http://localhost:2020/api/v2/metrics >/dev/null || exit 1"]
interval: 15s
timeout: 10s
retries: 30

View File

@@ -0,0 +1,27 @@
#!/usr/bin/env bash
set -euo pipefail
root="$(cd "$(dirname "${BASH_SOURCE[0]}")/../" && pwd)"
# 创建新的private目录结构 (基于argus目录结构)
echo "[INFO] Creating private directory structure for supervisor-based containers..."
mkdir -p "$root/private/argus/log/elasticsearch"
mkdir -p "$root/private/argus/log/kibana"
# 设置数据目录权限(ES 和 Kibana 容器都使用 UID 1000)
echo "[INFO] Setting permissions for data directories..."
sudo chown -R 1000:1000 "$root/private/argus/log/elasticsearch" 2>/dev/null || true
sudo chown -R 1000:1000 "$root/private/argus/log/kibana" 2>/dev/null || true
echo "[INFO] Supervisor-based containers will manage their own scripts and configurations"
# 检查fluent-bit相关文件是否存在
if [[ ! -f "$root/../fluent-bit/fluent-bit-bundle.tar.gz" ]]; then
echo "[WARN] fluent-bit/fluent-bit-bundle.tar.gz 不存在,请确保已创建该文件"
fi
if [[ ! -f "$root/../fluent-bit/start-fluent-bit.sh" ]]; then
echo "[WARN] fluent-bit/start-fluent-bit.sh 不存在,请确保已创建该启动脚本"
fi
echo "[OK] 初始化完成: private/argus/log/{elasticsearch,kibana}"
echo "[INFO] Fluent-bit files should be in fluent-bit/ directory"

10
src/log/tests/scripts/02_up.sh Executable file
View File

@@ -0,0 +1,10 @@
#!/usr/bin/env bash
set -euo pipefail
cd "$(dirname "$0")/.."
compose_cmd="docker compose"
if ! $compose_cmd version >/dev/null 2>&1; then
if command -v docker-compose >/dev/null 2>&1; then compose_cmd="docker-compose"; else
echo "需要 Docker Compose,请安装后重试" >&2; exit 1; fi
fi
$compose_cmd -p logging-mvp up -d --remove-orphans
echo "[OK] 服务已启动:ES http://localhost:9200 Kibana http://localhost:5601 Fluent-Bit host01 http://localhost:2020 Fluent-Bit host02 http://localhost:2021"

View File

@@ -0,0 +1,31 @@
#!/usr/bin/env bash
set -euo pipefail
# 获取fluent-bit-host01容器名称
container_name="logging-mvp-fluent-bit-host01-1"
# 检查容器是否存在并运行
if ! docker ps | grep -q "$container_name"; then
echo "[ERROR] Fluent Bit容器 $container_name 未运行"
exit 1
fi
# 创建日志目录
docker exec "$container_name" mkdir -p /logs/train /logs/infer
# 写入训练日志 (host01)
docker exec "$container_name" sh -c "printf '%s INFO [host01] training step=1 loss=1.23 model=bert\n' \"\$(date '+%F %T')\" >> /logs/train/train-demo.log"
docker exec "$container_name" sh -c "printf '%s INFO [host01] training step=2 loss=1.15 model=bert\n' \"\$(date '+%F %T')\" >> /logs/train/train-demo.log"
# 写入推理日志 (host01)
docker exec "$container_name" sh -c "printf '%s ERROR [host01] inference failed on batch=1\n' \"\$(date '+%F %T')\" >> /logs/infer/infer-demo.log"
docker exec "$container_name" sh -c "cat <<'STACK' >> /logs/infer/infer-demo.log
Traceback (most recent call last):
File \"inference.py\", line 15, in <module>
raise RuntimeError(\"CUDA out of memory on host01\")
RuntimeError: CUDA out of memory on host01
STACK"
echo "[OK] 已通过docker exec写入测试日志到 host01 容器内:"
echo " - /logs/train/train-demo.log"
echo " - /logs/infer/infer-demo.log"

View File

@@ -0,0 +1,27 @@
#!/usr/bin/env bash
set -euo pipefail
# 获取fluent-bit-host02容器名称
container_name="logging-mvp-fluent-bit-host02-1"
# 检查容器是否存在并运行
if ! docker ps | grep -q "$container_name"; then
echo "[ERROR] Fluent Bit容器 $container_name 未运行"
exit 1
fi
# 创建日志目录
docker exec "$container_name" mkdir -p /logs/train /logs/infer
# 写入训练日志 (host02)
docker exec "$container_name" sh -c "printf '%s INFO [host02] training step=1 loss=1.45 model=gpt\n' \"\$(date '+%F %T')\" >> /logs/train/train-demo.log"
docker exec "$container_name" sh -c "printf '%s INFO [host02] training step=2 loss=1.38 model=gpt\n' \"\$(date '+%F %T')\" >> /logs/train/train-demo.log"
docker exec "$container_name" sh -c "printf '%s INFO [host02] training step=3 loss=1.32 model=gpt\n' \"\$(date '+%F %T')\" >> /logs/train/train-demo.log"
# 写入推理日志 (host02)
docker exec "$container_name" sh -c "printf '%s WARN [host02] inference slow on batch=5 latency=2.3s\n' \"\$(date '+%F %T')\" >> /logs/infer/infer-demo.log"
docker exec "$container_name" sh -c "printf '%s INFO [host02] inference completed batch=6 latency=0.8s\n' \"\$(date '+%F %T')\" >> /logs/infer/infer-demo.log"
echo "[OK] 已通过docker exec写入测试日志到 host02 容器内:"
echo " - /logs/train/train-demo.log"
echo " - /logs/infer/infer-demo.log"

View File

@@ -0,0 +1,7 @@
#!/usr/bin/env bash
set -euo pipefail
ES="${ES:-http://localhost:9200}"
echo "[i] 查询 ES 端点:$ES"
curl -fsS "$ES/_cat/indices?v" | egrep 'train-|infer-|logstash' || true
printf "train-* 计数:"; curl -fsS "$ES/train-*/_count" | sed -E 's/.*"count":([0-9]+).*/\1/'; echo
printf "infer-* 计数:"; curl -fsS "$ES/infer-*/_count" | sed -E 's/.*"count":([0-9]+).*/\1/'; echo

View File

@@ -0,0 +1,10 @@
#!/usr/bin/env bash
set -euo pipefail
cd "$(dirname "$0")/.."
compose_cmd="docker compose"
if ! $compose_cmd version >/dev/null 2>&1; then
if command -v docker-compose >/dev/null 2>&1; then compose_cmd="docker-compose"; else
echo "需要 Docker Compose,请安装后重试" >&2; exit 1; fi
fi
$compose_cmd -p logging-mvp down
echo "[OK] 已停止所有容器"

168
src/log/tests/scripts/e2e_test.sh Executable file
View File

@@ -0,0 +1,168 @@
#!/usr/bin/env bash
set -euo pipefail
echo "======================================="
echo "ARGUS Log System End-to-End Test"
echo "======================================="
echo ""
# 记录测试开始时间
test_start_time=$(date +%s)
# 函数:获取ES中的日志计数
get_log_count() {
local train_count=$(curl -s "http://localhost:9200/train-*/_count" 2>/dev/null | grep -o '"count":[0-9]*' | cut -d':' -f2 || echo "0")
local infer_count=$(curl -s "http://localhost:9200/infer-*/_count" 2>/dev/null | grep -o '"count":[0-9]*' | cut -d':' -f2 || echo "0")
echo "$((train_count + infer_count))"
}
# 函数:等待服务就绪
wait_for_services() {
echo "[INFO] Waiting for all services to be ready..."
local max_attempts=60
local attempt=1
while [ $attempt -le $max_attempts ]; do
if curl -fs http://localhost:9200/_cluster/health >/dev/null 2>&1 && \
curl -fs http://localhost:5601/api/status >/dev/null 2>&1 && \
curl -fs http://localhost:2020/api/v2/metrics >/dev/null 2>&1 && \
curl -fs http://localhost:2021/api/v2/metrics >/dev/null 2>&1; then
echo "[OK] All services are ready!"
return 0
fi
echo " Waiting for services... ($attempt/$max_attempts)"
sleep 5
((attempt++))
done
echo "[ERROR] Services not ready after $max_attempts attempts"
return 1
}
# 函数:显示测试步骤
show_step() {
echo ""
echo "🔄 Step $1: $2"
echo "----------------------------------------"
}
# 函数:验证步骤结果
verify_step() {
if [ $? -eq 0 ]; then
echo "✅ $1 - SUCCESS"
else
echo "❌ $1 - FAILED"
exit 1
fi
}
# 开始端到端测试
show_step "1" "Bootstrap - Initialize environment"
./scripts/01_bootstrap.sh
verify_step "Bootstrap"
show_step "2" "Startup - Start all services"
./scripts/02_up.sh
verify_step "Service startup"
# 等待服务完全就绪
wait_for_services || exit 1
# 记录发送测试数据前的日志计数
initial_count=$(get_log_count)
echo "[INFO] Initial log count: $initial_count"
show_step "3a" "Send test data - Host01"
./scripts/03_send_test_host01.sh
verify_step "Test data sending (host01)"
show_step "3b" "Send test data - Host02"
./scripts/03_send_test_host02.sh
verify_step "Test data sending (host02)"
# 等待数据被处理
echo "[INFO] Waiting for data to be processed..."
sleep 10
show_step "4" "Verify data - Query Elasticsearch"
./scripts/04_query_es.sh
verify_step "Data verification"
# 记录发送测试数据后的日志计数
final_count=$(get_log_count)
echo "[INFO] Final log count: $final_count"
# 验证日志数量是否增加
if [ "$final_count" -gt "$initial_count" ]; then
added_logs=$((final_count - initial_count))
echo "✅ Log count verification - SUCCESS: Added $added_logs logs (from $initial_count to $final_count)"
else
echo "❌ Log count verification - FAILED: Expected count to increase, but got $initial_count -> $final_count"
exit 1
fi
# 验证预期的最小日志数量(每个主机应该发送一些日志)
expected_min_logs=4 # 至少应该有几条日志
if [ "$final_count" -ge "$expected_min_logs" ]; then
echo "✅ Minimum log threshold - SUCCESS: $final_count logs (>= $expected_min_logs expected)"
else
echo "❌ Minimum log threshold - FAILED: Only $final_count logs (>= $expected_min_logs expected)"
exit 1
fi
# 检查服务健康状态
show_step "Health" "Check service health"
echo "[INFO] Checking service health..."
# 检查 Elasticsearch 健康状态
es_health=$(curl -s "http://localhost:9200/_cluster/health" | grep -o '"status":"[^"]*"' | cut -d'"' -f4)
if [ "$es_health" = "green" ] || [ "$es_health" = "yellow" ]; then
echo "✅ Elasticsearch health: $es_health"
else
echo "❌ Elasticsearch health: $es_health"
fi
# 检查 Kibana 状态
if curl -fs "http://localhost:5601/api/status" >/dev/null 2>&1; then
kb_status="available"
echo "✅ Kibana status: $kb_status"
else
kb_status="unavailable"
echo "⚠️ Kibana status: $kb_status"
fi
# 检查 Fluent-Bit 指标
fb_host01_uptime=$(curl -s "http://localhost:2020/api/v2/metrics" | grep "fluentbit_uptime" | head -1 | grep -o "[0-9]\+$" || echo "0")
fb_host02_uptime=$(curl -s "http://localhost:2021/api/v2/metrics" | grep "fluentbit_uptime" | head -1 | grep -o "[0-9]\+$" || echo "0")
if [ "$fb_host01_uptime" -gt 0 ] && [ "$fb_host02_uptime" -gt 0 ]; then
echo "✅ Fluent-Bit services: host01 uptime=${fb_host01_uptime}s, host02 uptime=${fb_host02_uptime}s"
else
echo "⚠️ Fluent-Bit services: host01 uptime=${fb_host01_uptime}s, host02 uptime=${fb_host02_uptime}s"
fi
verify_step "Service health check"
show_step "5" "Cleanup - Stop all services"
./scripts/05_down.sh
verify_step "Service cleanup"
# 计算总测试时间
test_end_time=$(date +%s)
total_time=$((test_end_time - test_start_time))
echo ""
echo "======================================="
echo "🎉 END-TO-END TEST COMPLETED SUCCESSFULLY!"
echo "======================================="
echo "📊 Test Summary:"
echo " • Initial logs: $initial_count"
echo " • Final logs: $final_count"
echo " • Added logs: $added_logs"
echo " • Total time: ${total_time}s"
echo " • ES health: $es_health"
echo " • Kibana status: $kb_status"
echo " • All services started and stopped successfully"
echo ""
echo "✅ The ARGUS log system is working correctly!"
echo ""