Welcome to the fourth installment of our series on implementing a sophisticated order processing system! In our previous posts, we laid the foundation for our project, explored advanced Temporal workflows, and delved into advanced database operations. Today, we’re focusing on an equally crucial aspect of any production-ready system: monitoring and alerting.
In a microservices architecture, especially one handling complex processes like order management, effective monitoring and alerting are crucial. They allow us to:
Prometheus is an open-source systems monitoring and alerting toolkit. It’s become a standard in the cloud-native world due to its powerful features and extensive ecosystem. Key components include:
We’ll also be using Grafana, a popular open-source platform for monitoring and observability, to create dashboards and visualize our Prometheus data.
By the end of this post, you’ll be able to:
Let’s dive in!
Before we start implementing, let’s review some key concepts that will be crucial for our monitoring and alerting setup.
Observability refers to the ability to understand the internal state of a system by examining its outputs. In distributed systems like our order processing system, observability typically encompasses three main pillars:
In this post, we’ll focus primarily on metrics, though we’ll touch on how these can be integrated with logs and traces.
Prometheus follows a pull-based architecture:
Prometheus offers four core metric types:
PromQL (Prometheus Query Language) は、Prometheus データをクエリするための強力な関数型言語です。時系列データをリアルタイムで選択して集計できます。主な機能は次のとおりです:
ダッシュボードとアラートを構築するときに、PromQL クエリの例を見ていきます。
Grafana は、マルチプラットフォームのオープンソース分析およびインタラクティブな視覚化 Web アプリケーションです。 Prometheus もその 1 つである、サポートされているデータ ソースに接続すると、Web にチャート、グラフ、アラートを提供します。主な機能は次のとおりです:
これらの概念を説明したので、監視および警告システムの実装を開始しましょう。
注文処理システムを監視するために Prometheus を設定することから始めましょう。
まず、Prometheus を docker-compose.yml ファイルに追加しましょう。
services: # ... other services ... prometheus: image: prom/prometheus:v2.30.3 volumes: - ./prometheus:/etc/prometheus - prometheus_data:/prometheus command: - '--config.file=/etc/prometheus/prometheus.yml' - '--storage.tsdb.path=/prometheus' - '--web.console.libraries=/usr/share/prometheus/console_libraries' - '--web.console.templates=/usr/share/prometheus/consoles' ports: - 9090:9090 volumes: # ... other volumes ... prometheus_data: {}
次に、./prometheus ディレクトリに prometheus.yml ファイルを作成します。
global: scrape_interval: 15s evaluation_interval: 15s scrape_configs: - job_name: 'prometheus' static_configs: - targets: ['localhost:9090'] - job_name: 'order_processing_api' static_configs: - targets: ['order_processing_api:8080'] - job_name: 'postgres' static_configs: - targets: ['postgres_exporter:9187']
この構成は、Prometheus に、Prometheus 自体、注文処理 API、および Postgres エクスポーター (後でセットアップします) からメトリクスを収集するように指示します。
Go サービスからのメトリクスを公開するには、Prometheus クライアント ライブラリを使用します。まず、go.mod に追加します:
go get github.com/prometheus/client_golang
次に、メインの Go ファイルを変更してメトリクスを公開しましょう:
package main import ( "net/http" "github.com/gin-gonic/gin" "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promhttp" ) var ( httpRequestsTotal = prometheus.NewCounterVec( prometheus.CounterOpts{ Name: "http_requests_total", Help: "Total number of HTTP requests", }, []string{"method", "endpoint", "status"}, ) httpRequestDuration = prometheus.NewHistogramVec( prometheus.HistogramOpts{ Name: "http_request_duration_seconds", Help: "Duration of HTTP requests in seconds", Buckets: prometheus.DefBuckets, }, []string{"method", "endpoint"}, ) ) func init() { prometheus.MustRegister(httpRequestsTotal) prometheus.MustRegister(httpRequestDuration) } func main() { r := gin.Default() // Middleware to record metrics r.Use(func(c *gin.Context) { timer := prometheus.NewTimer(httpRequestDuration.WithLabelValues(c.Request.Method, c.FullPath())) c.Next() timer.ObserveDuration() httpRequestsTotal.WithLabelValues(c.Request.Method, c.FullPath(), string(c.Writer.Status())).Inc() }) // Expose metrics endpoint r.GET("/metrics", gin.WrapH(promhttp.Handler())) // ... rest of your routes ... r.Run(":8080") }
このコードは 2 つのメトリクスを設定します:
より動的な環境のために、Prometheus はさまざまなサービス検出メカニズムをサポートしています。たとえば、Kubernetes 上で実行している場合は、Kubernetes SD 構成を使用できます。
scrape_configs: - job_name: 'kubernetes-pods' kubernetes_sd_configs: - role: pod relabel_configs: - source_labels: [__meta_kubernetes_pod_annotation_prometheus_io_scrape] action: keep regex: true - source_labels: [__meta_kubernetes_pod_annotation_prometheus_io_path] action: replace target_label: __metrics_path__ regex: (.+)
この構成は、適切なアノテーションを持つポッドからメトリクスを自動的に検出して収集します。
Prometheus は、ローカル ファイル システム上の時系列データベースにデータを保存します。 Prometheus 構成で保持時間とストレージ サイズを構成できます:
global: scrape_interval: 15s evaluation_interval: 15s storage: tsdb: retention.time: 15d retention.size: 50GB # ... rest of the configuration ...
この構成では、保存期間が 15 日間、最大ストレージ サイズが 50 GB に設定されます。
次のセクションでは、注文処理システムのカスタム指標の定義と実装について詳しく説明します。
Prometheus がセットアップされ、基本的な HTTP メトリクスが実装されたので、注文処理システムに固有のカスタム メトリクスを定義して実装しましょう。
メトリクスを設計するときは、システムからどのような洞察を得たいかを考えることが重要です。弊社の注文処理システムでは、以下を追跡する必要がある場合があります:
これらのメトリクスを実装しましょう:
package metrics import ( "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promauto" ) var ( OrdersCreated = promauto.NewCounter(prometheus.CounterOpts{ Name: "orders_created_total", Help: "The total number of created orders", }) OrderProcessingTime = promauto.NewHistogram(prometheus.HistogramOpts{ Name: "order_processing_seconds", Help: "Time taken to process an order", Buckets: prometheus.LinearBuckets(0, 30, 10), // 0-300 seconds, 30-second buckets }) OrderStatusGauge = promauto.NewGaugeVec(prometheus.GaugeOpts{ Name: "orders_by_status", Help: "Number of orders by status", }, []string{"status"}) PaymentProcessed = promauto.NewCounterVec(prometheus.CounterOpts{ Name: "payments_processed_total", Help: "The total number of processed payments", }, []string{"status"}) InventoryUpdates = promauto.NewCounter(prometheus.CounterOpts{ Name: "inventory_updates_total", Help: "The total number of inventory updates", }) ShippingArrangementTime = promauto.NewHistogram(prometheus.HistogramOpts{ Name: "shipping_arrangement_seconds", Help: "Time taken to arrange shipping", Buckets: prometheus.LinearBuckets(0, 60, 5), // 0-300 seconds, 60-second buckets }) )
メトリクスを定義したので、サービスに実装してみましょう:
package main import ( "time" "github.com/yourusername/order-processing-system/metrics" ) func createOrder(order Order) error { startTime := time.Now() // Order creation logic... metrics.OrdersCreated.Inc() metrics.OrderProcessingTime.Observe(time.Since(startTime).Seconds()) metrics.OrderStatusGauge.WithLabelValues("pending").Inc() return nil } func processPayment(payment Payment) error { // Payment processing logic... if paymentSuccessful { metrics.PaymentProcessed.WithLabelValues("success").Inc() } else { metrics.PaymentProcessed.WithLabelValues("failure").Inc() } return nil } func updateInventory(item Item) error { // Inventory update logic... metrics.InventoryUpdates.Inc() return nil } func arrangeShipping(order Order) error { startTime := time.Now() // Shipping arrangement logic... metrics.ShippingArrangementTime.Observe(time.Since(startTime).Seconds()) return nil }
メトリクスに名前を付けてラベルを付けるときは、次のベスト プラクティスを考慮してください。
For API endpoints, we’ve already implemented basic instrumentation. For database operations, we can add metrics like this:
func (s *Store) GetOrder(ctx context.Context, id int64) (Order, error) { startTime := time.Now() defer func() { metrics.DBOperationDuration.WithLabelValues("GetOrder").Observe(time.Since(startTime).Seconds()) }() // Existing GetOrder logic... }
For Temporal workflows, we can add metrics in our activity implementations:
func ProcessOrderActivity(ctx context.Context, order Order) error { startTime := time.Now() defer func() { metrics.WorkflowActivityDuration.WithLabelValues("ProcessOrder").Observe(time.Since(startTime).Seconds()) }() // Existing ProcessOrder logic... }
Now that we have our metrics set up, let’s visualize them using Grafana.
First, let’s add Grafana to our docker-compose.yml:
services: # ... other services ... grafana: image: grafana/grafana:8.2.2 ports: - 3000:3000 volumes: - grafana_data:/var/lib/grafana volumes: # ... other volumes ... grafana_data: {}
Let’s create a dashboard for our order processing system:
For our first panel, let’s create a graph of order creation rate:
Let’s add another panel for order processing time:
For order status distribution:
Continue adding panels for other metrics we’ve defined.
Grafana allows us to create variables that can be used across the dashboard. Let’s create a variable for time range:
Now we can use this in our queries like this: rate(orders_created_total[$time_range])
In the next section, we’ll set up alerting rules to notify us of potential issues in our system.
Now that we have our metrics and dashboards set up, let’s implement alerting to proactively notify us of potential issues in our system.
When designing alerts, consider the following principles:
For our order processing system, we might want to alert on:
Let’s create an alerts.yml file in our Prometheus configuration directory:
groups: - name: order_processing_alerts rules: - alert: HighOrderProcessingErrorRate expr: rate(order_processing_errors_total[5m]) / rate(orders_created_total[5m]) > 0.05 for: 5m labels: severity: critical annotations: summary: High order processing error rate description: "Error rate is over the last 5 minutes" - alert: SlowOrderProcessing expr: histogram_quantile(0.95, rate(order_processing_seconds_bucket[5m])) > 300 for: 10m labels: severity: warning annotations: summary: Slow order processing description: "95th percentile of order processing time is over the last 5 minutes" - alert: UnusualOrderRate expr: abs(rate(orders_created_total[1h]) - rate(orders_created_total[1h] offset 1d)) > (rate(orders_created_total[1h] offset 1d) * 0.3) for: 30m labels: severity: warning annotations: summary: Unusual order creation rate description: "Order creation rate has changed by more than 30% compared to the same time yesterday" - alert: LowInventory expr: inventory_level < 10 for: 5m labels: severity: warning annotations: summary: Low inventory level description: "Inventory level for is " - alert: HighPaymentFailureRate expr: rate(payments_processed_total{status="failure"}[15m]) / rate(payments_processed_total[15m]) > 0.1 for: 15m labels: severity: critical annotations: summary: High payment failure rate description: "Payment failure rate is over the last 15 minutes"
Update your prometheus.yml to include this alerts file:
rule_files: - "alerts.yml"
Now, let’s set up Alertmanager to handle our alerts. Add Alertmanager to your docker-compose.yml:
services: # ... other services ... alertmanager: image: prom/alertmanager:v0.23.0 ports: - 9093:9093 volumes: - ./alertmanager:/etc/alertmanager command: - '--config.file=/etc/alertmanager/alertmanager.yml'
Create an alertmanager.yml in the ./alertmanager directory:
route: group_by: ['alertname'] group_wait: 30s group_interval: 5m repeat_interval: 1h receiver: 'email-notifications' receivers: - name: 'email-notifications' email_configs: - to: 'team@example.com' from: 'alertmanager@example.com' smarthost: 'smtp.example.com:587' auth_username: 'alertmanager@example.com' auth_identity: 'alertmanager@example.com' auth_password: 'password'
Update your prometheus.yml to point to Alertmanager:
alerting: alertmanagers: - static_configs: - targets: - alertmanager:9093
In the Alertmanager configuration above, we’ve set up email notifications. You can also configure other channels like Slack, PagerDuty, or custom webhooks.
In our alerts, we’ve used severity labels. We can use these in Alertmanager to implement different routing or notification strategies based on severity:
route: group_by: ['alertname'] group_wait: 30s group_interval: 5m repeat_interval: 1h receiver: 'email-notifications' routes: - match: severity: critical receiver: 'pagerduty-critical' - match: severity: warning receiver: 'slack-warnings' receivers: - name: 'email-notifications' email_configs: - to: 'team@example.com' - name: 'pagerduty-critical' pagerduty_configs: - service_key: '<your-pagerduty-service-key>' - name: 'slack-warnings' slack_configs: - api_url: '<your-slack-webhook-url>' channel: '#alerts'
Monitoring database performance is crucial for maintaining a responsive and reliable system. Let’s set up monitoring for our PostgreSQL database.
First, add the Postgres exporter to your docker-compose.yml:
services: # ... other services ... postgres_exporter: image: wrouesnel/postgres_exporter:latest environment: DATA_SOURCE_NAME: "postgresql://user:password@postgres:5432/dbname?sslmode=disable" ports: - 9187:9187
Make sure to replace user, password, and dbname with your actual PostgreSQL credentials.
Some important PostgreSQL metrics to monitor include:
Let’s create a new dashboard for database performance:
Let’s add some database-specific alerts to our alerts.yml:
- alert: HighDatabaseConnections expr: pg_stat_activity_count > 100 for: 5m labels: severity: warning annotations: summary: High number of database connections description: "There are active database connections" - alert: LowCacheHitRatio expr: pg_stat_database_blks_hit / (pg_stat_database_blks_hit + pg_stat_database_blks_read) < 0.9 for: 15m labels: severity: warning annotations: summary: Low database cache hit ratio description: "Cache hit ratio is "
Monitoring Temporal workflows is essential for ensuring the reliability and performance of our order processing system.
Temporal provides a metrics client that we can use to expose metrics to Prometheus. Let’s update our Temporal worker to include metrics:
import ( "go.temporal.io/sdk/client" "go.temporal.io/sdk/worker" "go.temporal.io/sdk/contrib/prometheus" ) func main() { // ... other setup ... // Create Prometheus metrics handler metricsHandler := prometheus.NewPrometheusMetricsHandler() // Create Temporal client with metrics c, err := client.NewClient(client.Options{ MetricsHandler: metricsHandler, }) if err != nil { log.Fatalln("Unable to create Temporal client", err) } defer c.Close() // Create worker with metrics w := worker.New(c, "order-processing-task-queue", worker.Options{ MetricsHandler: metricsHandler, }) // ... register workflows and activities ... // Run the worker err = w.Run(worker.InterruptCh()) if err != nil { log.Fatalln("Unable to start worker", err) } }
Important Temporal metrics to monitor include:
Let’s create a dashboard for Temporal workflows:
Let’s add some Temporal-specific alerts to our alerts.yml:
- alert: HighWorkflowFailureRate expr: rate(temporal_workflow_failed_total[15m]) / rate(temporal_workflow_completed_total[15m]) > 0.05 for: 15m labels: severity: critical annotations: summary: High workflow failure rate description: "Workflow failure rate is over the last 15 minutes" - alert: LongRunningWorkflow expr: histogram_quantile(0.95, rate(temporal_workflow_execution_time_bucket[1h])) > 3600 for: 30m labels: severity: warning annotations: summary: Long-running workflows detected description: "95th percentile of workflow execution time is over 1 hour"
These alerts will help you detect issues with your Temporal workflows, such as high failure rates or unexpectedly long-running workflows.
In the next sections, we’ll cover some advanced Prometheus techniques and discuss testing and validation of our monitoring setup.
As our monitoring system grows more complex, we can leverage some advanced Prometheus techniques to improve its efficiency and capabilities.
Recording rules allow you to precompute frequently needed or computationally expensive expressions and save their result as a new set of time series. This can significantly speed up the evaluation of dashboards and alerts.
Let’s add some recording rules to our Prometheus configuration. Create a rules.yml file:
groups: - name: example_recording_rules interval: 5m rules: - record: job:order_processing_rate:5m expr: rate(orders_created_total[5m]) - record: job:order_processing_error_rate:5m expr: rate(order_processing_errors_total[5m]) / rate(orders_created_total[5m]) - record: job:payment_success_rate:5m expr: rate(payments_processed_total{status="success"}[5m]) / rate(payments_processed_total[5m])
Add this file to your Prometheus configuration:
rule_files: - "alerts.yml" - "rules.yml"
Now you can use these precomputed metrics in your dashboards and alerts, which can be especially helpful for complex queries that you use frequently.
The Pushgateway allows you to push metrics from jobs that can’t be scraped, such as batch jobs or serverless functions. Let’s add a Pushgateway to our docker-compose.yml:
services: # ... other services ... pushgateway: image: prom/pushgateway ports: - 9091:9091
Now, you can push metrics to the Pushgateway from your batch jobs or short-lived processes. Here’s an example using the Go client:
import ( "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/push" ) func runBatchJob() { // Define a counter for the batch job batchJobCounter := prometheus.NewCounter(prometheus.CounterOpts{ Name: "batch_job_processed_total", Help: "Total number of items processed by the batch job", }) // Run your batch job and update the counter // ... // Push the metric to the Pushgateway pusher := push.New("http://pushgateway:9091", "batch_job") pusher.Collector(batchJobCounter) if err := pusher.Push(); err != nil { log.Printf("Could not push to Pushgateway: %v", err) } }
Don’t forget to add the Pushgateway as a target in your Prometheus configuration:
scrape_configs: # ... other configs ... - job_name: 'pushgateway' static_configs: - targets: ['pushgateway:9091']
For large-scale systems, you might need to set up Prometheus federation, where one Prometheus server scrapes data from other Prometheus servers. This allows you to aggregate metrics from multiple Prometheus instances.
Here’s an example configuration for a federated Prometheus setup:
scrape_configs: - job_name: 'federate' scrape_interval: 15s honor_labels: true metrics_path: '/federate' params: 'match[]': - '{job="order_processing_api"}' - '{job="postgres_exporter"}' static_configs: - targets: - 'prometheus-1:9090' - 'prometheus-2:9090'
This configuration allows a higher-level Prometheus server to scrape specific metrics from other Prometheus servers.
Exemplars allow you to link metrics to trace data, providing a way to drill down from a high-level metric to a specific trace. This is particularly useful when integrating Prometheus with distributed tracing systems like Jaeger or Zipkin.
To use exemplars, you need to enable them in your Prometheus configuration:
global: scrape_interval: 15s evaluation_interval: 15s exemplar_storage: enable: true
Then, when instrumenting your code, you can add exemplars to your metrics:
import ( "github.com/prometheus/client_golang/prometheus" "github.com/prometheus/client_golang/prometheus/promauto" ) var ( orderProcessingDuration = promauto.NewHistogramVec( prometheus.HistogramOpts{ Name: "order_processing_duration_seconds", Help: "Duration of order processing in seconds", Buckets: prometheus.DefBuckets, }, []string{"status"}, ) ) func processOrder(order Order) { start := time.Now() // Process the order... duration := time.Since(start) orderProcessingDuration.WithLabelValues(order.Status).Observe(duration.Seconds(), prometheus.Labels{ "traceID": getCurrentTraceID(), }, ) }
This allows you to link from a spike in order processing duration directly to the trace of a slow order, greatly aiding in debugging and performance analysis.
Ensuring the reliability of your monitoring system is crucial. Let’s explore some strategies for testing and validating our Prometheus setup.
When unit testing your Go code, you can use the prometheus/testutil package to verify that your metrics are being updated correctly:
import ( "testing" "github.com/prometheus/client_golang/prometheus/testutil" ) func TestOrderProcessing(t *testing.T) { // Process an order processOrder(Order{ID: 1, Status: "completed"}) // Check if the metric was updated expected := ` # HELP order_processing_duration_seconds Duration of order processing in seconds # TYPE order_processing_duration_seconds histogram order_processing_duration_seconds_bucket{status="completed",le="0.005"} 1 order_processing_duration_seconds_bucket{status="completed",le="0.01"} 1 # ... other buckets ... order_processing_duration_seconds_sum{status="completed"} 0.001 order_processing_duration_seconds_count{status="completed"} 1 ` if err := testutil.CollectAndCompare(orderProcessingDuration, strings.NewReader(expected)); err != nil { t.Errorf("unexpected collecting result:\n%s", err) } }
To test that Prometheus is correctly scraping your metrics, you can set up an integration test that starts your application, waits for Prometheus to scrape it, and then queries Prometheus to verify the metrics:
func TestPrometheusIntegration(t *testing.T) { // Start your application go startApp() // Wait for Prometheus to scrape (adjust the sleep time as needed) time.Sleep(30 * time.Second) // Query Prometheus client, err := api.NewClient(api.Config{ Address: "http://localhost:9090", }) if err != nil { t.Fatalf("Error creating client: %v", err) } v1api := v1.NewAPI(client) ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second) defer cancel() result, warnings, err := v1api.Query(ctx, "order_processing_duration_seconds_count", time.Now()) if err != nil { t.Fatalf("Error querying Prometheus: %v", err) } if len(warnings) > 0 { t.Logf("Warnings: %v", warnings) } // Check the result if result.(model.Vector).Len() == 0 { t.Errorf("Expected non-empty result") } }
It’s important to verify that your monitoring system performs well under load. You can use tools like hey or vegeta to generate load on your system while observing your metrics:
hey -n 10000 -c 100 http://localhost:8080/orders
While the load test is running, observe your Grafana dashboards and check that your metrics are updating as expected and that Prometheus is able to keep up with the increased load.
To test your alerting rules, you can temporarily adjust the thresholds to trigger alerts, or use Prometheus’s API to manually fire alerts:
curl -H "Content-Type: application/json" -d '{ "alerts": [ { "labels": { "alertname": "HighOrderProcessingErrorRate", "severity": "critical" }, "annotations": { "summary": "High order processing error rate" } } ] }' http://localhost:9093/api/v1/alerts
This will send a test alert to your Alertmanager, allowing you to verify that your notification channels are working correctly.
As you implement and scale your monitoring system, keep these challenges and considerations in mind:
High cardinality can lead to performance issues in Prometheus. Be cautious when adding labels to metrics, especially labels with many possible values (like user IDs or IP addresses). Instead, consider using histogram metrics or reducing the cardinality by grouping similar values.
For large-scale systems, consider:
Your monitoring system is critical infrastructure. Consider:
次のことを確認してください:
アラートノイズを軽減するには:
この投稿では、Prometheus と Grafana を使用した注文処理システムの包括的な監視とアラートについて説明しました。私たちはカスタム指標を設定し、有益なダッシュボードを作成し、アラートを実装し、高度な技術と考慮事項を検討しました。
シリーズの次のパートでは、分散トレースとロギングに焦点を当てます。以下について説明します:
今後も注文処理システムの強化を継続し、分散システムの動作とパフォーマンスについてより深い洞察を得ることに注力していきますので、ご期待ください!
困難な問題に直面していますか、それとも新しいアイデアやプロジェクトに関して外部の視点が必要ですか?お手伝いできます!大規模な投資を行う前にテクノロジーの概念実証を構築したい場合でも、難しい問題についてのガイダンスが必要な場合でも、私がお手伝いいたします。
私と協力することに興味がある場合は、hungaikevin@gmail.com まで電子メールでご連絡ください。
課題をチャンスに変えましょう!
The above is the detailed content of Implementing an Order Processing System: Part Monitoring and Alerting. For more information, please follow other related articles on the PHP Chinese website!