configure-log-aggregation（配置日志聚合）

Configure Log Aggregation

Implement centralized log collection, parsing, and querying with Loki/Promtail or ELK stack for operational visibility.

When to Use

• Consolidating logs from multiple services or hosts into a searchable system
• Replacing local log files with centralized, queryable log storage
• Correlating logs with metrics and traces for full observability
• Implementing structured logging with label extraction from unstructured logs
• Setting retention policies for log data based on storage and compliance needs
• Troubleshooting production incidents requiring log analysis across services

Inputs

• Required: Log sources (application logs, system logs, container logs)
• Required: Log format patterns (JSON, plaintext, syslog, etc.)
• Optional: Label extraction rules for structured querying
• Optional: Retention and compression policies
• Optional: Existing log shipper configuration (Fluentd, Filebeat, Promtail)

Procedure

See Extended Examples for complete configuration files and templates.

Step 1: Choose Log Aggregation Stack

Select between Loki (Prometheus-style) or ELK (Elasticsearch-based) based on requirements.

Loki advantages:

• Lightweight, designed for Kubernetes and cloud-native environments
• Label-based indexing (like Prometheus) for low storage overhead
• Native integration with Grafana for unified dashboards
• Horizontal scalability with object storage (S3, GCS)
• Lower resource consumption compared to Elasticsearch

ELK advantages:

• Full-text search across all log content (not just labels)
• Rich query DSL and aggregations
• Mature ecosystem with beats, logstash plugins
• Better for compliance/audit logs requiring deep historical search

For this guide, we'll focus on Loki + Promtail (recommended for most modern setups).

Decision criteria:

Use Loki if:
- You want label-based queries similar to Prometheus
- Storage costs are a concern (Loki indexes only labels)
- You already use Grafana for metrics
- Kubernetes/container-native deployment

Use ELK if:
- You need full-text search across all log content
- You have complex log parsing and enrichment requirements
- You require advanced analytics and aggregations
- Legacy systems with existing Logstash pipelines

Expected: Clear choice made based on requirements, team downloads appropriate installation artifacts.

On failure:

• Benchmark storage requirements: Loki ~10x less than Elasticsearch for same logs
• Evaluate query patterns: full-text search needs vs label filtering
• Consider operational overhead: ELK requires more tuning and resources

Step 2: Deploy Loki

Install and configure Loki with appropriate storage backend.

Docker Compose deployment (docker-compose.yml):

version: '3.8'

services:
  loki:
    image: grafana/loki:2.9.0
    ports:
      - "3100:3100"
    volumes:
      - ./loki-config.yml:/etc/loki/local-config.yaml
      - loki-data:/loki
    command: -config.file=/etc/loki/local-config.yaml
    restart: unless-stopped

  promtail:
    image: grafana/promtail:2.9.0
    volumes:
      - ./promtail-config.yml:/etc/promtail/config.yml
      - /var/log:/var/log:ro
      - /var/lib/docker/containers:/var/lib/docker/containers:ro
    command: -config.file=/etc/promtail/config.yml
    restart: unless-stopped
    depends_on:
      - loki

volumes:
  loki-data:

Loki configuration (loki-config.yml):

auth_enabled: false

server:
  http_listen_port: 3100
  grpc_listen_port: 9096

# ... (see EXAMPLES.md for complete configuration)

For production with S3 storage:

storage_config:
  aws:
    s3: s3://us-east-1/my-loki-bucket
    s3forcepathstyle: true
  boltdb_shipper:
    active_index_directory: /loki/index
    cache_location: /loki/cache
    shared_store: s3

Expected: Loki starts successfully, health check passes at http://localhost:3100/ready, logs stored according to retention policy.

On failure:

• Check Loki logs: docker logs loki
• Verify storage directories exist and are writable
• Test config syntax: docker run grafana/loki:2.9.0 -config.file=/etc/loki/local-config.yaml -verify-config
• Ensure retention settings don't exceed disk capacity
• For S3: verify IAM permissions and bucket access

Step 3: Configure Promtail for Log Shipping

Set up Promtail to scrape logs and forward to Loki with label extraction.

Promtail configuration (promtail-config.yml):

server:
  http_listen_port: 9080
  grpc_listen_port: 0

positions:
  filename: /tmp/positions.yaml
# ... (see EXAMPLES.md for complete configuration)

Key Promtail concepts:

• Scrape configs: Define log sources and how to discover them
• Pipeline stages: Transform and label logs before sending to Loki
• Relabel configs: Dynamic labeling based on metadata
• Positions file: Tracks read offsets to avoid re-processing logs

Expected: Promtail scrapes configured log files, labels applied correctly, logs visible in Loki via LogQL queries.

On failure:

• Check Promtail logs: docker logs promtail
• Verify file paths are accessible: docker exec promtail ls /var/log
• Test regex patterns independently with sample log lines
• Monitor Promtail metrics: curl http://localhost:9080/metrics | grep promtail
• Check positions file for progress: cat /tmp/positions.yaml

Step 4: Query Logs with LogQL

Learn LogQL syntax for filtering and aggregating logs.

Basic queries:

# All logs from a job
{job="app"}

# Logs with specific label values
{job="app", level="error"}

# Regex filter on log line content
{job="app"} |~ "authentication failed"

# Case-insensitive regex
{job="app"} |~ "(?i)error"

# Line filter (doesn't parse, just includes/excludes)
{job="app"} |= "user"  # Contains "user"
{job="app"} != "debug" # Doesn't contain "debug"

Parsing and filtering:

# JSON parsing
{job="app"} | json | level="error"

# Regex parsing with named groups
{job="app"} | regexp "user_id=(?P<user_id>\\d+)" | user_id="12345"

# Logfmt parsing (key=value format)
{job="app"} | logfmt | level="error", service="auth"

# Pattern parsing
{job="nginx"} | pattern `<ip> - <user> [<timestamp>] "<method> <path> <protocol>" <status> <size>` | status >= 500

Aggregations (metrics from logs):

# Count log lines per level
sum by (level) (count_over_time({job="app"}[5m]))

# Rate of error logs
rate({job="app", level="error"}[5m])

# Bytes processed per service
sum by (service) (bytes_over_time({job="app"}[1h]))

# Average request duration from logs
avg_over_time({job="app"} | json | unwrap duration [5m])

# Top 10 error messages
topk(10, sum by (message) (count_over_time({level="error"} [1h])))

Filtering by extracted fields:

# Find specific trace in logs
{job="app"} | json | trace_id="abc123def456"

# HTTP 5xx errors from nginx
{job="nginx"} | pattern `<_> "<_> <_> <_>" <status> <_>` | status >= 500

# Failed authentication attempts
{job="app"} | json | message=~"authentication failed" | user_id != ""

Create Grafana explore queries or dashboard panels using these patterns.

Expected: Queries return expected log lines, filtering works correctly, aggregations produce metrics from logs.

On failure:

• Use Grafana Explore to debug queries interactively
• Check label names: curl http://localhost:3100/loki/api/v1/labels
• Verify label values: curl http://localhost:3100/loki/api/v1/label/{label_name}/values
• Simplify query: start with basic label selector, add filters incrementally
• Check time range: logs might not exist in selected window

Step 5: Integrate Logs with Metrics and Traces

Correlate logs with Prometheus metrics and distributed traces for unified observability.

Add trace IDs to logs (application instrumentation):

# Python with OpenTelemetry
import logging
from opentelemetry import trace

logger = logging.getLogger(__name__)

def handle_request():
    span = trace.get_current_span()
    trace_id = span.get_span_context().trace_id

    logger.info(
        "Processing request",
        extra={"trace_id": format(trace_id, "032x")}
    )

// Go with OpenTelemetry
import (
    "go.opentelemetry.io/otel/trace"
    "go.uber.org/zap"
)

func handleRequest(ctx context.Context) {
    span := trace.SpanFromContext(ctx)
    traceID := span.SpanContext().TraceID().String()

    logger.Info("Processing request",
        zap.String("trace_id", traceID),
    )
}

Configure Grafana data links from metrics to logs:

In Prometheus panel field config:

{
  "fieldConfig": {
    "defaults": {
      "links": [
        {
          "title": "View Logs",
          "url": "/explore?left={\"datasource\":\"Loki\",\"queries\":[{\"refId\":\"A\",\"expr\":\"{job=\\\"app\\\",instance=\\\"${__field.labels.instance}\\\"} |= `${__field.labels.trace_id}`\"}],\"range\":{\"from\":\"${__from}\",\"to\":\"${__to}\"}}",
          "targetBlank": false
        }
      ]
    }
  }
}

Configure Grafana data links from logs to traces:

In Loki datasource config:

datasources:
  - name: Loki
    type: loki
    url: http://loki:3100
    jsonData:
      derivedFields:
        - datasourceName: Tempo
          matcherRegex: "trace_id=(\\w+)"
          name: TraceID
          url: "$${__value.raw}"

Correlate logs in Grafana Explore:

1. Query metrics in Prometheus
2. Click on data point
3. Select "View Logs" from context menu
4. Loki query auto-populated with relevant labels and time range
5. Click trace ID in logs
6. Tempo trace view opens with full distributed trace

Expected: Clicking metrics opens related logs, trace IDs in logs link to trace viewer, single pane for metrics/logs/traces navigation.

On failure:

• Verify trace ID format matches regex in derived fields
• Check that trace_id label extracted by Promtail pipeline
• Ensure Tempo datasource configured in Grafana
• Test URL encoding for complex filter expressions
• Validate data link URLs in incognito/private browser window

Step 6: Set Up Log Retention and Compaction

Configure retention policies and compaction to manage storage costs.

Retention by stream (in Loki config):

limits_config:
  retention_period: 720h  # Global default: 30 days

  # Per-tenant retention (requires multi-tenancy enabled)
  per_tenant_override_config: /etc/loki/overrides.yaml

# overrides.yaml
overrides:
  production:
    retention_period: 2160h  # 90 days for production
  staging:
    retention_period: 360h   # 15 days for staging
  development:
    retention_period: 168h   # 7 days for dev

Retention by stream labels (requires compactor):

compactor:
  working_directory: /loki/compactor
  shared_store: filesystem
  compaction_interval: 10m
  retention_enabled: true
  retention_delete_delay: 2h
# ... (see EXAMPLES.md for complete configuration)

Priority determines which rule applies when multiple match (lower number = higher priority).

Compression settings:

chunk_store_config:
  chunk_cache_config:
    enable_fifocache: true
    fifocache:
      max_size_bytes: 1GB
      ttl: 24h
# ... (see EXAMPLES.md for complete configuration)

Monitor retention:

# Check chunk stats
curl http://localhost:3100/loki/api/v1/status/chunks | jq

# Check compactor metrics
curl http://localhost:3100/metrics | grep loki_compactor

# Verify deleted chunks
curl http://localhost:3100/metrics | grep loki_boltdb_shipper_retention_deleted

Expected: Old logs automatically deleted per retention policy, storage usage stabilizes, compaction reduces index size.

On failure:

• Enable compactor in Loki config if retention not working
• Check compactor logs: docker logs loki | grep compactor
• Verify retention_enabled: true and retention_deletes_enabled: true
• Monitor disk usage: du -sh /loki/
• For S3: check bucket lifecycle policies don't conflict with Loki retention

Validation

• Loki API health check returns 200: curl http://localhost:3100/ready
• Promtail successfully scraping logs from all configured sources
• Labels extracted correctly from log lines (visible in Grafana Explore)
• LogQL queries return expected results with proper filtering
• Log retention policy enforced (old logs deleted after retention period)
• Logs accessible from Grafana dashboards and Explore view
• Trace IDs from logs link to Tempo trace viewer
• Metrics panels have data links to relevant logs
• Compaction running and reducing storage overhead
• Storage usage within allocated disk/S3 budget

Common Pitfalls

• High cardinality labels: Using unbounded label values (user IDs, request IDs) causes index explosion. Use fixed labels (level, service, env) and put variables in log lines.
• Missing log parsing: Sending raw logs without label extraction limits query capabilities. Always parse structured logs (JSON, logfmt) or use regex for unstructured.
• Incorrect time parsing: Mismatched timestamp formats cause logs to be out of order or rejected. Test timestamp parsing with sample logs.
• Retention not working: Compactor must be enabled for retention to delete old data. Check retention_enabled: true and retention_deletes_enabled: true.
• Ingestion rate limits: Default limits (10MB/s) may be too low for high-volume systems. Adjust ingestion_rate_mb and ingestion_burst_size_mb.
• Query timeouts: Broad queries over long time ranges can timeout. Use more specific label selectors and shorter time windows.
• Log duplication: Multiple Promtail instances scraping same logs create duplicates. Use unique labels or positions file coordination.

Related Skills

• correlate-observability-signals - Unified debugging across metrics, logs, and traces using trace IDs
• build-grafana-dashboards - Visualize log-derived metrics and create log panels in dashboards
• setup-prometheus-monitoring - Metrics provide context for when to query logs during incidents
• instrument-distributed-tracing - Add trace IDs to logs for correlation with distributed traces