Network, cloud and databases

1. Network #

Who talks to whom.

Network Performance Monitoring #

Thanks to eBPF, the agent measures flows between services (volume, TCP latency, retransmissions, DNS resolutions) and aggregates them by tags: you can see that one service talks too much to another, or that an availability zone is losing packets.

2. Cloud #

Managed services.

Integrations #

• AWS: read-only IAM role and/or streaming metrics (CloudWatch Metric Streams via Firehose); logs via the Forwarder Lambda or Firehose.
• Azure: app registration and Event Hub for logs; GCP: service account and Pub/Sub for logs.
• Serverless: Lambda extension and libraries for function traces and metrics, without a host agent.

Common mistake: A cloud integration with all namespaces enabled collects thousands of useless metrics and slows down the interface: limit it to the services actually used and filter by resource tags.

3. Databases #

Understanding queries.

Database Monitoring #

Database Monitoring (PostgreSQL, MySQL, SQL Server, Oracle, MongoDB) collects normalized queries, their execution plans and wait events, and links a slow query to the APM service that issued it.

# conf.d/postgres.d/conf.yaml
init_config:
instances:
  - dbm: true
    host: db.internal
    port: 5432
    username: datadog
    password: "ENC[postgres_password]"

Good to know: The datadog account needs read permissions on pg_stat_statements and on the activity views; the documentation provides the creation queries to run.