PingTower Metrics Writer

Batches ping history into ClickHouse — up to 500 rows or once a second, whichever comes first.

Stack: C#, .NET 10, ClickHouse, RabbitMQ, Polly.

Role in the system

Every probe made by ping-service ends up here. metrics-writer turns the high-frequency event stream into efficient bulk inserts, so ClickHouse gets a few large writes instead of thousands of tiny ones. The api then reads this table for latency charts, uptime and ping history.

flowchart LR
  PS[ping-service] -->|server.ping.recorded| S((metrics-writer))
  S -->|bulk INSERT| CH[(ClickHouse<br>server_pings)]
  CH -.->|charts, uptime| API[api]

Loading

Features

• Size-or-time batching — a batch is flushed at BatchSize messages (default 500) or after FlushIntervalMs (default 1000 ms).
• At-least-once delivery — messages are acked only after the batch is written; a failed insert nacks the whole batch back to the queue.
• Resilient writes — Polly retry (3×, exponential backoff), 10 s timeout and a circuit breaker (opens for 2 min after 5 failures).
• Back-pressure — bounded in-memory channel and RabbitMQ prefetch (must be greater than the batch size).
• Poison messages — payloads that cannot be parsed are rejected without requeue.

Contracts

Direction	Channel	Name	Payload
In	queue ← pingEventsExchange	q.metrics-writer.ping-events (server.ping.recorded)	ping result
Storage	ClickHouse	pingtower_analytics.server_pings	one row per ping; MergeTree, partitioned by day, TTL 30 days

The table is created by infra/clickhouse/clickhouse-init/init.sql.

Quick start

Whole stack — via infra (all repos cloned side by side):

make -C infra up

This service only (broker and ClickHouse already running from infra). Create .env with the variables referenced in docker-compose.yml, then:

docker compose up -d --build

Local development:

dotnet run --project src/MetricsWriter
dotnet test src/src.sln

Structure

metrics-writer/
├── src/
│   ├── Domain/            # PingRecord, IPingRecordWriter
│   ├── Application/       # incoming message DTOs
│   ├── Infrastructure/    # ClickHouse bulk writer + Polly decorator
│   └── MetricsWriter/     # hosted worker: consume → batch → flush → ack
└── tests/                 # unit and integration tests