Offloading local storage
A scraping server has limited local disk and is not built for very long retention or horizontal scale. Remote write streams samples out to an external, scalable storage backend in near real time.
How remote write behaves
- After each scrape, samples are queued and sent in batches to a remote endpoint.
- An in memory write ahead queue buffers samples and retries on failure.
- The protocol is typically compressed and sent over HTTP.
Reliability concerns
- Backpressure when the remote is slow can fill the queue, so shards and queue limits matter.
- Ordering and dedupe must be handled if multiple servers write for high availability.
- On crash a write ahead log lets unsent samples be recovered.
Why use it
It separates collection from storage, letting a fleet of small scrapers feed one large multi tenant store that handles retention, downsampling, and querying.
Key idea
Remote write streams batched samples from local scrapers to a scalable backend, separating collection from long term storage while handling retries and backpressure.