Proper Thanos Receive Setup with Remote Write and Duplication Concerns

[lordoe]

Hi Thanos team and community,

I'm currently working on setting up a Thanos Receive setup for long-term storage of metrics in a multi-cluster OpenShift environment, and I’m looking for clarification around replication factor and metric deduplication in the context of remote write.

---

I have the following setup:

Source (SRC) OpenShift Cluster

• Prometheus with remote write enabled
• Each metric is scraped twice (due to two Prometheus replicas scraping the same targets)

Destination (DST) OpenShift Cluster:

• Thanos Receive (with multiple replicas)
• Thanos Querier, Compactor, and Store set up properly

Questions

1. Is it correct that Thanos neither deduplicates metrics at ingest time (via Receive), nor at compaction (via Compactor), and that deduplication only happens at query time via the Querier?
2. Given that metrics are already duplicated at the Prometheus level (2x), does setting a --receive.replication-factor=2 lead to a 4x storage duplication? That is, each of the already-duplicated samples gets written to two Receive replicas?
3. If I reduce the --receive.replication-factor to 1, do I:

• Still need more than one Receive replica?
• Lose high availability? (e.g., if a single Receive replica goes down, would data be lost?)

4. I tested with replication factor 1 and two Receive replicas. After killing one replica, I observed that ingestion wasn't evenly distributed — is this expected behavior?

If it helps, here more context about what I want to do:

I want to observe my SRC cluster's metrics long-term in the DST cluster, using remote write to Thanos Receive. I'm trying to find the right balance between storage efficiency and high availability, and currently unsure whether replication-factor > 1 makes sense in my scenario.

Any insights or corrections to my understanding are highly appreciated!

Thank you in advance!

Kind Regards