Auditing My Own Estimate: Langfuse from $3.4K a Month to $1.0K
The first Langfuse design on EKS was full HA and about $3,400 a month at the load I expected. Scenarios ran from about $1.8K to $7.7K. I re-checked it before buying the hardware. The estimate was about 3.5× too high. The tier that shipped is about $1,000 a month, roughly 71% below the original number.
What the $3.4K was buying
- ClickHouse, 1 shard and 3 replicas, one per AZ, on r8g.2xlarge, each with a 1,500 GiB gp3 volume at 8,000 IOPS.
- ClickHouse Keeper on 3 separate small nodes.
- RDS PostgreSQL, ElastiCache Valkey, and S3 for events, media, exports, and backups.
- Separate web and ingest deployments, because heavy ingest hurts the UI.
Where the money actually was
About $1,000 a month needed no redesign:
- Monitoring nodes that duplicated a stack that already existed.
- Provisioned IOPS that the workload did not need.
- 90-day retention when 30 days was enough.
About $1,400 a month was availability I was not ready to pay for yet: a single ClickHouse replica, no Keeper, and a deferred split between ingest and UI.
| Tier | Monthly cost | What you get |
|---|---|---|
| Pilot | ~$0.55K | 20% trace sampling |
| Lean (shipped) | ~$1.0K | Single ClickHouse node, single-AZ Postgres, one queue node |
| AZ-resilient | ~$1.6K | Survives the loss of an availability zone |
| Full HA | ~$2.5K | Replicated ClickHouse with Keeper, split ingest and UI |
The single queue node looked like silent trace loss in the first design. Langfuse writes the event to S3 before it enqueues it, and the queue holds about 12 hours. A dead queue delays traces. It does not delete them. That is the risk the lean tier accepted.
The as-built runbook overrides the design doc wherever they differ. If you only read the ADR, you will build the $3.4K version.
How the cluster is wired, and the seven outages worth preventing, is in Self-hosting Langfuse on Kubernetes.