Migrate to ClickHouse

Migrate from Snowflake to ClickHouse

Snowflake is good at batch BI. It is not built for real time, high concurrency or latency sensitive analytics, and it is expensive. ClickHouse is, and our pricing analysis shows savings of 40% to 55%.

How the migration works
Sub second
Queries over billions of rows, from columnar storage and vectorised execution.
50%+
Cost savings observed in the migrations we have delivered from SQL Server, OpenSearch and Snowflake.
100s
Of concurrent users served with consistent latency, so analytics can go into your product.
Anywhere
Open source. Run it on any cloud, on premise or as ClickHouse Cloud.
Why teams move

Why Snowflake users come to us

A lot of companies have spent the last few years implementing Snowflake, and it is often the newest thing in the stack. It is genuinely good at internal BI and batch reporting. The conversation we have is about the workloads it is not designed for: real time analytics, many users issuing queries at the same time, and anything that needs sub second latency over fresh data.

The cost side matters too. We compared the two platforms on published pricing and found ClickHouse Cloud to be 40% to 55% cheaper for equivalent compute, before taking into account that ClickHouse needs less of it. Snowflake warehouses double in cost with each size step, whereas ClickHouse scales linearly in small increments, so the gap widens as you grow.

You do not have to move everything. Many teams keep Snowflake for the batch and finance reporting and move the real time, product and customer facing analytics to ClickHouse, with Iceberg as the shared format between them. Others move entirely. We will tell you honestly which makes sense.

Sound familiar?
  • [01]Credit consumption climbs every quarter and warehouses are sized up to keep dashboards responsive.
  • [02]Dashboards and embedded analytics cannot deliver sub second response with many concurrent users.
  • [03]Data is minutes or hours old by the time it is queryable, so real time use cases are off the table.
  • [04]Warehouses are left running to avoid cold start latency, which is paid for by the hour.
  • [05]Serving analytics to customers or into a product from Snowflake is too slow and too costly.
What changes

Snowflake to ClickHouse: faster, cheaper, simpler

ClickHouse is the fastest open source analytical database on the market. These are the differences that matter most for teams coming from this platform.

[01]

40% to 55% lower compute cost

Our published analysis modelled small, medium and large deployments and found ClickHouse Cloud cheaper by 40% to 55% on like for like compute, with the gap growing with utilisation.

[02]

Built for concurrency

ClickHouse serves hundreds of concurrent queries with consistent sub second latency, which is what makes it usable as the backend for a product rather than only an internal BI tool.

[03]

Real time by default

Data is queryable within seconds of arriving, whether from Kafka, CDC or object storage, and materialised views keep aggregates current without scheduled transformation jobs.

[04]

Scale in small steps, run anywhere

ClickHouse Cloud scales compute linearly in 8GB increments rather than doubling, and the open source engine runs on any cloud or on premise if you want to leave managed services behind.

How we migrate

Accelerated migration with CDC

Snowflake migrations are export driven rather than CDC driven. We unload tables to Parquet or Iceberg in S3 or GCS, where ClickHouse can read them directly, then use ClickPipes to keep loading new files or to connect the upstream sources, such as Kafka or Postgres, directly to ClickHouse so Snowflake is no longer in the path at all. Snowflake SQL, dbt models and semantic layers are translated as part of the work.

  1. 01

    Assess

    We review the schemas, the query workload and the cost of the current platform, then agree the scope, the target architecture and the business case before anything is built.

  2. 02

    Design the ClickHouse schema

    Tables are redesigned for columnar storage rather than copied across one to one. Sorting keys, partitioning, codecs and materialised views are chosen against your real queries.

  3. 03

    Load and replicate

    Historical data is bulk loaded, then changes are streamed continuously so ClickHouse stays in line with the source for as long as the two need to run side by side.

  4. 04

    Run in parallel

    Dashboards, applications and reports are pointed at ClickHouse and validated against the source. Performance and cost are measured rather than assumed.

  5. 05

    Cut over

    Workloads move across in stages, the old platform is scaled down and the remaining spend is retired. Nothing is switched off until the numbers reconcile.

  6. 06

    Support

    After go live we can monitor the environment, tune it as the workload grows and provide an escalation path under SLA through our proactive support service.

Tooling we use
Parquet and object storage

Warehouse exports land in S3 or GCS as Parquet or Iceberg. ClickHouse reads them directly with its table functions, or ClickPipes keeps ingesting new files as they arrive.

ClickPipes

The managed ingestion service built into ClickHouse Cloud. Its CDC connectors take an initial snapshot and then stream every insert, update and delete, with no infrastructure for you to run.

We are tool agnostic. The right choice depends on whether you are going to ClickHouse Cloud or self hosted, how much control you want over the pipeline and what else is in the stack. Zero downtime is the norm, not the exception.

Beyond the migration

One database, many use cases

Moving off Snowflake is usually the start. Because ClickHouse is so versatile, the same platform tends to pick up more workloads once it is in place.

Real time analytics

Dashboards and embedded analytics over data that is seconds old, served to many users at once.

Data warehousing

A warehouse that is fast enough to query directly, without the overnight batch or the cube layer.

Observability

Logs, metrics and traces at a fraction of the cost of Datadog, Splunk or Elastic, with ClickStack on top.

Time series

Market data, IoT, telemetry and events at hundreds of thousands of rows per second with long retention.

Search and security analytics

Full text search, SIEM and threat hunting over high volume event data using plain SQL.

Agentic AI

A data platform fast and cheap enough for agents that turn one question into dozens of queries.

Related work

Migrations we have delivered

Gaming

Helping An Online Gaming Company Migrate From Postgres To ClickHouse Cloud

Building a real-time data warehouse for a sports betting and gaming platform processing ~300 million events per day, scaling to trillions of rows.

Manufacturing & Retail

Helping A Food Manufacturer And Retailer Deliver A Real Time Data Warehouse Using ClickHouse

Migrating 20 years of sales data from a legacy Oracle data warehouse to ClickHouse, resolving performance and memory issues while accelerating business intelligence.

Further reading

Ready to move off Snowflake?

Start with a short call. We will look at your schemas, queries and current spend, tell you honestly whether ClickHouse is the right fit, and put a number on the performance and cost you can expect.

Support after go live