Running on Snowflake?

Snowflake credits running away from you? We bring it back.

A warehouse that was supposed to idle is somehow running at XL on a Sunday morning, the finance team is asking why the bill tripled, and the dbt job that used to finish by 9am is still going at noon. We open the query history, find the handful of workloads that are costing the real money, and quietly turn the invoice around.

  • Since 2015Fixing technology
  • FreeEstimates
  • US teamReal engineers

snowflake-usage.log

still failing

What it is

Snowflake at a glance

Snowflake is a cloud data warehouse built for analytics on large structured and semi-structured datasets. Compute and storage are separated: data sits once in cloud object storage, and virtual warehouses of varying sizes read from it on demand, which is why most teams end up with several warehouses sized for different workloads. It runs on AWS, Azure and Google Cloud, and queries are written in a dialect of ANSI SQL with native support for JSON, Parquet and other semi-structured formats through the VARIANT type.

The company started in 2012 and went public in 2020. Pricing is credit-based: a warehouse consumes credits per second while it is running, and storage is billed separately per compressed terabyte. Features such as Snowpipe for streaming ingestion, Time Travel for historical queries, zero-copy cloning, Snowpark for Python and Scala, dynamic tables and Cortex for in-warehouse AI have been bolted on over the years, which makes the platform powerful and also makes the bill easy to misread.

Type
Cloud data warehouse
First released
2014 general availability
License
Proprietary SaaS, credit-based pricing
Maintained by
Snowflake, Inc.
Typically used for
BI reporting, analytics on large datasets, dbt pipelines, data sharing, embedded analytics

The honest version

Snowflake pros and cons

No technology is all upside. This is where Snowflake earns its place and where it tends to cause trouble.

Where Snowflake is strong

  • Compute and storage scale independentlyYou can resize a warehouse up to crunch a report and park it back down without touching the data itself, which is a different mental model from a classic MPP database.
  • Semi-structured data is a first-class citizenVARIANT, OBJECT and ARRAY columns query JSON and Parquet with ordinary SQL, which saves a lot of ETL that other warehouses still demand.
  • Time Travel and zero-copy clones are genuinely usefulYou can query a table as it was an hour ago, or clone a terabyte of production for testing without doubling the storage bill.
  • Concurrency is built in, not tacked onMulti-cluster warehouses handle unpredictable BI traffic by spinning up extra clusters under load, rather than queueing every analyst behind the slowest query.
  • Data sharing without exportsSecure Data Sharing lets a partner query your data in place with no copy, no SFTP, and no nightly refresh job to maintain.
  • A well-stocked ecosystemEvery major BI tool, ingestion service and dbt setup talks to Snowflake natively, which keeps the plumbing short.

Where Snowflake bites

  • Credits disappear faster than people expectAn XL warehouse left on through the weekend quietly spends thousands of dollars, and the default auto-suspend of 10 minutes is long enough to notice on the invoice.
  • Query tuning is different and still necessaryThere are no traditional indexes. Clustering keys, micro-partition pruning and result caching are the real levers, and most teams reach for a bigger warehouse long before they reach for those.
  • Per-second billing is not free during idleA warehouse resumed by a dashboard every minute runs almost continuously, even if each individual query is quick, because of the minimum per-second charge and the auto-suspend floor.
  • Lock-in grows with every new featureSnowpark procedures, Cortex functions, dynamic tables and Streams are productive and rarely portable, so moving off the platform is an engineering project rather than a configuration change.
  • Access controls get complex quicklyThe role hierarchy, future grants and ownership rules trip up teams that treat permissions as an afterthought, and audits often find surprises.
  • Partner tools can run up a bill without you seeing itAn ingestion tool or BI product with a small scheduled job can wake a warehouse every few minutes all night, which shows up as a line item nobody budgeted for.

Sound familiar?

Common Snowflake problems we fix

The Snowflake problems that bring people to us, and what you have once each one is fixed.

Credit consumption: 184 credits yesterday (baseline 22)

The Snowflake bill spiked and nobody is sure why

Usually one warehouse running oversized, a dashboard refreshing every minute against a large table, or a dbt job that started scanning far more than it should. Query history and warehouse metering show exactly which workload is responsible once someone knows where to look.

Costs traced to the queries that caused them

WH did not auto-suspend, 14 hours idle

Warehouses stay up overnight

A connector or partner tool holds a session open, which counts as activity, so the warehouse never hits the idle threshold and keeps billing through the night.

Warehouses suspend when nothing is really running

Query scanned 4.1 TB

A query that used to be fast now scans the whole table

Clustering drifted, or the predicate no longer lines up with the micro-partition layout, so pruning does almost nothing and every run reads the full table.

Pruning restored, run time and cost both down

statement_timeout_in_seconds: 3600

Long queries keep timing out

A transformation that outgrew its warehouse, an unnecessary cross join, or a window function with no partition. Throwing a bigger warehouse at it is the expensive answer, not the right one.

Query rewritten, warehouse size matched to real need

exceeded memory per partition

Queries spill to disk and crawl

Memory-heavy joins and sorts exceeding what the warehouse size can hold, spilling to local then remote storage and multiplying run time.

Joins reshaped or warehouse sized once, deliberately

Resource monitor TRIGGERED

A resource monitor is cutting off real work

Credit guardrails were set once and forgotten, and a legitimate end-of-month report is being killed because the monitor was tuned for last quarter.

Guardrails that stop runaways, not reports

no owner for role ANALYTICS_RW

The role hierarchy is a mess

Grants were handed out as needed, future grants were never set, and nobody can say with confidence who can read which schema today.

Roles cleaned up, grants documented

dbt run time doubled in a week

The dbt project slowed down for no obvious reason

A model picked up a new source table, incremental logic silently fell back to a full refresh, or a merge key changed and the warehouse is now rewriting far more rows than it needs to.

Incrementals correct, runs back under the window

What we do

Snowflake help, from cost rescue to steady operation

Call us for one runaway warehouse, or hand us the whole account.

Cost rescue
A bill that jumped, a warehouse running unchecked, or a connector that never sleeps. We open query history, find the cause, and show the savings in credits and dollars.
Query and warehouse tuning
Clustering keys, pruning, warehouse sizing and multi-cluster policies set against your actual workloads rather than default settings.
dbt and pipeline work
dbt projects that are slow, fragile or doing full refreshes when they should be incremental, rebuilt to run within the window you need.
Access and governance
Role hierarchy, future grants, masking policies and row access policies set up so audits are a short conversation rather than a scramble.
Migrations
Moving from Redshift, BigQuery, or a legacy warehouse onto Snowflake, or moving data in or out without breaking downstream dashboards.
Managed Tech
Credit alerts that someone actually reads, monthly cost reviews, and a team on call when a connector decides to hold a warehouse open all night.

How it works

From first message to fixed

  1. Look Free estimate

    Tell us what's going wrong. We look at it, then tell you plainly what's wrong and what it will cost.

  2. Fix

    We fix the cause, not the symptom. If something is down, getting it back up comes first.

  3. Harden

    Backups, monitoring, updates and access, so the same problem can't quietly come back.

  4. Hand back or stay on

    You get a plain-language write-up. Or we stay on for hosting, upkeep and new work.

Often found together

Running Snowflake alongside something else?

We work on the whole stack, so the problem can be wherever it actually is.

Questions

Snowflake questions

Something else? Call+1 737 727 7327

Why is our Snowflake bill so much higher than expected?

In our experience it is almost always one of three things. A warehouse that is sized larger than the workload needs and runs for hours, a dashboard or connector that wakes the warehouse constantly so auto-suspend never fires, or a single heavy dbt model that scans the whole fact table every run. Query history, warehouse metering and the ACCOUNT_USAGE views show which one it is, and the fix is usually a small configuration change rather than a rewrite.

How do we stop Snowflake warehouses from running all night?

First, confirm the warehouse is really idle by looking at QUERY_HISTORY rather than trusting the dashboard. Then tighten auto-suspend to a minute or two for development warehouses, review which service accounts keep sessions open, and schedule the connectors that insist on waking the warehouse so they batch rather than poll. A resource monitor as a safety net catches anything that slips through.

Can you take over a Snowflake account we inherited?

Yes, and inherited accounts are a large part of this work. We review the warehouses, roles, resource monitors and network policies, document which workloads run where, rotate credentials, and bring the whole thing back into a state you can explain to your auditors and your CFO.

Should we move from Redshift or BigQuery to Snowflake (or the other way)?

Only when there is a real reason. Snowflake tends to win on concurrency, semi-structured data and operational simplicity. BigQuery can be cheaper for very bursty, read-heavy analytics in the Google ecosystem. Redshift still makes sense for teams deep in AWS. We will model the actual query mix and pricing before recommending a move.

How do we tune a slow Snowflake query without just upsizing the warehouse?

By looking at the query profile rather than guessing. Pruning, spill to remote storage, cardinality estimates and the shape of the join tree all show up there. A larger warehouse hides the symptom and multiplies the cost. Most of the slow queries we see need a clustering key reconsidered, a predicate rewritten or an incremental strategy fixed, not a bigger machine.

How much does Snowflake help cost?

The estimate is free. Send a few weeks of query history or share read access to the account, and we come back with a price and a shortlist of the workloads worth fixing first.

Free estimate

Snowflake costs or queries worrying you?

Tell us what you're running and what's going wrong. A person reads it and comes back with an estimate.

  • No charge to find out what it will take
  • If it's urgent, say so

Replies come from a person. By sending, you authorize us to email you back. No marketing.