If you run Snowflake in production, you've probably asked yourself: which queries are burning credits? Why did costs spike last Tuesday? Are any warehouses just sitting there idle?
Most observability tools answer these questions by shipping your query logs to an external SaaS — data egress, vendor lock-in, another subscription. SnowLens takes a different approach: it's a Streamlit app that installs directly into your Snowflake account and runs on your own compute. Your data never leaves your environment.
And it's completely free.
What It Detects — 9 Rules
Performance:
Slow queries — configurable threshold (10s/20s/30s/60s) with SELECT-only toggle
Cancelled queries — manually stopped or killed by timeout
Failed queries — non-cancellation errors (syntax, permissions, resource limits)
Disk spill — local vs remote classification, remote flagged as high-risk
Queuing — warehouse contention when QUEUED_OVERLOAD_TIME exceeds 60s
Cost:
Credit spikes — hourly usage > 2 standard deviations above warehouse baseline
Idle-but-running — credits burned with zero queries
Oversized warehouses — big warehouse, tiny workload
Workload cost spikes — cost outliers by QUERY_TAG or role
Plus a cost attribution engine that groups credit usage by QUERY_TAG, dbt model, or role.
How It Works
SnowLens reads from three Snowflake metadata views that every account already has:
SNOWFLAKE.ACCOUNT_USAGE.QUERY_HISTORY
SNOWFLAKE.ACCOUNT_USAGE.WAREHOUSE_METERING_HISTORY
SNOWFLAKE.ACCOUNT_USAGE.QUERY_ATTRIBUTION_HISTORY
These are Snowflake's own internal metadata — not your business data. No tables, stages, or query results are touched.
Privacy Model
Only reads ACCOUNT_USAGE metadata
No outbound network calls (Streamlit-in-Snowflake runs in isolation)
Runs on XSMALL warehouse with 60s auto-suspend (cents per session)
Least-privilege role with only the grants the app needs
Clean uninstall in 3 lines of SQL
Memory Safety
Streamlit-in-Snowflake has a ~32 MB memory limit. SnowLens handles large accounts with:
Query text truncated to 300 characters
Each detector capped at top 200 findings via QUALIFY ROW_NUMBER()
Cost attribution aggregated in SQL before loading into Python
Warehouse scope selector to limit detection to individual warehouses
Install in 5 Minutes
Download the ZIP from snowlens.vizcanvaz.com
Run sql/01_setup.sql as ACCOUNTADMIN in Snowsight
Upload streamlit_app.py and environment.yml to the stage, run sql/02_create_app.sql
Done. Open Projects > Streamlit > SNOWLENS_APP.
Links
Product page: snowlens.vizcanvaz.com
GitHub: github.com/abysabin/snowlens
Full blog post with SQL logic: vizcanvaz.com
Built by Abin Alex — 15+ years in data engineering. Questions or feedback welcome at vizcanvas@gmail.com.
Top comments (0)