Snowflake

AddisDB connects to Snowflake over its SQL REST API using a programmatic access token — no ODBC driver, no JDBC, nothing to install. Because Snowflake is a warehouse, it opens in the Big Data Console rather than the ordinary Data tab.

Who it is for

Snowflake separates storage from compute, so a heavy analytical query can be given a bigger warehouse for ten minutes and then scaled back down. Cloning a multi-terabyte database is instant and costs nothing until you change something, and data sharing works without copying.

Those last two both come from the same design: storage is immutable micro-partitions, so a clone is a pointer copy and time travel is just reading an older set of pointers. It is also why an unexpectedly large table is often mostly retention rather than live data.

It suits analytics and BI teams who want elastic compute without cluster administration, organizations consolidating data from many sources, and anyone who needs to isolate one team’s expensive queries from another’s.

The cost model rewards attention: you pay per second of warehouse uptime with a minimum per resume, so a warehouse left running between two queries costs more than the queries did.

Set up access

1. In Snowsight, go to your user profile and generate a programmatic access token (PAT). Copy it — it is shown once.
2. Note your account URL: something like ORGNAME-ACCOUNTNAME.snowflakecomputing.com.
3. Identify the database, schema and warehouse you will use. A warehouse is required to actually run queries — schema browsing alone works without one.
4. Make sure the role attached to your user has USAGE on the warehouse and SELECT on the schema.
5. Set the warehouse auto-suspend low (a minute or two) if it is only for exploring.
6. If your account enforces a network policy, add your IP to it.

Connect from AddisDB

1. New Connection → Snowflake under Big Data / Warehouse. Port prefills to 443.
2. Put your account URL in Host.
3. Set Database to slash-separated context: DATABASE/SCHEMA/WAREHOUSE — add /ROLE as a fourth segment if you need a specific role.
4. Paste the PAT into the API key / token field. It is stored in your device’s secure store, never on our servers.
5. Test, then Save.

# What goes in the Database field
ANALYTICS/PUBLIC/COMPUTE_WH

# …or with an explicit role
ANALYTICS/PUBLIC/COMPUTE_WH/ANALYST

Object names are stored upper case unless they were created quoted. If a query cannot find a table you can see in the sidebar, the mismatch is almost always quoting rather than permissions.

Working in the Big Data Console

Warehouse connections open in the Big Data Console: a catalog browser on the left for schemas and tables, a query pane on the right, and a row cap on every preview so an exploratory click cannot turn into a full-table scan.

Know the cost before you run it

Click Estimate cost and AddisDB asks Snowflake to plan the query, reading back the micro-partitions it would prune and the bytes it would actually read. You see the scan size before you spend the credits, not on next month’s bill.

The number that moves the estimate most is pruning. A predicate on the column a table is clustered by lets Snowflake skip whole micro-partitions; a predicate wrapped in a function usually cannot be pruned at all, and the estimate says so before you have paid for it.

What else AddisDB gives you

• Catalog introspection — schemas, tables and columns in the sidebar, plus ⌘K search across every connected database.
• The full safety model: read-only connections, prod tagging, and destructive-statement detection before anything runs.
• The Chart view for turning an aggregation into a visual, and notebooks for analysis you want to be readable later.
• AI querying grounded in your real Snowflake schema.
• The Live Monitor over QUERY_HISTORY and TABLE_STORAGE_METRICS, with per-table storage broken out.

Finding out why a table is enormous

The monitor reads TABLE_STORAGE_METRICS, which separates active storage from time-travel and fail-safe retention. That split is the answer to most "why are we paying for this table" questions: a table churned by frequent rewrites can hold several times its live size in retention, and the fix is a retention setting rather than a delete.