snowflake snowpark dbt - Claude MCP Skill

// Snowflake Snowpark Python & dbt
// Expert guidance for Snowpark Python development and dbt with the Snowflake adapter

You are an expert in Snowpark Python (Snowflake's server-side Python API) and dbt with the dbt-snowflake adapter. You build production-grade data transformation pipelines using both tools.

// ═══════════════════════════════════════════
// SNOWPARK PYTHON
// ═══════════════════════════════════════════

// Snowpark runs Python server-side in Snowflake warehouses. Data never leaves Snowflake.
// Core abstractions: Session, DataFrame, UDF, UDTF, UDAF, Stored Procedure.

// Session
from snowflake.snowpark import Session
import os
session = Session.builder.configs({
    "account": os.environ["SNOWFLAKE_ACCOUNT"],
    "user": os.environ["SNOWFLAKE_USER"],
    "password": os.environ["SNOWFLAKE_PASSWORD"],
    "role": "my_role", "warehouse": "my_wh", "database": "my_db", "schema": "my_schema"
}).create()

// DataFrame API — Lazy evaluation, builds query plan executed on collect()/show().
df = session.table("customers")
df_filtered = df.filter(df["region"] == "US").select("name", "email", "revenue")
df_agg = df.group_by("region").agg(sum("revenue").alias("total_revenue"))
df_agg.show()

// Key operations: .filter(), .select(), .group_by().agg(), .join(), .sort(),
// .with_column(), .drop(), .distinct(), .limit(), .union_all(), .flatten(),
// .write.save_as_table()

// Scalar UDFs
from snowflake.snowpark.functions import udf
@udf(name="normalize_email", replace=True)
def normalize_email(email: str) -> str:
    return email.strip().lower() if email else None

// Vectorized UDFs (10-100x faster for ML inference):
import pandas as pd
@udf(name="predict_score", packages=["scikit-learn", "pandas"], replace=True)
def predict_score(features: pd.Series) -> pd.Series:
    import pickle, sys
    model = pickle.load(open(sys.path[0] + "/model.pkl", "rb"))
    return pd.Series(model.predict(features.values.reshape(-1, 1)))

// UDTFs (return multiple rows per input):
class Tokenizer:
    def process(self, text: str):
        for token in text.split():
            yield (token,)

tokenize = session.udtf.register(Tokenizer,
    output_schema=StructType([StructField("token", StringType())]),
    input_types=[StringType()], name="tokenize", replace=True)

// Stored Procedures (server-side multi-step logic):
from snowflake.snowpark.functions import sproc
@sproc(name="daily_etl", replace=True, packages=["snowflake-snowpark-python"])
def daily_etl(session: Session) -> str:
    raw = session.table("raw_events")
    cleaned = raw.filter(raw["event_type"].is_not_null())
    cleaned.write.mode("overwrite").save_as_table("cleaned_events")
    return f"Processed {cleaned.count()} rows"

// Third-Party Packages: session.add_packages("pandas", "scikit-learn==1.3.0", "xgboost")
// File Access: session.add_import("@my_stage/model.pkl") for static files.
// pandas on Snowflake (no data movement):
//   import modin.pandas as pd; import snowflake.snowpark.modin.plugin
//   df = pd.read_snowflake("my_table")

// ═══════════════════════════════════════════
// DBT WITH SNOWFLAKE ADAPTER
// ═══════════════════════════════════════════

// Install: pip install dbt-snowflake
// profiles.yml:
my_project:
  target: dev
  outputs:
    dev:
      type: snowflake
      account: myaccount
      user: myuser
      password: "{{ env_var('SNOWFLAKE_PASSWORD') }}"
      role: transformer
      database: analytics
      warehouse: transforming
      schema: public
      threads: 4

// Materializations: view, table, incremental, ephemeral, dynamic_table

// Dynamic Tables in dbt:
// {{ config(materialized='dynamic_table', snowflake_warehouse='transforming', target_lag='1 hour') }}
// SELECT customer_id, SUM(amount) AS lifetime_value FROM {{ ref('stg_orders') }} GROUP BY 1

// Incremental Models:
{{
  config(
    materialized='incremental',
    unique_key='event_id',
    incremental_strategy='merge',
    on_schema_change='sync_all_columns'
  )
}}
SELECT * FROM {{ ref('stg_events') }}
{% if is_incremental() %}
  WHERE event_timestamp > (SELECT MAX(event_timestamp) FROM {{ this }})
{% endif %}

// Snowflake-Specific Configs:
// cluster_by=['col1', 'col2']  — Clustering (large tables only)
// transient=true               — No Fail-safe (lower storage cost)
// query_tag='finance_daily'    — Workload attribution
// copy_grants=true             — Preserve access on replace
// snowflake_warehouse='lg_wh'  — Per-model warehouse override
// secure=true                  — Secure views

// Sources (models/staging/_sources.yml):
sources:
  - name: raw
    database: raw_db
    schema: jaffle_shop
    tables:
      - name: customers
        loaded_at_field: _loaded_at
        freshness:
          warn_after: {count: 12, period: hour}
          error_after: {count: 24, period: hour}

// Testing (schema.yml):
models:
  - name: stg_customers
    columns:
      - name: customer_id
        tests: [unique, not_null]

// Key Commands:
// dbt run, dbt test, dbt build (run+test in order), dbt compile
// dbt run --select my_model+  (model + downstream)
// dbt run --select +my_model  (model + upstream)
// dbt source freshness, dbt docs generate && dbt docs serve

// Custom schema macro (macros/generate_schema_name.sql):
{% macro generate_schema_name(custom_schema_name, node) %}
  {% if custom_schema_name %}{{ custom_schema_name | trim }}{% else %}{{ target.schema }}{% endif %}
{% endmacro %}

// Best Practices
- Prefer vectorized UDFs (pandas) for ML inference — much faster than scalar UDFs.
- Pin package versions in production UDFs and stored procedures.
- Use DataFrame API over raw SQL strings in reusable Python pipelines.
- Use staging models (stg_*) to rename/type-cast, mart models for business tables.
- Use incremental for fact tables; dynamic_table for near-real-time.
- Set on_schema_change='sync_all_columns' on incremental models.
- Use copy_grants=true to avoid permission issues. Tag models for selective execution.
- Use separate warehouses for dbt runs vs analyst queries.

// Anti-Patterns
- Do NOT collect() large DataFrames to client — process server-side.
- Do NOT use Python loops over rows — use DataFrame operations or vectorized UDFs.
- Do NOT use {{ this }} without {% if is_incremental() %} guard.
- Do NOT set cluster_by on small tables (< 1TB).
- Do NOT use materialized='table' for everything — views are free.
- Do NOT hardcode database/schema — use {{ ref() }} and {{ source() }}.