Connecting To Data Source

Connect to Data Source

Register an external data source with AWS Glue so downstream skills (ingesting-into-data-lake) can move data from it. A Glue connection stores the network config, driver, and credential reference for one source. Create once per source, reuse across jobs.

Philosophy

A connection is a named pipe, not a pipeline. This skill produces a tested, reusable Glue connection. It does not move data.

Common Tasks

You MUST execute commands using AWS MCP server tools when connected – they provide validation, sandboxed execution, and audit logging. Fall back to AWS CLI only if MCP is unavailable. You MUST explain each step before executing.

Workflow

1. Verify Dependencies and Context

• You MUST check whether AWS MCP tools or AWS CLI are available and inform the user if missing
• You MUST confirm target AWS region and verify credentials with aws sts get-caller-identity

2. Classify the Source

Ask the user which source type they want to connect to, or infer from hints:

If the user names DynamoDB or a local file, stop and tell them: DynamoDB is read directly by Glue without a connection, and local files belong in the ingesting-into-data-lake skill’s local-upload workflow.

3. Gather Connection Hints from the User

You MUST ask for hints the user can provide – do not guess.

For all sources:

• Desired connection name (lowercase, hyphens: oracle-prod-sales, snowflake-analytics)
• Existing Secrets Manager secret, or create one
• Is source reachable from a Glue VPC (same, peered, VPN, Direct Connect)

JDBC: hostname/endpoint, port, database, whether RDS/Aurora/self-managed, IAM DB auth enabled (Aurora/RDS MySQL/Postgres), SSL required.

Snowflake: account identifier, warehouse, role, default database, auth (password, key-pair, OAuth).

BigQuery: GCP project ID, location, whether service account JSON is provisioned.

4. Discover Existing Connections and Candidate Sources

Check what exists before creating.

Existing Glue connections:

aws glue get-connections --filter ConnectionType=<TYPE> --region <REGION>

If a suitable one exists, confirm and skip to Step 7.

Candidate sources in account (JDBC/Redshift only):

• RDS: aws rds describe-db-instances
• Aurora: aws rds describe-db-clusters
• Redshift: aws redshift describe-clusters

Present candidates to user; let them pick. See discovery.md (opens in a new tab).

5. Register Credentials

You MUST encourage AWS Secrets Manager over plaintext passwords. You SHOULD prefer IAM database authentication where supported (Aurora/RDS MySQL and PostgreSQL, Redshift). See credential-security.md (opens in a new tab).

• You MUST confirm with user before creating a new Secrets Manager secret
• You MUST NOT write plaintext credentials into chat or logs
• For IAM DB auth, no secret is needed

6. Create the Glue Connection

Follow the source-specific reference for connection properties:

aws glue create-connection --connection-input '<JSON>' --region <REGION>

Private sources require PhysicalConnectionRequirements (SubnetId, SecurityGroupIdList, AvailabilityZone). See network-setup.md (opens in a new tab).

7. Test the Connection

You MUST test before handing off. Testing is two-phase: a quick API check, then an engine-level verification.

Phase A: Glue TestConnection (network and credential sanity check)

aws glue test-connection --connection-name <NAME> --region <REGION>

This validates that Glue can reach the source and authenticate. It does NOT prove the connection works end-to-end with the query engine the user plans to use.

Phase B: Engine-level verification

After TestConnection passes, verify the connection works with the user’s intended engine by running a minimal query through it:

• Glue ETL (default): Run a smoke-test Glue job that reads one row via the connection. See troubleshooting.md (opens in a new tab).
• Athena: If the user plans to query via Athena with a federated connector, run a SELECT 1 through the Athena connection to confirm the Lambda-based connector can reach the source.
• Glue Crawler: If the user plans to crawl the source, run a test crawl on a single table.

Phase B catches issues that TestConnection misses: driver compatibility at job runtime, catalog configuration, Spark-level serialization, and engine-specific auth flows (e.g., Snowflake SNOWFLAKE type works in ETL but not via JDBC crawlers).

On success in both phases, tell user the connection name is ready for ingesting-into-data-lake. On failure in either phase, Step 8.

8. Troubleshoot (only if test failed)

Diagnose in order: network, credentials, driver. See troubleshooting.md (opens in a new tab).

Constraints:

• You MUST check VPC routing, security groups, and S3 VPC endpoint before blaming credentials
• You MUST verify Glue role can read the Secrets Manager secret
• You MUST NOT rotate credentials without user confirmation

Argument Routing

• No args: Walk through Steps 1-7 interactively
• Source type keyword (e.g., snowflake, oracle): Skip to Step 2 with the type prefilled
• Existing connection name: Skip to Step 7 (test) then Step 8 if failing
• Hostname or RDS endpoint: Skip to Step 4 with the candidate prefilled

Gotchas

• Glue’s SNOWFLAKE connection type is distinct from JDBC configured for Snowflake. You MUST use SNOWFLAKE for Spark ETL jobs; do not use JDBC.
• Connection names are immutable. Choose carefully.
• PhysicalConnectionRequirements.AvailabilityZone MUST match the subnet’s AZ or the connection fails at job runtime, not creation time.
• IAM database authentication tokens expire in 15 minutes. The Glue job generates a fresh token on each connection; do not cache.
• An S3 VPC gateway endpoint MUST exist in the VPC used by private-source connections. Without it, Glue jobs cannot read their scripts or write results to S3.

Troubleshooting

References

• jdbc-setup.md (opens in a new tab) – Oracle, SQL Server, PostgreSQL, MySQL, RDS, Redshift
• snowflake-setup.md (opens in a new tab) – Glue SNOWFLAKE type, auth modes
• bigquery-setup.md (opens in a new tab) – Glue BIGQUERY type, GCP service accounts
• discovery.md (opens in a new tab) – Finding existing connections and candidate sources
• credential-security.md (opens in a new tab) – Secrets Manager and IAM DB auth
• network-setup.md (opens in a new tab) – VPC, subnets, security groups, endpoints
• troubleshooting.md (opens in a new tab) – Connection errors and diagnostic flow