fix(redshift-mcp-server): handle missing DBName field in discover_clusters

[abhu85]

Summary

• Fixes KeyError: 'DBName' when discover_clusters() processes clusters without the DBName field present
• Uses safe dictionary access via .get('DBName', 'dev') with default to 'dev' (Redshift's default database name)
• Adds regression test to verify clusters without DBName are handled gracefully

Problem

The describe_clusters API may return cluster metadata without a DBName field in certain scenarios (e.g., clusters created via certain methods, or clusters in specific states). The current implementation used direct dictionary access cluster['DBName'] which raises KeyError when the field is missing.

Per AWS API Reference, DBName is marked as "Required: No".

Root Cause

Line 417 in redshift.py:

'database_name': cluster['DBName'],  # Direct access - raises KeyError if missing

Other fields in the same block already used safe access patterns (.get()), but DBName was missed.

Fix

Changed to safe dictionary access with default value:

'database_name': cluster.get('DBName', 'dev'),  # Returns 'dev' if missing

This matches the existing Serverless workgroup behavior which already defaults to 'dev'.

Tests Run

All 13 discover tests pass, including new test for missing DBName scenario.

Fixes #2331

---

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of the project license.

[abhu85]

Thanks @grayhemp for the guidance!

I've updated the PR to default to 'dev' instead of None:

'database_name': cluster.get('DBName', 'dev'),

This matches the existing Serverless workgroup behavior (lines 454-456) which already defaults to 'dev'.

Regarding Serverless - I checked and it already handles the missing database name case correctly via:

'database_name': workgroup_detail.get('configParameters', [{}])[0].get('parameterValue', 'dev'),

So only the provisioned cluster path needed this fix.

[grayhemp]

I've updated the PR to default to 'dev' instead of None:
...
This matches the existing Serverless workgroup behavior (lines 454-456) which already defaults to 'dev'.

Indeed, awesome. Thank you very much.

[grayhemp]

Additionally, could you please run E2E tests and post the results, similarly to how it is done here #1927 (see the description)? You might need to modify the prompts to exclude the PRs specific part as it's not merged yet. Thanks again.

[abhu85]

@grayhemp Thanks for the feedback! I've refactored the test as suggested:

• Modified test_discover_clusters_provisioned() to return 2 mock entries
• First entry (full_cluster) extends second entry (minimal_cluster) to avoid code duplication
• Minimal entry has DBName omitted to test the .get('DBName', 'dev') fix
• Removed the separate test_discover_clusters_provisioned_missing_dbname test method

All tests pass locally:

tests/test_redshift.py::TestDiscoverFunctions::test_discover_clusters_provisioned PASSED
tests/test_redshift.py::TestDiscoverFunctions::test_discover_clusters_provisioned_error PASSED
tests/test_redshift.py::TestDiscoverFunctions::test_discover_clusters_serverless PASSED
tests/test_redshift.py::TestDiscoverFunctions::test_discover_clusters_serverless_error PASSED

Regarding E2E tests - I don't have a Redshift cluster available for testing. Could you run E2E tests on your end, or is there a test cluster I could use?

[grayhemp]

kiro-cli chat \
    --model claude-opus-4.6-1m \
    --no-interactive \
    --trust-tools "@{awslabs.redshift-mcp-server}/*,execute_bash,fs_read" \
'Use Redshift MCP Server to run end-to-end (integration) tests covering the scenarios for the changes introduced in the current branch only. Get the test scenario ideas from the unit tests under the project directory. Provide a short testing summary, one line per scenario.'

Change Specific Integration Test Summary

#	Scenario	Result
1	Cluster discovery returns provisioned clusters with all fields populated (including database_name)	✅ PASS
2	Cluster discovery returns serverless workgroups alongside provisioned clusters	✅ PASS
3	database_name field is always present — no KeyError when DBName key is missing from API response (core fix for #2331)	✅ PASS
4	Optional fields (tags, endpoint, port, node_type, number_of_nodes) are handled gracefully with defaults	✅ PASS
5	Downstream list_databases works using the discovered database_name on both provisioned and serverless clusters	✅ PASS
6	Downstream list_schemas works using the discovered database on a provisioned cluster	✅ PASS
7	End-to-end query execution succeeds using the discovered database_name ("dev") from cluster discovery	✅ PASS

All 7 scenarios passed. The .get('DBName', 'dev') fix correctly defaults database_name to
"dev" when the key is absent, and the full discovery-to-query pipeline works without
errors.

[grayhemp]

kiro-cli chat \
    --model claude-opus-4.6-1m \
    --no-interactive \
    --trust-tools "@{awslabs.redshift-mcp-server}/*,execute_bash,fs_read" \
'Use Redshift MCP Server to run a complete set of end-to-end (integration) tests covering all its tools. Check both provisioned and serverless clusters, including the database schema exploration in both. Check the SQL read-only protection, transaction breaker protection, failed user SQL behavior. Get the test scenario ideas from the unit tests under the project directory. Provide a short testing summary, one line per tool.'

Complete Integration Test Summary

Tool	Result	Details
list_clusters	✅ PASS	Discovered 2 provisioned clusters + 1 serverless workgroup with correct metadata (type, status, endpoint, node_type, encryption)
list_databases (provisioned)	✅ PASS	Returned dev and sample_data_dev databases with owner, type, and isolation level on redshift-cluster-analytics
list_databases (serverless)	✅ PASS	Returned same databases on redshift-workgroup-etl serverless workgroup
list_schemas (provisioned)	✅ PASS	Returned 4 schemas (information_schema, pg_catalog, public, tickit) with correct ACLs on provisioned cluster
list_schemas (serverless)	✅ PASS	Returned same 4 schemas on serverless workgroup with IAM role-based ACLs
list_tables (provisioned)	✅ PASS	Returned 7 tickit tables (category, date, event, listing, sales, users, venue) with TABLE type on provisioned
list_tables (serverless)	✅ PASS	Returned identical 7 tables on serverless workgroup
list_columns (provisioned)	✅ PASS	Returned 18 columns for users table with correct data types (integer, character, boolean), ordinal positions, and nullability
list_columns (serverless)	✅ PASS	Returned identical 18 columns on serverless workgroup
execute_query (provisioned SELECT)	✅ PASS	Returned 3 rows with columns, row_count, execution_time_ms, and query_id on provisioned cluster
execute_query (serverless SELECT)	✅ PASS	Returned 3 rows with same structure on serverless workgroup
execute_query (read-only protection)	✅ PASS	INSERT blocked with "transaction is read-only" — BEGIN READ ONLY wrapper works
execute_query (transaction breaker: END)	✅ PASS	END; DROP TABLE rejected with "suspicious pattern" before reaching Redshift
execute_query (transaction breaker: COMMIT)	✅ PASS	COMMIT; INSERT rejected with "suspicious pattern" before reaching Redshift
execute_query (transaction breaker: ABORT)	✅ PASS	ABORT work; SELECT rejected with "suspicious pattern"
execute_query (transaction breaker: ROLLBACK)	✅ PASS	ROLLBACK TRANSACTION; SELECT rejected with "suspicious pattern"
execute_query (failed SQL: syntax error)	✅ PASS	Returns clear Redshift error "syntax error at or near" — transaction properly closed via END
execute_query (failed SQL: missing table)	✅ PASS	Returns "relation does not exist" — transaction properly closed, no orphaned sessions
Data integrity verification	✅ PASS	49,990 users confirmed intact after all attack/error tests — no data was modified

All 19 test scenarios passed across both provisioned and serverless cluster types. The three protection layers (read-only transactions, transaction breaker regex, graceful error handling with END cleanup) all function correctly in production.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 91.29%. Comparing base (6fda1e9) to head (8dec995).
⚠️ Report is 22 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #2685      +/-   ##
==========================================
- Coverage   91.30%   91.29%   -0.02%     
==========================================
  Files        1001     1001              
  Lines       73725    73725              
  Branches    11880    11880              
==========================================
- Hits        67316    67305      -11     
- Misses       3959     3966       +7     
- Partials     2450     2454       +4     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[grayhemp]

@abhu85 Thank you. I ran E2E tests and they all look good. Let me get the second reviewer/approval.

[scottschreckengaust]

LGTM