May 2026 monthly summary focused on delivering scalable, observable, and test-covered enhancements to DB monitoring for PostgreSQL across two DataDog repos (integrations-core and datadog-agent). Delivered a new per-column statistics capability and a corresponding event-platform routing path, with emphasis on multi-database support, safe opt-in defaults, and strong testing. Key highlights: - Implemented Column Statistics Collector integrated with the dbm-column-stats pipeline in DataDog/integrations-core. The collector queries pg_stats via datadog.column_stats(), streams results with a cursor, and chunks payloads to limit memory usage (5,000 columns per batch). It supports multi-database autodiscovery with per-database error isolation, and includes health events for missing functions/privileges and recovery detection. Also added include/exclude filtering logic, parameterized SQL, and tracing via APM. Default was changed to collect_column_stats.enabled: false to maintain opt-in behavior, with tests updated accordingly. - Renamed collector to column_statistics for clarity and consistency across config keys, symbols, and metrics naming, paired with broader test coverage. - Implemented Event Platform Passthrough for Per-Column Statistics Routing to Metrics Intake in DataDog/datadog-agent. A new dbm-column-statistics passthrough pipeline routes per-column statistics from the Postgres DBM check to dbm-metrics-intake (dbmcolumnstats track). The solution reuses existing database_monitoring.metrics.* endpoint config and applies high-throughput defaults (20 MB, 10 parallel sends, 500-deep channel). End-to-end validation against a local DBM stack showed correct routing and reliable ingestion (200 OK against dbmcolumnstatistics endpoint). Internal measurements indicate column-stats overhead of ~67 MB p99 on small and ~68 MB p99 on medium stacks, with reproducibility across runs. - Strengthened filtering and test coverage. Refactored filters to support parameterized queries, added shared helpers for include/exclude patterns across schemas and column_stats collectors, and expanded unit/integration tests (test_filters.py, test_column_stats.py) to cover include/exclude combinations and edge cases. Updated test_config defaults accordingly. - Business impact: gained granular visibility into per-column statistics across multi-database environments, enabling earlier detection of column-level anomalies, capacity planning signals, and targeted optimizations. The changes also improve reliability and maintainability of the DBM monitoring stack through opt-in defaults, unified filtering helpers, and robust end-to-end tests. Technologies/skills demonstrated: PostgreSQL column statistics (pg_stats), Python (collector implementation, autodiscovery, filtering logic, tests), SQL (parameterized queries and filters), APM tracing, event-platform pipelines and dbm-metrics ingestion, test automation (unit/integration tests), multi-database architecture, performance testing and validation.