erikdarlingdata/PerformanceMonitor v3.13.0 on GitHub

See CHANGELOG.md for the summary, and docs/changelog/3.13.md for the full entries.

Darling: the first start after the upgrade runs store migrations V230 to V266 (schema 229 to 266). No step needs an offline window. Two steps build an index on the collection log and on the oversized plan backlog. On a store where those tables hold many rows, the first start takes longer.
Lite: the DuckDB schema moves from 78 to 81 on the first start. Each step adds one column, and rows stored before the upgrade read NULL in it.
Self-managed stores: if you set up your own PostgreSQL store, start the service on 3.13.0 first, then run Darling/tools/provision-roles.sql again.
Docker: the Darling container now runs as user 1654, not root. A new compose install needs nothing extra. An install that ran as root needs two commands before the new image starts. The 3.13 changelog gives them under Important.
Alert paging: several alerts now page less. High CPU now waits for 5 samples in a row. Blocking needs 3 events with a 15 second wait, or one event that waits 60 seconds. Alerts held back are kept in the alert history and the daily digest.

SQL Server targets with CDC: the query snapshots collector can read msdb.dbo.cdc_jobs, so it leaves CDC capture jobs out of the Long-Running Query alert. Without the grant, it matches capture jobs by name instead. The grant block in docs/permissions.md includes it, guarded for servers that have never used CDC.
Self-hosted PostgreSQL targets: a grant of EXECUTE on pg_read_binary_file(text, bigint, bigint) lets the three server log collectors use the binary read. Without it, they keep the current read.

See docs/permissions.md for the full list.