Postgres metrics refreshes skip the legacy runs column

Agno's calculate_metrics() on PostgresDb and AsyncPostgresDb now reads only each run's model and model_provider, so a refresh leaves the legacy sessions.runs column in Postgres. The runs migration keeps that column as a backup until someone calls cleanup_legacy_runs_column(), and earlier releases selected it on every refresh. Every stored run came along, messages and tool calls included, only to be thrown away because the runs table already held the same data.

That hurt most on the first refresh after migrating, which covers the full history and also fires on its own when the metrics page opens. In one reported case, 51 GB of legacy runs ran an 8 GB container out of memory. An earlier metrics fix dealt with bind-parameter limits and the migration's own memory; this one shrinks what the refresh reads.

Runs that never reached the runs table still count. Postgres expands the legacy column with jsonb_array_elements inside the query and sends back two fields per run.

A script seeded an embedded Postgres with 1,500 sessions of 20 runs each, ran the migration, kept the legacy column in place, then called calculate_metrics() in a fresh process and measured peak memory growth.

The sessions query shows where the memory went.

# real output, first sessions query of the refresh, captured with a SQLAlchemy event
 
Before
SELECT demo.agno_sessions.session_id, demo.agno_sessions.user_id, demo.agno_sessions.session_data, demo.agno_sessions.created_at, demo.agno_sessions.session_type, demo.agno_sessions.runs
 
Now
SELECT demo.agno_sessions.session_id, demo.agno_sessions.user_id, demo.agno_sessions.session_data, demo.agno_sessions.created_at, demo.agno_sessions.session_type

See the cookbook, and learn more about refreshing AgentOS metrics and migrating your Agno database in the documentation.