next-tjwater-server/resources/sql/015_workbench_events_and_task_runs.sql

-- Durable workbench event log and independent Agent execution runs.

BEGIN;

CREATE SCHEMA IF NOT EXISTS workbench;

CREATE TABLE IF NOT EXISTS workbench.domain_event_log (
    event_seq bigserial PRIMARY KEY,
    project_id uuid NOT NULL,
    event_id uuid NOT NULL,
    topic text NOT NULL CHECK (topic IN (
        'operational.task.changed',
        'operational.run.queued',
        'operational.run.changed',
        'operational.report.published',
        'operational.action.changed',
        'operational.notification.changed',
        'workbench.alert.changed',
        'workbench.work_order.changed',
        'analysis.run.changed'
    )),
    aggregate_type text NOT NULL CHECK (length(aggregate_type) > 0),
    aggregate_id text NOT NULL CHECK (length(aggregate_id) > 0),
    aggregate_revision bigint NOT NULL CHECK (aggregate_revision >= 1),
    schema_version integer NOT NULL DEFAULT 1 CHECK (schema_version = 1),
    occurred_at timestamptz NOT NULL DEFAULT now(),
    metadata jsonb NOT NULL DEFAULT '{}'::jsonb,
    UNIQUE (project_id,event_id)
);

CREATE INDEX IF NOT EXISTS domain_event_log_project_cursor_idx
    ON workbench.domain_event_log(project_id,event_seq);
CREATE INDEX IF NOT EXISTS domain_event_log_retention_idx
    ON workbench.domain_event_log(occurred_at,event_seq);

ALTER TABLE IF EXISTS analysis.runs
    ADD COLUMN IF NOT EXISTS revision bigint NOT NULL DEFAULT 1
    CHECK (revision >= 1);

CREATE TABLE IF NOT EXISTS workbench.operational_task_runs (
    project_id uuid NOT NULL,
    run_id uuid NOT NULL,
    task_id uuid NOT NULL,
    generation bigint NOT NULL CHECK (generation >= 1),
    execution_request_id uuid NOT NULL,
    status text NOT NULL CHECK (status IN (
        'queued','claimed','running','succeeded','failed','timed_out','cancelled'
    )),
    revision bigint NOT NULL CHECK (revision >= 1),
    agent_session_id text,
    lease_expires_at timestamptz,
    heartbeat_at timestamptz,
    started_at timestamptz,
    finished_at timestamptz,
    failure_code text,
    failure_detail text,
    dispatch_attempts integer NOT NULL DEFAULT 0 CHECK (dispatch_attempts >= 0),
    next_dispatch_at timestamptz NOT NULL DEFAULT now(),
    created_at timestamptz NOT NULL,
    updated_at timestamptz NOT NULL,
    PRIMARY KEY (project_id,run_id),
    UNIQUE (project_id,task_id,generation),
    UNIQUE (project_id,execution_request_id),
    FOREIGN KEY (project_id,task_id)
        REFERENCES workbench.operational_tasks(project_id,task_id) ON DELETE CASCADE
);

CREATE UNIQUE INDEX IF NOT EXISTS operational_task_runs_one_active_idx
    ON workbench.operational_task_runs(project_id,task_id)
    WHERE status IN ('queued','claimed','running');
CREATE INDEX IF NOT EXISTS operational_task_runs_queue_idx
    ON workbench.operational_task_runs(
        project_id,next_dispatch_at,created_at,run_id
    ) WHERE status='queued';
CREATE INDEX IF NOT EXISTS operational_task_runs_lease_idx
    ON workbench.operational_task_runs(project_id,lease_expires_at,run_id)
    WHERE status IN ('claimed','running') AND lease_expires_at IS NOT NULL;

ALTER TABLE workbench.operational_tasks
    ADD COLUMN IF NOT EXISTS lifecycle_status text;

UPDATE workbench.operational_tasks
SET lifecycle_status = CASE status
    WHEN 'waiting_human' THEN 'waiting_human'
    WHEN 'completed' THEN 'completed'
    WHEN 'failed' THEN 'failed'
    WHEN 'cancelled' THEN 'cancelled'
    ELSE 'open'
END
WHERE lifecycle_status IS NULL;

UPDATE workbench.operational_tasks
SET document = document || jsonb_build_object('lifecycle_status', lifecycle_status)
WHERE document->>'lifecycle_status' IS DISTINCT FROM lifecycle_status;

ALTER TABLE workbench.operational_tasks
    ALTER COLUMN lifecycle_status SET NOT NULL;

DO $$
BEGIN
    IF NOT EXISTS (
        SELECT 1 FROM pg_constraint
        WHERE conrelid='workbench.operational_tasks'::regclass
          AND conname='operational_tasks_lifecycle_status_check'
    ) THEN
        ALTER TABLE workbench.operational_tasks
            ADD CONSTRAINT operational_tasks_lifecycle_status_check
            CHECK (lifecycle_status IN
                ('open','waiting_human','completed','failed','cancelled'));
    END IF;
END $$;

-- Deterministically preserve the currently active or last known execution as
-- generation history. uuid values are derived without requiring an extension.
INSERT INTO workbench.operational_task_runs (
    project_id,run_id,task_id,generation,execution_request_id,status,revision,
    agent_session_id,lease_expires_at,heartbeat_at,started_at,finished_at,
    failure_code,failure_detail,dispatch_attempts,next_dispatch_at,created_at,updated_at
)
SELECT
    project_id,
    (substr(md5(project_id::text || ':' || task_id::text || ':' || execution_generation::text),1,8)
      || '-' || substr(md5(project_id::text || ':' || task_id::text || ':' || execution_generation::text),9,4)
      || '-5' || substr(md5(project_id::text || ':' || task_id::text || ':' || execution_generation::text),14,3)
      || '-a' || substr(md5(project_id::text || ':' || task_id::text || ':' || execution_generation::text),18,3)
      || '-' || substr(md5(project_id::text || ':' || task_id::text || ':' || execution_generation::text),21,12))::uuid,
    task_id,
    execution_generation,
    execution_request_id,
    CASE status
        WHEN 'queued' THEN 'queued'
        WHEN 'running' THEN 'running'
        WHEN 'waiting_human' THEN 'succeeded'
        WHEN 'completed' THEN 'succeeded'
        WHEN 'failed' THEN 'failed'
        WHEN 'cancelled' THEN 'cancelled'
    END,
    GREATEST(revision,1),
    agent_session_id,
    agent_lease_expires_at,
    CASE WHEN status='running' THEN updated_at ELSE NULL END,
    CASE WHEN status IN ('running','waiting_human','completed','failed','cancelled')
        THEN created_at ELSE NULL END,
    CASE WHEN status IN ('waiting_human','completed','failed','cancelled')
        THEN updated_at ELSE NULL END,
    CASE WHEN status='failed' THEN 'MIGRATED_FAILURE' ELSE NULL END,
    NULL,
    0,
    created_at,
    created_at,
    updated_at
FROM workbench.operational_tasks
ON CONFLICT (project_id,task_id,generation) DO NOTHING;

COMMIT;