flatrender/backend/db/migrations/11_render_nodes.sql

-- =====================================================================
-- RENDER SCHEMA — Part 1: Farm Nodes & Health
-- =====================================================================

SET search_path TO render, public;

CREATE TYPE node_status      AS ENUM ('Ready','Busy','Offline','Maintenance','Crashed','Updating','Disabled');
CREATE TYPE node_kind        AS ENUM ('Shared','Dedicated','Spot');
CREATE TYPE render_type      AS ENUM ('Free','Paid','Snapshot','Mockup');
CREATE TYPE ae_version       AS ENUM ('2020','2021','2022','2023','2024','2025');

-- ---------------------------------------------------------------------
-- render_nodes — registry of farm machines
-- ---------------------------------------------------------------------
CREATE TABLE render_nodes (
    id                          UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    name                        TEXT NOT NULL,
    region                      TEXT NOT NULL,                          -- 'tehran','frankfurt',...

    -- Network
    node_ip                     INET NOT NULL,
    worker_port                 INT NOT NULL DEFAULT 5555,
    public_endpoint             TEXT,

    -- Spec
    ram_gb                      INT,
    cpu_cores                   INT,
    gpu_model                   TEXT,
    storage_gb                  INT,

    -- Software
    current_ae_version          ae_version NOT NULL,
    available_ae_versions       TEXT[] NOT NULL DEFAULT '{}',
    node_agent_version          TEXT,
    last_update_at              TIMESTAMPTZ,
    last_update_error           TEXT,

    -- Ownership
    node_kind                   node_kind NOT NULL DEFAULT 'Shared',
    owner_user_id               UUID,                                   -- references identity.users (for Dedicated)
    owner_tenant_id             UUID,                                   -- references identity.tenants

    -- State
    status                      node_status NOT NULL DEFAULT 'Offline',
    current_job_id              UUID,                                   -- references render_jobs
    current_frame_job_id        UUID,                                   -- references frame_jobs
    job_started_at              TIMESTAMPTZ,
    render_type                 render_type,                            -- which queue it's serving now

    -- Health (denormalized for hot reads)
    last_heartbeat_at           TIMESTAMPTZ,
    last_cpu_pct                INT,
    last_ram_available_mb       INT,
    ae_running                  BOOLEAN NOT NULL DEFAULT FALSE,

    -- Stats
    lifetime_task_count         BIGINT NOT NULL DEFAULT 0,
    lifetime_crash_count        INT NOT NULL DEFAULT 0,
    consecutive_failures        INT NOT NULL DEFAULT 0,

    -- Scheduling
    priority                    INT NOT NULL DEFAULT 100,
    is_active                   BOOLEAN NOT NULL DEFAULT TRUE,
    accepts_new_jobs            BOOLEAN NOT NULL DEFAULT TRUE,

    -- Maintenance
    last_maintenance_at         TIMESTAMPTZ,
    next_maintenance_at         TIMESTAMPTZ,
    maintenance_reason          TEXT,

    -- Local cache state (templates the node has downloaded)
    cached_template_md5s        TEXT[] NOT NULL DEFAULT '{}',
    cache_used_gb               INT NOT NULL DEFAULT 0,

    metadata                    JSONB NOT NULL DEFAULT '{}'::jsonb,
    created_at                  TIMESTAMPTZ NOT NULL DEFAULT NOW(),
    updated_at                  TIMESTAMPTZ NOT NULL DEFAULT NOW()
);

CREATE INDEX idx_nodes_region_status ON render_nodes(region, status) WHERE is_active = TRUE;
CREATE INDEX idx_nodes_ready         ON render_nodes(region, priority DESC)
       WHERE status = 'Ready' AND accepts_new_jobs = TRUE AND is_active = TRUE;
CREATE INDEX idx_nodes_owner         ON render_nodes(owner_user_id) WHERE node_kind = 'Dedicated';
CREATE INDEX idx_nodes_heartbeat     ON render_nodes(last_heartbeat_at) WHERE is_active = TRUE;
CREATE UNIQUE INDEX uq_nodes_ip_port ON render_nodes(node_ip, worker_port);

CREATE TRIGGER tg_render_nodes_updated_at
BEFORE UPDATE ON render_nodes FOR EACH ROW EXECUTE FUNCTION public.tg_set_updated_at();

-- ---------------------------------------------------------------------
-- node_health_logs — historical heartbeat data (partitioned monthly)
-- ---------------------------------------------------------------------
CREATE TABLE node_health_logs (
    id                  BIGSERIAL,
    node_id             UUID NOT NULL,
    recorded_at         TIMESTAMPTZ NOT NULL DEFAULT NOW(),

    status              node_status NOT NULL,
    cpu_pct             INT,
    ram_available_mb    INT,
    ae_running          BOOLEAN,
    current_job_id      UUID,
    current_frame       INT,

    -- Templates cached (size summary only)
    cache_used_gb       INT,

    PRIMARY KEY (id, recorded_at)
) PARTITION BY RANGE (recorded_at);

CREATE INDEX idx_node_health_node ON node_health_logs(node_id, recorded_at DESC);

CREATE TABLE node_health_logs_y2026m01
    PARTITION OF node_health_logs
    FOR VALUES FROM ('2026-01-01') TO ('2026-02-01');

-- ---------------------------------------------------------------------
-- node_crashes — every detected AE crash
-- ---------------------------------------------------------------------
CREATE TABLE node_crashes (
    id                  UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    node_id             UUID NOT NULL REFERENCES render_nodes(id) ON DELETE CASCADE,
    render_job_id       UUID,                       -- which job was running
    frame_job_id        UUID,

    crashed_at          TIMESTAMPTZ NOT NULL DEFAULT NOW(),
    last_known_frame    INT,
    crash_signal        TEXT,                       -- exit code or signal
    error_log           TEXT,                       -- last N lines of AE log
    log_file_url        TEXT,                       -- MinIO upload of full log

    -- Recovery
    auto_recovered      BOOLEAN NOT NULL DEFAULT FALSE,
    recovery_action     TEXT,                       -- 'reset_prefs','restart_ae','reassign_job'
    recovered_at        TIMESTAMPTZ,

    created_at          TIMESTAMPTZ NOT NULL DEFAULT NOW()
);

CREATE INDEX idx_node_crashes_node ON node_crashes(node_id, crashed_at DESC);
CREATE INDEX idx_node_crashes_job  ON node_crashes(render_job_id) WHERE render_job_id IS NOT NULL;

-- ---------------------------------------------------------------------
-- node_updates — software/AE update tracking
-- ---------------------------------------------------------------------
CREATE TABLE node_updates (
    id                          UUID PRIMARY KEY DEFAULT gen_random_uuid(),
    update_file_name            TEXT NOT NULL,
    update_number               INT NOT NULL,
    description                 TEXT,
    target_ae_version           ae_version,
    in_update_queue             BOOLEAN NOT NULL DEFAULT FALSE,
    rolled_out_to_node_ids      UUID[] NOT NULL DEFAULT '{}',
    last_update_queue_date      TIMESTAMPTZ,
    create_date                 TIMESTAMPTZ NOT NULL DEFAULT NOW(),
    created_at                  TIMESTAMPTZ NOT NULL DEFAULT NOW()
);

-- ---------------------------------------------------------------------
-- node_template_cache — what's currently cached on each node
-- ---------------------------------------------------------------------
CREATE TABLE node_template_cache (
    id                  BIGSERIAL PRIMARY KEY,
    node_id             UUID NOT NULL REFERENCES render_nodes(id) ON DELETE CASCADE,
    project_id          UUID NOT NULL,                      -- references content.projects
    aep_file_md5        TEXT NOT NULL,
    file_size_bytes     BIGINT NOT NULL,
    cached_at           TIMESTAMPTZ NOT NULL DEFAULT NOW(),
    last_used_at        TIMESTAMPTZ NOT NULL DEFAULT NOW(),
    use_count           INT NOT NULL DEFAULT 1,
    local_path          TEXT NOT NULL,
    UNIQUE (node_id, aep_file_md5)
);

CREATE INDEX idx_node_cache_node_lru ON node_template_cache(node_id, last_used_at);
CREATE INDEX idx_node_cache_md5      ON node_template_cache(aep_file_md5);