|
| 1 | +/** |
| 2 | + * Cluster daemon — leader election via Neon `cc_node_leases`. |
| 3 | + * |
| 4 | + * Mirrors the lease pattern in |
| 5 | + * chittyentity/workers/shared/agent-tasks.ts (`task_leases`): |
| 6 | + * - atomic claim via UPDATE ... RETURNING with expired-lease takeover |
| 7 | + * - heartbeat extends lease_expires_at |
| 8 | + * - explicit release nullifies the holder |
| 9 | + * |
| 10 | + * Schema: see `cc_node_leases` in src/db/schema.ts. |
| 11 | + * |
| 12 | + * @canonical-uri chittycanon://docs/architecture/chittycommand/ADR-001 |
| 13 | + */ |
| 14 | + |
| 15 | +import { neon, type NeonQueryFunction } from '@neondatabase/serverless'; |
| 16 | + |
| 17 | +export interface LeaderEnv { |
| 18 | + DATABASE_URL?: string; |
| 19 | + HYPERDRIVE?: { connectionString: string }; |
| 20 | +} |
| 21 | + |
| 22 | +/** The canonical foundation-PR role. New roles can be added later. */ |
| 23 | +export const META_LEADER_ROLE = 'meta-orchestrator-leader' as const; |
| 24 | + |
| 25 | +export interface NodeLease { |
| 26 | + role: string; |
| 27 | + nodeId: string; |
| 28 | + nodeDescriptor: string | null; |
| 29 | + sessionId: string | null; |
| 30 | + claimedAt: Date; |
| 31 | + heartbeatAt: Date; |
| 32 | + leaseExpiresAt: Date; |
| 33 | + metadata: Record<string, unknown>; |
| 34 | +} |
| 35 | + |
| 36 | +export interface ClaimOptions { |
| 37 | + /** ChittyID of the node attempting to claim (Location type — L). */ |
| 38 | + nodeId: string; |
| 39 | + /** Free-form descriptor for ops (e.g. "chittymini-03"). */ |
| 40 | + nodeDescriptor?: string; |
| 41 | + /** Process/session id for this attempt. */ |
| 42 | + sessionId?: string; |
| 43 | + /** Lease length in seconds. Defaults to 30s. */ |
| 44 | + leaseSeconds?: number; |
| 45 | + /** The role to claim. Defaults to META_LEADER_ROLE. */ |
| 46 | + role?: string; |
| 47 | + /** Optional metadata persisted with the lease. */ |
| 48 | + metadata?: Record<string, unknown>; |
| 49 | +} |
| 50 | + |
| 51 | +function getSql(env: LeaderEnv): NeonQueryFunction<false, false> { |
| 52 | + const conn = env.DATABASE_URL || env.HYPERDRIVE?.connectionString; |
| 53 | + if (!conn) { |
| 54 | + throw new Error('[daemon/leader] No DATABASE_URL or HYPERDRIVE connection string'); |
| 55 | + } |
| 56 | + return neon(conn); |
| 57 | +} |
| 58 | + |
| 59 | +function normalizeLeaseSeconds(input: number | undefined): number { |
| 60 | + if (!input || !Number.isFinite(input) || input <= 0) return 30; |
| 61 | + // Clamp to 1s .. 1h to avoid pathological leases. |
| 62 | + return Math.max(1, Math.min(3600, Math.floor(input))); |
| 63 | +} |
| 64 | + |
| 65 | +/** |
| 66 | + * Atomically claim leadership for `role`. |
| 67 | + * |
| 68 | + * Logic: |
| 69 | + * - Insert the role row if missing (idempotent via ON CONFLICT DO NOTHING). |
| 70 | + * - UPDATE the row to set node_id/sessionId IFF the current holder is the |
| 71 | + * same node (re-claim) OR the lease is unset/expired. |
| 72 | + * - Returns the new lease if the UPDATE affected a row, else null. |
| 73 | + * |
| 74 | + * No other Neon round-trip happens between the conditional SELECT and the |
| 75 | + * UPDATE — the WHERE clause inside UPDATE is itself the gate, so concurrent |
| 76 | + * claimers from different nodes will see exactly one winner. |
| 77 | + */ |
| 78 | +export async function claimLeadership( |
| 79 | + env: LeaderEnv, |
| 80 | + options: ClaimOptions, |
| 81 | +): Promise<NodeLease | null> { |
| 82 | + if (!options.nodeId) throw new Error('[daemon/leader] nodeId is required'); |
| 83 | + |
| 84 | + const sql = getSql(env); |
| 85 | + const role = options.role ?? META_LEADER_ROLE; |
| 86 | + const leaseSeconds = normalizeLeaseSeconds(options.leaseSeconds); |
| 87 | + const sessionId = options.sessionId ?? null; |
| 88 | + const descriptor = options.nodeDescriptor ?? null; |
| 89 | + const metadata = JSON.stringify(options.metadata ?? {}); |
| 90 | + |
| 91 | + // 1. Ensure a row exists for this role. Idempotent. |
| 92 | + await sql` |
| 93 | + INSERT INTO cc_node_leases (role, metadata) |
| 94 | + VALUES (${role}, ${metadata}::jsonb) |
| 95 | + ON CONFLICT (role) DO NOTHING`; |
| 96 | + |
| 97 | + // 2. Atomic claim. |
| 98 | + const rows = await sql` |
| 99 | + UPDATE cc_node_leases |
| 100 | + SET node_id = ${options.nodeId}, |
| 101 | + node_descriptor = ${descriptor}, |
| 102 | + session_id = ${sessionId}, |
| 103 | + claimed_at = COALESCE(claimed_at, NOW()), |
| 104 | + heartbeat_at = NOW(), |
| 105 | + lease_expires_at = NOW() + (${leaseSeconds} * INTERVAL '1 second'), |
| 106 | + metadata = ${metadata}::jsonb, |
| 107 | + updated_at = NOW() |
| 108 | + WHERE role = ${role} |
| 109 | + AND ( |
| 110 | + node_id IS NULL |
| 111 | + OR node_id = ${options.nodeId} |
| 112 | + OR lease_expires_at IS NULL |
| 113 | + OR lease_expires_at < NOW() |
| 114 | + ) |
| 115 | + RETURNING *`; |
| 116 | + |
| 117 | + if (rows.length === 0) return null; |
| 118 | + return rowToLease(rows[0]); |
| 119 | +} |
| 120 | + |
| 121 | +/** |
| 122 | + * Extend the lease. Returns null if this node is no longer the holder |
| 123 | + * (another node took over). |
| 124 | + */ |
| 125 | +export async function heartbeat( |
| 126 | + env: LeaderEnv, |
| 127 | + nodeId: string, |
| 128 | + options: { role?: string; leaseSeconds?: number } = {}, |
| 129 | +): Promise<NodeLease | null> { |
| 130 | + if (!nodeId) throw new Error('[daemon/leader] nodeId is required for heartbeat'); |
| 131 | + const sql = getSql(env); |
| 132 | + const role = options.role ?? META_LEADER_ROLE; |
| 133 | + const leaseSeconds = normalizeLeaseSeconds(options.leaseSeconds); |
| 134 | + |
| 135 | + const rows = await sql` |
| 136 | + UPDATE cc_node_leases |
| 137 | + SET heartbeat_at = NOW(), |
| 138 | + lease_expires_at = NOW() + (${leaseSeconds} * INTERVAL '1 second'), |
| 139 | + updated_at = NOW() |
| 140 | + WHERE role = ${role} AND node_id = ${nodeId} |
| 141 | + RETURNING *`; |
| 142 | + return rows[0] ? rowToLease(rows[0]) : null; |
| 143 | +} |
| 144 | + |
| 145 | +/** |
| 146 | + * Release leadership. Only this node can release — if a different node |
| 147 | + * holds the role, this is a no-op (returns false). |
| 148 | + */ |
| 149 | +export async function releaseLeadership( |
| 150 | + env: LeaderEnv, |
| 151 | + nodeId: string, |
| 152 | + options: { role?: string } = {}, |
| 153 | +): Promise<boolean> { |
| 154 | + if (!nodeId) throw new Error('[daemon/leader] nodeId is required for release'); |
| 155 | + const sql = getSql(env); |
| 156 | + const role = options.role ?? META_LEADER_ROLE; |
| 157 | + const rows = await sql` |
| 158 | + UPDATE cc_node_leases |
| 159 | + SET node_id = NULL, |
| 160 | + session_id = NULL, |
| 161 | + node_descriptor = NULL, |
| 162 | + claimed_at = NULL, |
| 163 | + heartbeat_at = NULL, |
| 164 | + lease_expires_at = NULL, |
| 165 | + updated_at = NOW() |
| 166 | + WHERE role = ${role} AND node_id = ${nodeId} |
| 167 | + RETURNING role`; |
| 168 | + return rows.length > 0; |
| 169 | +} |
| 170 | + |
| 171 | +/** |
| 172 | + * Inspect the current lease row (no mutation). Useful for diagnostics. |
| 173 | + */ |
| 174 | +export async function describeLease( |
| 175 | + env: LeaderEnv, |
| 176 | + options: { role?: string } = {}, |
| 177 | +): Promise<NodeLease | null> { |
| 178 | + const sql = getSql(env); |
| 179 | + const role = options.role ?? META_LEADER_ROLE; |
| 180 | + const rows = await sql`SELECT * FROM cc_node_leases WHERE role = ${role} LIMIT 1`; |
| 181 | + if (!rows[0] || !rows[0].node_id) return null; |
| 182 | + return rowToLease(rows[0]); |
| 183 | +} |
| 184 | + |
| 185 | +function rowToLease(row: Record<string, unknown>): NodeLease { |
| 186 | + return { |
| 187 | + role: String(row.role), |
| 188 | + nodeId: String(row.node_id), |
| 189 | + nodeDescriptor: (row.node_descriptor as string) ?? null, |
| 190 | + sessionId: (row.session_id as string) ?? null, |
| 191 | + claimedAt: new Date(row.claimed_at as string), |
| 192 | + heartbeatAt: new Date(row.heartbeat_at as string), |
| 193 | + leaseExpiresAt: new Date(row.lease_expires_at as string), |
| 194 | + metadata: (row.metadata as Record<string, unknown>) ?? {}, |
| 195 | + }; |
| 196 | +} |
0 commit comments