|
| 1 | +import { readFile } from 'node:fs/promises'; |
| 2 | +import { loadavg } from 'node:os'; |
| 3 | +import { z } from 'zod'; |
| 4 | +import type { OpenclawVersionInfo } from './openclaw-version'; |
| 5 | +import { CONTROLLER_COMMIT, CONTROLLER_VERSION } from './version'; |
| 6 | +import type { SupervisorStats } from './supervisor'; |
| 7 | + |
| 8 | +const CHECKIN_INTERVAL_MS = 5 * 60 * 1000; |
| 9 | +const INITIAL_DELAY_MS = 2 * 60 * 1000; |
| 10 | +const REQUEST_TIMEOUT_MS = 10 * 1000; |
| 11 | + |
| 12 | +export type NetStats = { bytesIn: number; bytesOut: number }; |
| 13 | + |
| 14 | +const NetStatsSchema = z.object({ |
| 15 | + bytesIn: z.number().int().min(0), |
| 16 | + bytesOut: z.number().int().min(0), |
| 17 | +}); |
| 18 | + |
| 19 | +function normalizeNetStats(value: unknown): NetStats { |
| 20 | + const parsed = NetStatsSchema.safeParse(value); |
| 21 | + if (!parsed.success) { |
| 22 | + return { bytesIn: 0, bytesOut: 0 }; |
| 23 | + } |
| 24 | + return parsed.data; |
| 25 | +} |
| 26 | + |
| 27 | +export type CheckinDeps = { |
| 28 | + getApiKey: () => string; |
| 29 | + getGatewayToken: () => string; |
| 30 | + getSandboxId: () => string; |
| 31 | + getCheckinUrl: () => string; |
| 32 | + getSupervisorStats: () => SupervisorStats; |
| 33 | + getOpenclawVersion: () => Promise<OpenclawVersionInfo>; |
| 34 | + getMachineId?: () => string; |
| 35 | +}; |
| 36 | + |
| 37 | +export function parseNetLine(line: string): NetStats { |
| 38 | + const parts = line.trim().split(/\s+/); |
| 39 | + return normalizeNetStats({ |
| 40 | + bytesIn: Number.parseInt(parts[1] ?? '', 10) || 0, |
| 41 | + bytesOut: Number.parseInt(parts[9] ?? '', 10) || 0, |
| 42 | + }); |
| 43 | +} |
| 44 | + |
| 45 | +export function parseNetDevText(raw: string): NetStats { |
| 46 | + const lines = raw.split('\n'); |
| 47 | + |
| 48 | + const eth0Line = lines.find(line => line.trim().startsWith('eth0:')); |
| 49 | + if (eth0Line) { |
| 50 | + return parseNetLine(eth0Line); |
| 51 | + } |
| 52 | + |
| 53 | + let bytesIn = 0; |
| 54 | + let bytesOut = 0; |
| 55 | + |
| 56 | + for (const line of lines) { |
| 57 | + const trimmed = line.trim(); |
| 58 | + if (!trimmed || trimmed.includes('|') || !trimmed.includes(':') || trimmed.startsWith('lo:')) { |
| 59 | + continue; |
| 60 | + } |
| 61 | + const stats = parseNetLine(trimmed); |
| 62 | + bytesIn += stats.bytesIn; |
| 63 | + bytesOut += stats.bytesOut; |
| 64 | + } |
| 65 | + |
| 66 | + return normalizeNetStats({ bytesIn, bytesOut }); |
| 67 | +} |
| 68 | + |
| 69 | +export async function readNetStats(): Promise<NetStats> { |
| 70 | + try { |
| 71 | + const raw = await readFile('/proc/net/dev', 'utf8'); |
| 72 | + return parseNetDevText(raw); |
| 73 | + } catch { |
| 74 | + return { bytesIn: 0, bytesOut: 0 }; |
| 75 | + } |
| 76 | +} |
| 77 | + |
| 78 | +export function startCheckin(deps: CheckinDeps): () => void { |
| 79 | + const checkinUrl = deps.getCheckinUrl(); |
| 80 | + if (!checkinUrl) { |
| 81 | + return () => {}; |
| 82 | + } |
| 83 | + |
| 84 | + let previousRestarts = deps.getSupervisorStats().restarts; |
| 85 | + let previousNetStats: NetStats = { bytesIn: 0, bytesOut: 0 }; |
| 86 | + let checkinInFlight = false; |
| 87 | + |
| 88 | + void readNetStats().then(stats => { |
| 89 | + previousNetStats = stats; |
| 90 | + }); |
| 91 | + |
| 92 | + const doCheckin = async (): Promise<void> => { |
| 93 | + if (checkinInFlight) { |
| 94 | + return; |
| 95 | + } |
| 96 | + checkinInFlight = true; |
| 97 | + |
| 98 | + try { |
| 99 | + const apiKey = deps.getApiKey(); |
| 100 | + const gatewayToken = deps.getGatewayToken(); |
| 101 | + const sandboxId = deps.getSandboxId(); |
| 102 | + if (!apiKey || !gatewayToken || !sandboxId) { |
| 103 | + return; |
| 104 | + } |
| 105 | + |
| 106 | + const stats = deps.getSupervisorStats(); |
| 107 | + const openclawVersion = await deps.getOpenclawVersion(); |
| 108 | + const currentNetStats = await readNetStats(); |
| 109 | + |
| 110 | + const restartsSinceLastCheckin = Math.max(0, stats.restarts - previousRestarts); |
| 111 | + const bandwidthBytesIn = Math.max(0, currentNetStats.bytesIn - previousNetStats.bytesIn); |
| 112 | + const bandwidthBytesOut = Math.max(0, currentNetStats.bytesOut - previousNetStats.bytesOut); |
| 113 | + |
| 114 | + const lastExitReason = stats.lastExit |
| 115 | + ? stats.lastExit.signal |
| 116 | + ? `signal:${stats.lastExit.signal}` |
| 117 | + : stats.lastExit.code !== null |
| 118 | + ? `code:${stats.lastExit.code}` |
| 119 | + : '' |
| 120 | + : ''; |
| 121 | + |
| 122 | + const controller = new AbortController(); |
| 123 | + const timeout = setTimeout(() => { |
| 124 | + controller.abort(); |
| 125 | + }, REQUEST_TIMEOUT_MS); |
| 126 | + |
| 127 | + let response: Response; |
| 128 | + try { |
| 129 | + response = await fetch(checkinUrl, { |
| 130 | + method: 'POST', |
| 131 | + headers: { |
| 132 | + 'content-type': 'application/json', |
| 133 | + authorization: `Bearer ${apiKey}`, |
| 134 | + 'x-kiloclaw-gateway-token': gatewayToken, |
| 135 | + }, |
| 136 | + signal: controller.signal, |
| 137 | + body: JSON.stringify({ |
| 138 | + sandboxId, |
| 139 | + machineId: deps.getMachineId?.() ?? process.env.FLY_MACHINE_ID ?? '', |
| 140 | + controllerVersion: CONTROLLER_VERSION, |
| 141 | + controllerCommit: CONTROLLER_COMMIT, |
| 142 | + openclawVersion: openclawVersion.version, |
| 143 | + openclawCommit: openclawVersion.commit, |
| 144 | + supervisorState: stats.state, |
| 145 | + totalRestarts: stats.restarts, |
| 146 | + restartsSinceLastCheckin, |
| 147 | + uptimeSeconds: stats.uptime, |
| 148 | + loadAvg5m: loadavg()[1] ?? 0, |
| 149 | + bandwidthBytesIn, |
| 150 | + bandwidthBytesOut, |
| 151 | + lastExitReason, |
| 152 | + }), |
| 153 | + }); |
| 154 | + } finally { |
| 155 | + clearTimeout(timeout); |
| 156 | + } |
| 157 | + |
| 158 | + if (!response.ok) { |
| 159 | + const errorText = await response.text().catch(() => ''); |
| 160 | + console.error(`[checkin] HTTP ${response.status}: ${errorText}`); |
| 161 | + return; |
| 162 | + } |
| 163 | + |
| 164 | + // Only advance baselines after a successful checkin. |
| 165 | + previousRestarts = stats.restarts; |
| 166 | + previousNetStats = currentNetStats; |
| 167 | + } catch (err) { |
| 168 | + console.error('[checkin] failed:', err); |
| 169 | + } finally { |
| 170 | + checkinInFlight = false; |
| 171 | + } |
| 172 | + }; |
| 173 | + |
| 174 | + let interval: ReturnType<typeof setInterval> | undefined; |
| 175 | + |
| 176 | + const initialTimeout = setTimeout(() => { |
| 177 | + void doCheckin(); |
| 178 | + interval = setInterval(() => { |
| 179 | + void doCheckin(); |
| 180 | + }, CHECKIN_INTERVAL_MS); |
| 181 | + }, INITIAL_DELAY_MS); |
| 182 | + |
| 183 | + return () => { |
| 184 | + clearTimeout(initialTimeout); |
| 185 | + if (interval) { |
| 186 | + clearInterval(interval); |
| 187 | + } |
| 188 | + }; |
| 189 | +} |
0 commit comments