Spaces:
Paused
Paused
File size: 8,435 Bytes
9284ad7 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 | // Node 20 includes native fetch support
// AI Client for Groq with Failover Support + Circuit Breaker
// Fallback shims for Probot runtime
/* eslint-disable no-console */
import {
ChatGPTAPI as AIClient,
ChatGPTError as AIError,
ChatMessage,
SendMessageOptions
} from 'chatgpt'
import pRetry from 'p-retry'
import CircuitBreaker from 'opossum'
import {AIOptions, Options} from './options'
import {TokenLimits} from './limits'
import pino from 'pino'
const logger = pino({ level: process.env.LOG_LEVEL || 'info' })
// Circuit breaker configuration
const CIRCUIT_BREAKER_OPTIONS = {
timeout: 60000, // 60 seconds
errorThresholdPercentage: 50,
resetTimeout: 30000,
volumeThreshold: 5,
}
const logInfo = (msg: string) => logger.info(msg)
const logWarning = (msg: string) => logger.warn(msg)
const logError = (msg: string) => logger.error(msg)
const setFailed = (msg: string) => {
logger.error(msg)
process.exit(1)
}
export interface Ids {
parentMessageId?: string
conversationId?: string
}
export class Bot {
private api: AIClient | null = null
private readonly options: Options
private readonly aiOptions: AIOptions
private currentModel: string
private fallbackModels: string[]
private circuitBreaker: CircuitBreaker | null = null
constructor(options: Options, aiOptions: AIOptions) {
this.options = options
this.aiOptions = aiOptions
this.currentModel = aiOptions.model
// Setup fallback chain based on tier
if (Options.HEAVY_MODELS.includes(this.currentModel)) {
this.fallbackModels = [...Options.HEAVY_MODELS]
} else {
this.fallbackModels = [...Options.LIGHT_MODELS]
}
// Remove current model from fallbacks to avoid immediate repeat
this.fallbackModels = this.fallbackModels.filter(
m => m !== this.currentModel
)
this.initClient()
this.initCircuitBreaker()
}
private initCircuitBreaker(): void {
if (!this.api) return
// Wrap the sendMessage method with circuit breaker
this.circuitBreaker = new CircuitBreaker(
async ({ message, opts }: { message: string; opts: SendMessageOptions }) => {
return this.api!.sendMessage(message, opts)
},
CIRCUIT_BREAKER_OPTIONS
)
// Circuit breaker event handlers
this.circuitBreaker.on('open', () => {
logWarning(`Circuit breaker opened for model: ${this.currentModel}`)
})
this.circuitBreaker.on('halfOpen', () => {
logInfo(`Circuit breaker half-open for model: ${this.currentModel}`)
})
this.circuitBreaker.on('close', () => {
logInfo(`Circuit breaker closed for model: ${this.currentModel}`)
})
this.circuitBreaker.on('fallback', (_result: unknown) => {
logWarning(`Circuit breaker fallback executed for model: ${this.currentModel}`)
})
}
private initClient(): void {
const apiKey = process.env.GROQ_API_KEY || process.env.AI_API_KEY
if (apiKey) {
const currentDate = new Date().toISOString().split('T')[0]
const limits = new TokenLimits(this.currentModel)
const systemMessage = `${this.options.systemMessage}
Knowledge cutoff: ${limits.knowledgeCutOff}
Current date: ${currentDate}
IMPORTANT: Entire response must be in ISO code: ${this.options.language}
`
this.api = new AIClient({
apiBaseUrl: this.options.apiBaseUrl,
systemMessage,
apiKey,
apiOrg: process.env.AI_API_ORG || undefined,
debug: this.options.debug,
maxModelTokens: limits.maxTokens,
maxResponseTokens: limits.responseTokens,
completionParams: {
temperature: this.options.modelTemperature,
model: this.currentModel
}
})
logInfo(`Initialized AI client with model: ${this.currentModel}`)
} else {
throw new Error("Missing 'GROQ_API_KEY' or 'AI_API_KEY'")
}
}
chat = async (message: string, ids: Ids): Promise<[string, Ids]> => {
let res: [string, Ids] = ['', {}]
try {
res = await this.chat_(message, ids)
return res
} catch (e: unknown) {
if (e instanceof AIError) {
logWarning(`AI communication error: ${e.message}`)
}
return res
}
}
private readonly chat_ = async (
message: string,
ids: Ids
): Promise<[string, Ids]> => {
const start = Date.now()
if (!message) return ['', {}]
let response: ChatMessage | undefined
if (this.api != null) {
const opts: SendMessageOptions = {
timeoutMs: this.options.timeoutMS
}
if (ids.parentMessageId) {
opts.parentMessageId = ids.parentMessageId
}
try {
// Check if circuit breaker is open
if (this.circuitBreaker?.opened) {
logWarning(`Circuit breaker is OPEN for ${this.currentModel}, attempting fallback...`)
if (this.fallbackModels.length > 0) {
const nextModel = this.fallbackModels.shift()
if (nextModel) {
this.currentModel = nextModel
this.initClient()
this.initCircuitBreaker()
// Recursive retry with new model
return this.chat_(message, ids)
}
}
throw new Error(`Circuit breaker open and no fallback models available for ${this.currentModel}`)
}
const result = await pRetry<ChatMessage>(
async () => {
if (!this.circuitBreaker) {
throw new Error('Circuit breaker not initialized')
}
try {
const messageResult = await this.circuitBreaker.fire({ message, opts })
return messageResult as ChatMessage
} catch (e: any) {
// Check if circuit breaker opened
if (this.circuitBreaker.opened && this.fallbackModels.length > 0) {
const nextModel = this.fallbackModels.shift()
if (nextModel) {
logWarning(
`Circuit breaker opened for ${this.currentModel}. Switching to fallback: ${nextModel}`
)
this.currentModel = nextModel
this.initClient()
this.initCircuitBreaker()
// Throw to trigger retry with new model
throw new Error(`Model switched to ${nextModel}, retrying...`)
}
}
// Handle rate limiting / context size
if ((e?.status === 429 || e?.status === 413) && this.fallbackModels.length > 0) {
const nextModel = this.fallbackModels.shift()
if (nextModel) {
logWarning(
`Rate limit or context size error (${e.status}) hit for ${this.currentModel}. Switching to fallback: ${nextModel}`
)
this.currentModel = nextModel
this.initClient()
this.initCircuitBreaker()
throw new Error(`Model switched to ${nextModel}, retrying...`)
}
}
throw e
}
},
{
retries: this.options.retries,
minTimeout: 20000, // 20s delay to allow TPM refresh
onFailedAttempt: (err: any) => {
logInfo(
`Attempt ${err.attemptNumber} failed. ${err.retriesLeft} retries left. Error: ${err.message}`
)
}
}
)
response = result
} catch (e: unknown) {
if (e instanceof AIError) {
logWarning(
`AI Error after retries: ${e.message} (Last model used: ${this.currentModel})`
)
} else if (e instanceof Error) {
logError(`Failed to get AI response: ${e.message}`)
}
}
const end = Date.now()
logInfo(`AI response time: ${end - start} ms`)
} else {
setFailed('AI client is not initialized')
}
let responseText = ''
if (response != null) {
responseText = response.text
}
const newIds: Ids = {
parentMessageId: response?.id,
conversationId: response?.conversationId
}
return [responseText, newIds]
}
}
|