build: add gemini and xai provider modules
Expands token-monitor with two new provider types:
- providers/gemini.js — Google Gemini API (body-based quota, no headers)
- Probes generateContent endpoint (1 token), falls back gemini-2.0-flash → gemini-2.5-flash
- Parses QuotaFailure violations + RetryInfo from 429 JSON body
- Returns: status, quota_violations[], retry_delay_seconds, severity
- providers/xai.js — x.ai/Grok (OpenAI-compatible header schema)
- Reads x-ratelimit-{limit,remaining}-{requests,tokens} headers
- Handles: no_key, ok, rate_limited, invalid_key states
- Warning threshold: < 10% remaining on requests or tokens
Both providers handle missing API keys gracefully (status: no_key).
Classification via providers/index.js using baseUrl patterns.
140/140 tests passing.
Closes recon findings from trentuna/a-team#91.
This commit is contained in:
parent
988618e165
commit
1b4e299461
2 changed files with 270 additions and 0 deletions
131
providers/gemini.js
Normal file
131
providers/gemini.js
Normal file
|
|
@ -0,0 +1,131 @@
|
|||
/**
|
||||
* gemini.js — Google Gemini API provider
|
||||
*
|
||||
* Gemini returns zero rate-limit headers. Quota state is embedded in 429 JSON body.
|
||||
* On 200: liveness confirmed, quota depth unavailable.
|
||||
* On 429 (RESOURCE_EXHAUSTED): parse error.details for QuotaFailure + RetryInfo.
|
||||
*
|
||||
* Probe endpoint: POST /v1beta/models/gemini-2.0-flash:generateContent?key=<KEY>
|
||||
* Body: {"contents":[{"parts":[{"text":"Hi"}]}],"generationConfig":{"maxOutputTokens":1}}
|
||||
*
|
||||
* Provider type in index.js: 'gemini-api'
|
||||
*
|
||||
* Quota from 429 body:
|
||||
* error.details[].@type === 'type.googleapis.com/google.rpc.QuotaFailure'
|
||||
* → .violations[].quotaId (e.g. 'GenerateRequestsPerDayPerProjectPerModel-FreeTier')
|
||||
* → .violations[].quotaMetric
|
||||
* error.details[].@type === 'type.googleapis.com/google.rpc.RetryInfo'
|
||||
* → .retryDelay (e.g. '43s' — parse to integer seconds)
|
||||
*/
|
||||
|
||||
/**
|
||||
* Parse a Gemini API response body and HTTP status into a normalized result.
|
||||
* Synchronous — suitable for unit tests with mock data.
|
||||
*
|
||||
* @param {Object|null} body — parsed JSON response body
|
||||
* @param {number} httpStatus — HTTP status code
|
||||
* @returns {Object} normalized provider result
|
||||
*/
|
||||
export function parseGeminiBody(body, httpStatus) {
|
||||
if (httpStatus === 401 || httpStatus === 403) {
|
||||
return { type: 'gemini-api', status: 'invalid_key', severity: 'unknown' };
|
||||
}
|
||||
|
||||
if (httpStatus === 429) {
|
||||
const details = body?.error?.details || [];
|
||||
|
||||
// Extract quota violations from QuotaFailure detail
|
||||
const quotaFailure = details.find(
|
||||
(d) => d['@type'] === 'type.googleapis.com/google.rpc.QuotaFailure'
|
||||
);
|
||||
const quotaViolations = (quotaFailure?.violations || [])
|
||||
.map((v) => v.quotaId || v.quotaMetric || '')
|
||||
.filter(Boolean);
|
||||
|
||||
// Extract retry delay from RetryInfo detail (e.g. '43s' → 43)
|
||||
const retryInfo = details.find(
|
||||
(d) => d['@type'] === 'type.googleapis.com/google.rpc.RetryInfo'
|
||||
);
|
||||
let retryDelaySeconds = null;
|
||||
if (retryInfo?.retryDelay) {
|
||||
const match = String(retryInfo.retryDelay).match(/^(\d+)/);
|
||||
retryDelaySeconds = match ? parseInt(match[1], 10) : null;
|
||||
}
|
||||
|
||||
return {
|
||||
type: 'gemini-api',
|
||||
status: 'exhausted',
|
||||
quota_violations: quotaViolations,
|
||||
retry_delay_seconds: retryDelaySeconds,
|
||||
severity: 'critical',
|
||||
};
|
||||
}
|
||||
|
||||
if (httpStatus === 200) {
|
||||
return {
|
||||
type: 'gemini-api',
|
||||
status: 'ok',
|
||||
quota_violations: [],
|
||||
retry_delay_seconds: null,
|
||||
severity: 'ok',
|
||||
};
|
||||
}
|
||||
|
||||
// Other errors (404, 5xx, etc.)
|
||||
return {
|
||||
type: 'gemini-api',
|
||||
status: 'error',
|
||||
message: body?.error?.message || `HTTP ${httpStatus}`,
|
||||
severity: 'unknown',
|
||||
};
|
||||
}
|
||||
|
||||
/**
|
||||
* Probe a Gemini provider endpoint.
|
||||
* Tries gemini-2.0-flash first; falls back to gemini-2.5-flash on 404.
|
||||
*
|
||||
* @param {string} providerName
|
||||
* @param {string} baseUrl — base URL (e.g. https://generativelanguage.googleapis.com)
|
||||
* @param {string|null} apiKey — Google API key
|
||||
* @returns {Promise<Object>} normalized provider result
|
||||
*/
|
||||
export async function probeGeminiProvider(providerName, baseUrl, apiKey) {
|
||||
if (!apiKey) {
|
||||
return { type: 'gemini-api', status: 'no_key', severity: 'unknown' };
|
||||
}
|
||||
|
||||
const base = (baseUrl || 'https://generativelanguage.googleapis.com').replace(/\/$/, '');
|
||||
|
||||
async function tryModel(model) {
|
||||
const url = `${base}/v1beta/models/${model}:generateContent?key=${apiKey}`;
|
||||
const response = await fetch(url, {
|
||||
method: 'POST',
|
||||
headers: { 'content-type': 'application/json' },
|
||||
body: JSON.stringify({
|
||||
contents: [{ parts: [{ text: 'Hi' }] }],
|
||||
generationConfig: { maxOutputTokens: 1 },
|
||||
}),
|
||||
});
|
||||
let body = null;
|
||||
try { body = await response.json(); } catch (_) {}
|
||||
return { status: response.status, body };
|
||||
}
|
||||
|
||||
try {
|
||||
let { status, body } = await tryModel('gemini-2.0-flash');
|
||||
|
||||
// 404 = model not found, try fallback
|
||||
if (status === 404) {
|
||||
({ status, body } = await tryModel('gemini-2.5-flash'));
|
||||
}
|
||||
|
||||
return parseGeminiBody(body, status);
|
||||
} catch (err) {
|
||||
return {
|
||||
type: 'gemini-api',
|
||||
status: 'error',
|
||||
message: err.message,
|
||||
severity: 'unknown',
|
||||
};
|
||||
}
|
||||
}
|
||||
Loading…
Add table
Add a link
Reference in a new issue