rene.fichtmueller/llm-gateway
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: track bridge tokens in request_tracking for dashboard totals

Browse Source 
codex-bridge passthrough previously hardcoded tokens_in=0 and used
text.length (char count) for tokens_out, writing only to llm_calls.
request_tracking never received these entries so all Codex/Claude/
Copilot bridge traffic was invisible in dashboard token totals.

Fix:
- Extract real token counts from upstreamJson.usage (prompt_tokens /
  completion_tokens) with char/4 estimation as fallback
- Call requestLogger.logRequest() to write to request_tracking
- Pass real totals to recordSubscriptionUsage() (was 0)
- Both audit log (llm_calls) and dashboard log (request_tracking) now
  receive consistent bridgeCaller + real token counts
This commit is contained in:
root 2026-07-06 21:07:11 +00:00
parent 73da9d7391
commit ddb8779ec7
1 changed files with 28 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

33
packages/gateway/src/routes/completion.ts
View File

@ -1191,22 +1191,45 @@ export async function completionRoute(fastify: FastifyInstance): Promise<void> {
const text = upstreamJson?.content ?? upstreamJson?.response ?? upstreamJson?.choices?.[0]?.message?.content ?? '';
const respBody = toOpenAIResponsesResponse({ output: text, model: parsed.data.model, status: 'approved' }, parsed.data.model);
logger.info({ callId, model: parsed.data.model, len: text.length }, 'codex-bridge passthrough OK');
const bridgeCaller = (request.headers['x-llm-interceptor-caller'] as string) || 'codex-app';
const bridgeTokensIn = (upstreamJson?.usage?.prompt_tokens as number | undefined) ?? Math.ceil(inputText.length / 4);
const bridgeTokensOut = (upstreamJson?.usage?.completion_tokens as number | undefined) ?? Math.ceil(text.length / 4);
// Track against the merged OpenAI (ChatGPT+Codex) subscription pool.
try {
const subId = modelToSubscriptionId(parsed.data.model ?? '') ?? 'codex';
void recordSubscriptionUsage(getPool(), subId, 0);
void recordSubscriptionUsage(getPool(), subId, bridgeTokensIn + bridgeTokensOut);
} catch (e) {
logger.warn({ e, callId }, 'failed to record subscription usage for passthrough');
}
// Also write an audit row so the dashboard activity tab sees it.
// Write to request_tracking so dashboard token totals include bridge calls.
try {
const bridgeLogger = createRequestLogger(getPool());
void bridgeLogger.logRequest(
callId,
bridgeCaller,
'codex_passthrough',
parsed.data.model ?? 'gpt-5.5',
'approved',
bridgeTokensIn,
bridgeTokensOut,
0,
Date.now() - startMs,
0,
false,
undefined,
);
} catch (e) {
logger.warn({ e, callId }, 'failed to log bridge request to request_tracking');
}
// Also write an audit row for the activity tab.
try {
void writeAuditLog({
callId,
caller: (request.headers['x-llm-interceptor-caller'] as string) || 'codex-app',
caller: bridgeCaller,
task_type: 'codex_passthrough',
status: 'approved',
tokens_in: 0,
tokens_out: text.length,
tokens_in: bridgeTokensIn,
tokens_out: bridgeTokensOut,
latency_ms: Date.now() - startMs,
confidence: 0,
cost_usd: 0,