@bman654/clodex 1.0.3 → 1.0.4

package/dist/cli.js

@@ -202,7 +202,7 @@ import { join } from "path";
 // package.json
 var package_default = {
   name: "@bman654/clodex",
-  version: "1.0.3",
+  version: "1.0.4",
   publishConfig: {
     access: "public"
   },
@@ -7599,6 +7599,8 @@ function anthropicPromptTooLongMessage(body, contextWindow) {
 }
 
 // src/proxy.ts
+var STREAM_KEEPALIVE_INTERVAL_MS = 2e4;
+var STREAM_KEEPALIVE_PING = 'event: ping\ndata: {"type":"ping"}\n\n';
 function createTranslationLifecycle(logPath, requestId, modelId, provider) {
   if (!logPath || !requestId) return void 0;
   const startedAt = Date.now();
@@ -7942,6 +7944,9 @@ function startProxyCatalog(routes, defaultAliasId, debug = false, inferenceLogPa
           });
           translationLifecycle?.dispatched();
           if (clientWantsStream) {
+            const keepAliveMs = Number(process.env.CLODEX_STREAM_KEEPALIVE_INTERVAL_MS) || STREAM_KEEPALIVE_INTERVAL_MS;
+            let lastDownstreamWriteAt = Date.now();
+            let lastUpstreamPartAt = Date.now();
             const writeStreamChunk = (chunk) => {
               translationLifecycle?.onOutput(chunk);
               if (!res.headersSent) {
@@ -7951,23 +7956,43 @@ function startProxyCatalog(routes, defaultAliasId, debug = false, inferenceLogPa
                   "Connection": "keep-alive"
                 });
               }
+              lastDownstreamWriteAt = Date.now();
               res.write(chunk);
             };
-            await withResponsesWebSocketDiagnosticContext(
-              { requestId: relayRequestId, claudeSessionId },
-              () => streamAnthropicResponse(
-                model,
-                params,
-                originalModel,
-                writeStreamChunk,
-                plog,
-                {
-                  onPart: (partType) => translationLifecycle?.onPart(partType),
-                  initialInputTokens: estimateAnthropicInputTokens(anthropicBody),
-                  abortSignal: clientAbort.signal
-                }
-              )
-            );
+            const keepAlive = setInterval(() => {
+              if (res.writableEnded || !res.headersSent) return;
+              const now = Date.now();
+              const outputIdleMs = now - lastDownstreamWriteAt;
+              const upstreamIdleMs = now - lastUpstreamPartAt;
+              if (outputIdleMs >= keepAliveMs && upstreamIdleMs < keepAliveMs) {
+                lastDownstreamWriteAt = now;
+                res.write(STREAM_KEEPALIVE_PING);
+                plog(() => `stream keepalive ping: output idle ${outputIdleMs}ms, upstream active (${upstreamIdleMs}ms since last part)`);
+              }
+            }, keepAliveMs);
+            keepAlive.unref();
+            try {
+              await withResponsesWebSocketDiagnosticContext(
+                { requestId: relayRequestId, claudeSessionId },
+                () => streamAnthropicResponse(
+                  model,
+                  params,
+                  originalModel,
+                  writeStreamChunk,
+                  plog,
+                  {
+                    onPart: (partType) => {
+                      lastUpstreamPartAt = Date.now();
+                      translationLifecycle?.onPart(partType);
+                    },
+                    initialInputTokens: estimateAnthropicInputTokens(anthropicBody),
+                    abortSignal: clientAbort.signal
+                  }
+                )
+              );
+            } finally {
+              clearInterval(keepAlive);
+            }
             translationLifecycle?.complete();
             if (!res.headersSent) writeStreamChunk("");
             res.end();