pi-freeflow 1.1.8 → 1.2.0

package/extensions/index.ts

@@ -735,13 +735,37 @@ const LOG_MAX_BYTES = 5 * 1024 * 1024;
 const LOG_MAX_FILES = 3;
 const DEBUG_STATE_FILE = path.join(homedir(), ".pi", "agent", "pi-freeflow-debug.json");
 interface DebugState { debug: boolean; level?: LogLevel }
+let cachedDebugState: DebugState | null | undefined = undefined;
+let cachedDebugMtime = 0;
+let cachedDebugAt = 0;
 function loadDebugState(): DebugState | null {
+	const now = Date.now();
+	// cache for 1s to avoid per-chunk FS hit in hot pipe path
+	if (cachedDebugState !== undefined && now - cachedDebugAt < 1000) {
+		return cachedDebugState;
+	}
 	try {
-		if (!fs.existsSync(DEBUG_STATE_FILE)) return null;
+		if (!fs.existsSync(DEBUG_STATE_FILE)) {
+			cachedDebugState = null;
+			cachedDebugAt = now;
+			return null;
+		}
+		const stat = fs.statSync(DEBUG_STATE_FILE);
+		if (stat.mtimeMs === cachedDebugMtime && cachedDebugState !== undefined) {
+			cachedDebugAt = now;
+			return cachedDebugState;
+		}
 		const raw = fs.readFileSync(DEBUG_STATE_FILE, "utf8");
 		const d = JSON.parse(raw) as DebugState;
-		if (typeof d.debug === "boolean") return d;
+		if (typeof d.debug === "boolean") {
+			cachedDebugState = d;
+			cachedDebugMtime = stat.mtimeMs;
+			cachedDebugAt = now;
+			return d;
+		}
 	} catch {}
+	cachedDebugState = null;
+	cachedDebugAt = now;
 	return null;
 }
 function saveDebugState(s: DebugState): void {
@@ -751,6 +775,16 @@ function saveDebugState(s: DebugState): void {
 		const tmp = `${DEBUG_STATE_FILE}.${randomUUID()}.tmp`;
 		fs.writeFileSync(tmp, JSON.stringify(s, null, 2), "utf8");
 		fs.renameSync(tmp, DEBUG_STATE_FILE);
+		// update cache
+		try {
+			const stat = fs.statSync(DEBUG_STATE_FILE);
+			cachedDebugState = s;
+			cachedDebugMtime = stat.mtimeMs;
+			cachedDebugAt = Date.now();
+		} catch {
+			cachedDebugState = s;
+			cachedDebugAt = Date.now();
+		}
 	} catch {}
 }
 function getMinLogLevel(): number {
@@ -1419,8 +1453,8 @@ function startProxy(
 			} catch {}
 
 			const upstream: Upstream = isKilo ? "kilo" : "opencode";
+			const isStream = parsedBody?.stream === true;
 			if (!checkRateLimit(clientIP, upstream)) {
-				log("warn", `rate limit hit for ${upstream} from ${clientIP}`, { ip: clientIP, upstream }, reqId);
 				res.writeHead(429, { "content-type": "application/json" });
 				res.end(JSON.stringify({ error: "rate limit exceeded" }));
 				return;
@@ -1474,52 +1508,49 @@ function startProxy(
 							activeHost,
 						);
 						try {
-							if (relayBody && parsedBody) {
+							if (parsedBody) {
 								const relayBodyObj = structuredClone(parsedBody);
 								normalizeRequestBody(relayBodyObj, true, isKilo, reqId);
-								relayBody = Buffer.from(JSON.stringify(relayBodyObj));
-							}
-							const response = await relayFetch(fullUrl, {
-								method: req.method || "POST",
-								headers: relayHeaders,
-								body: relayBody,
-								signal: AbortSignal.timeout(300_000),
-							}, reqId);
-							const ct =
-								response.headers.get("content-type") || "application/json";
-							if (response.ok && response.body) {
-								const ct =
-									response.headers.get("content-type") || "text/event-stream";
-								res.writeHead(response.status, {
-									"content-type": ct,
-									"cache-control": "no-cache, no-transform",
-									"connection": "keep-alive",
-									"x-accel-buffering": "no",
-								});
-								pipeUpstreamStream(
-									Readable.fromWeb(
-										response.body as unknown as WebReadableStream,
-									),
-									res,
-									req,
-									reqId,
-								);
-							} else {
-								const data = await response.text();
-								const ct =
-									response.headers.get("content-type") || "application/json";
-								res.writeHead(response.status, { "content-type": ct });
-								res.end(data);
+								const relayBody = Buffer.from(JSON.stringify(relayBodyObj));
+								const response = await relayFetch(fullUrl, {
+									method: req.method || "POST",
+									headers: relayHeaders,
+									body: relayBody,
+									signal: AbortSignal.timeout(300_000),
+								}, reqId);
+								if (isStream && response.ok && response.body) {
+									const ct =
+										response.headers.get("content-type") || "text/event-stream";
+									res.writeHead(response.status, {
+										"content-type": ct,
+										"cache-control": "no-cache, no-transform",
+										"connection": "keep-alive",
+										"x-accel-buffering": "no",
+									});
+									pipeUpstreamStream(
+										Readable.fromWeb(
+											response.body as unknown as WebReadableStream,
+										),
+										res,
+										req,
+										reqId,
+									);
+								} else {
+									const data = await response.text();
+									const ct =
+										response.headers.get("content-type") || "application/json";
+									res.writeHead(response.status, { "content-type": ct });
+									res.end(data);
+								}
+								return; // relay handled the response
 							}
-							return; // relay handled the response
 						} catch (e) {
 							log("warn", "opencode relay failed, falling back to direct", {
 								error: String(e),
 							}, reqId);
- 							if (res.headersSent) return; // can't recover mid-stream
- 						}
+							if (res.headersSent) return; // can't recover mid-stream
+						}
  					}
- 					// direct path (existing, untouched)
 					// direct path — with debug trace and thinking-aware normalize
 					let directBody = Buffer.concat(bodyChunks);
 					if (parsedBody) {

package/package.json

@@ -1,7 +1,7 @@
 {
 	"name": "pi-freeflow",
 	"type": "module",
-	"version": "1.1.8",
+	"version": "1.2.0",
 	"description": "Personal multi-cloud rolling fallback relay for OpenCode Zen and KiloCode models in OMP/Pi",
 	"keywords": [
 		"pi-package",