pi-freeflow 1.42.0 → 1.42.1

package/CHANGELOG.md

@@ -1,5 +1,11 @@
 # Changelog
 
+## 1.42.1
+
+### Patch Changes
+
+- 5ab7509: Enable forced round-robin relay selection for stateless OVHcloud AI requests across all healthy pool relays, load-balancing requests to prevent hitting upstream per-IP rate limits (2 req/min) while preserving sticky session affinity for OpenCode Zen.
+
 ## 1.42.0
 
 ### Minor Changes

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "pi-freeflow",
   "type": "module",
-  "version": "1.42.0",
+  "version": "1.42.1",
   "description": "Thin provider for OMP/Pi — model list + dumb relay proxy + log; host pi-ai owns thinking/normalization",
   "main": "extensions/index.ts",
   "types": "src/index.ts",

package/src/proxy.ts

@@ -787,7 +787,10 @@ delete chatBody.max_output_tokens;
      signal: controller.signal,
     },
     reqId,
-    { onServed: (url) => { servedRelay = url; } },
+    {
+     roundRobin: true,
+     onServed: (url) => { servedRelay = url; },
+    },
     (failure) => audit?.attemptFailure({ attempt: failure.attempt, status: failure.status, reason: failure.reason }),
    );
   } else {

package/src/relay-state.ts

@@ -824,26 +824,32 @@ export function isSessionPinEligible(pin: string, ordered: string[]): boolean {
  * a key the cursor round-robins per request. Cooling relays stay at the tail,
  * so a rate-limited IP is never preferred.
  */
-export function orderedRelayCandidates(preferredRelay?: string, spreadKey?: string): string[] {
- const ordered = getOrderedRelayUrls();
- const preferred = (preferredRelay ?? "").trim();
- if (preferred) {
-  const index = ordered.indexOf(preferred);
-  if (index <= 0) return ordered;
-  if (!isSessionPinEligible(preferred, ordered)) return ordered;
-  return [ordered[index], ...ordered.slice(0, index), ...ordered.slice(index + 1)];
- }
- if (activeRelayState.mode !== "spread") return ordered;
- const healthy = ordered.filter((u) => isRelayHealthy(u));
- const cooling = ordered.filter((u) => !isRelayHealthy(u));
- if (healthy.length <= 1) return [...healthy, ...cooling];
- const k = spreadKey ? stableHash(spreadKey) % healthy.length : spreadCursor++ % healthy.length;
- return [...healthy.slice(k), ...healthy.slice(0, k), ...cooling];
+export function orderedRelayCandidates(
+	preferredRelay?: string,
+	spreadKey?: string,
+	forceRoundRobin?: boolean,
+): string[] {
+	const ordered = getOrderedRelayUrls();
+	const preferred = (preferredRelay ?? "").trim();
+	if (preferred) {
+		const index = ordered.indexOf(preferred);
+		if (index <= 0) return ordered;
+		if (!isSessionPinEligible(preferred, ordered)) return ordered;
+		return [ordered[index], ...ordered.slice(0, index), ...ordered.slice(index + 1)];
+	}
+	if (activeRelayState.mode !== "spread" && !forceRoundRobin) return ordered;
+	const healthy = ordered.filter((u) => isRelayHealthy(u));
+	const cooling = ordered.filter((u) => !isRelayHealthy(u));
+	if (healthy.length <= 1) return [...healthy, ...cooling];
+	const k = spreadKey ? stableHash(spreadKey) % healthy.length : spreadCursor++ % healthy.length;
+	return [...healthy.slice(k), ...healthy.slice(0, k), ...cooling];
 }
 
 /** Round-robin cursor for spread mode when no stable key is available. */
 let spreadCursor = 0;
 
+/** Test-only: reset spread / round-robin cursor */
+export function _resetSpreadCursorForTest(): void { spreadCursor = 0; }
 /** FNV-1a: stable string -> bucket index for per-conversation sharding. */
 function stableHash(s: string): number {
  let h = 2166136261;

package/src/relay.ts

@@ -151,6 +151,8 @@ export interface RelayAffinity {
 	preferred?: string;
 	/** Stable key used to shard requests across healthy relays in spread mode. */
 	spreadKey?: string;
+	/** When true, healthy relays rotate round-robin regardless of pool mode (e.g. for OVH). */
+	roundRobin?: boolean;
 	onServed?: (relay: string | null) => void;
 }
 
@@ -186,7 +188,7 @@ export async function relayFetch(
 		affinity.onServed?.(null);
 		return fetch(url, opts as unknown as RequestInit);
 	}
-	const candidates = orderedRelayCandidates(affinity.preferred, affinity.spreadKey);
+	const candidates = orderedRelayCandidates(affinity.preferred, affinity.spreadKey, affinity.roundRobin);
 
 	if (candidates.length === 0) {
 		// Empty pool: skip straight to upstream instead of logging a misleading
@@ -709,11 +711,12 @@ export async function relayFetch(
 			const affinityServed =
 				Boolean(affinity.preferred) &&
 				targetUrl.trim() === (affinity.preferred ?? "").trim();
-			// Spread mode has no single primary: every request picks its own
-			// healthy relay, so rewriting the sticky URL here would churn the
-			// state file and spam "active relay auto-switched" on every turn.
+			// Spread mode and forced round-robin have no single primary: every
+			// request picks its own healthy relay, so rewriting the sticky URL here
+			// would churn the state file and spam "active relay auto-switched" on every turn.
 			const spreadMode = relayState.mode === "spread";
-			if (relayState.url !== targetUrl && !affinityServed && !spreadMode) {
+			const roundRobin = Boolean(affinity.roundRobin);
+			if (relayState.url !== targetUrl && !affinityServed && !spreadMode && !roundRobin) {
 				log("info", `active relay auto-switched to ${targetUrl}`, {
 					previous: relayState.url,
 				}, rid);