llm-relay 0.73.1 → 0.74.0

package/dist/cli.js

@@ -15,6 +15,7 @@ import { providerCredentialSlots, slotAllowsModel } from "./credential-fleet.js"
 import { loadLaneManifest, rosterIsStale, verifyModel } from "./lane-manifest.js";
 import { probeLanes } from "./lane-probe.js";
 import { buildDispatch, allLadderRungs, normalizeCliCommand, restoreExhaustedRows, specContextWindow, CONTEXT_TOKEN, TASK_TOKEN, formatLaneStats } from "./dispatch.js";
+import { loadLaneAffinityRows, restoreLaneAffinityRows } from "./lane-affinity.js";
 import { McpDispatchServer } from "./mcp/server.js";
 import { loadExhaustedRows } from "./dispatch-exhaustion-persistence.js";
 import { detectHostRouting, parseHostRoutingState } from "./host-routing.js";
@@ -1669,12 +1670,13 @@ export async function resolveDispatchView(opts) {
     const cachedContextWindow = setupDispatchCatalog(cfg);
     const liveRaw = (await tryServer(cfg, `/dispatch?${qs}`));
     const live = liveRaw !== null && liveRaw.host === "bypassed" ? liveRaw : null;
-    restoreExhaustedRows(cfg, loadExhaustedRows());
     if (live !== null) {
         const view = normalizeDispatchCommands(live);
         return substituteTaskInView({ ...view, source: "daemon" }, opts.task);
     }
     const fallbackCfg = loadConfigSafely() ?? cfg;
+    restoreExhaustedRows(fallbackCfg, loadExhaustedRows());
+    restoreLaneAffinityRows(fallbackCfg, loadLaneAffinityRows());
     const fallbackView = normalizeDispatchCommands(buildDispatch(fallbackCfg, {
         ...(opts.lane ? { lane: opts.lane } : {}),
         ...(opts.tier ? { tier: opts.tier } : {}),
@@ -1791,6 +1793,7 @@ export async function runDispatch(arg) {
     const staleProxy = hostStale || windowStale;
     const live = staleProxy ? null : liveRaw;
     restoreExhaustedRows(cfg, loadExhaustedRows());
+    restoreLaneAffinityRows(cfg, loadLaneAffinityRows());
     const rawView = normalizeDispatchCommands(live ??
         buildDispatch(cfg, {
             ...(lane ? { lane } : {}),
@@ -1865,6 +1868,10 @@ export async function runDispatch(arg) {
         }
         if (l.note)
             process.stdout.write(`   note: ${l.note}\n`);
+        if (l.pinned)
+            process.stdout.write(`   pinned until ${l.pinned.until} (${l.pinned.reason})\n`);
+        if (l.demoted)
+            process.stdout.write(`   demoted until ${l.demoted.until} (${l.demoted.reason})\n`);
         if (l.stats)
             process.stdout.write(`   ${formatLaneStats(l.stats)}\n`);
     }