geonix 1.32.0 → 1.32.1

package/README.md

@@ -295,6 +295,7 @@ Each returned part has:
 | `GX_LEVEL` | `info` | Log level: `debug`, `info`, `warn`, `error` |
 | `GX_STREAM_TIMEOUT` | `90000` | Max wait for a stream consumer to connect (ms) |
 | `GX_INACTIVITY_TIMEOUT` | `90000` | TCP proxy (HTTP-over-NATS) inactivity timeout (ms) |
+| `GX_HEALTH_TIMEOUT` | `2000` | Registry health-probe timeout per advertised address (ms) |
 | `GX_SECRET` | — | Encryption key: AES-256-GCM payloads + HMAC-SHA256 subjects. Services without the same key cannot communicate. |
 | `GX_DEBUG_ENDPOINT` | — | Mount path for the debug router (e.g. `/_debug`). Disabled when unset. |
 

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "geonix",
-    "version": "1.32.0",
+    "version": "1.32.1",
     "type": "module",
     "description": "",
     "bin": {

package/src/Registry.js

@@ -8,6 +8,11 @@ import { logger } from "./Logger.js";
 
 const REGISTRY_ENTRY_TIMEOUT = 5000;
 const GARBAGE_COLLECTOR_INTERVAL = 500;
+const HEALTH_TIMEOUT = 2000;
+
+// Read via function so tests can mutate the env after module load. Mirrors the
+// pattern used by getTransportTimeout / getStreamTimeout (per CLAUDE.md N6).
+const getHealthTimeout = () => parseInt(process.env.GX_HEALTH_TIMEOUT, 10) || HEALTH_TIMEOUT;
 
 /**
  * Maintains a local, in-process view of all services currently present on the NATS bus.
@@ -43,7 +48,7 @@ class Registry extends EventEmitter {
         this.#isActive = false;
     }
 
-    async #checkHealth(address, addresses, onFirstHealthy, timeout = 500) {
+    async #checkHealth(address, addresses, onFirstHealthy, timeout = getHealthTimeout()) {
         try {
             const result = await (await fetchWithTimeout(`http://${address}/!!_gx/health`, {}, timeout)).json();
 
@@ -137,17 +142,22 @@ class Registry extends EventEmitter {
         }
     }
 
-    // Refresh timeout for a known entry; if the beacon carries an `.a` that differs from
-    // the cached advertised list, kick off a re-probe to rebuild the healthy subset.
+    // Refresh timeout for a known entry; kick off a re-probe when either (a) the advertised
+    // list changed, or (b) the entry's healthy `a` is empty and the source is still advertising
+    // addresses — that's the "all probes failed at registration" case that needs to retry.
     #refreshKnown(entry, advertised) {
         entry.timeout = Date.now() + REGISTRY_ENTRY_TIMEOUT;
         if (!Array.isArray(advertised)) {
             return;
         }
-        if (this.#sameAddresses(entry.advertised ?? [], advertised)) {
+        const changed = !this.#sameAddresses(entry.advertised ?? [], advertised);
+        const stuckEmpty = (entry.a?.length ?? 0) === 0 && advertised.length > 0;
+        if (!changed && !stuckEmpty) {
             return;
         }
-        entry.advertised = [...advertised];
+        if (changed) {
+            entry.advertised = [...advertised];
+        }
         this.#reprobe(entry).catch((e) => logger.error("registry.reprobe:", e));
     }
 

package/src/Util.js

@@ -211,6 +211,15 @@ export function getNetworkAddresses() {
     for (const interfaceAddresses of Object.values(interfaces)) {
         if (!interfaceAddresses) { continue; }
         for (const addressObject of interfaceAddresses) {
+            // Skip IPv6 link-local addresses (fe80::/10). They require a zone identifier
+            // (e.g. %eth0) to route, which `fetch()` doesn't accept — advertising them
+            // just wastes a probe slot.
+            if (
+                addressObject.family === "IPv6" &&
+                addressObject.address.toLowerCase().startsWith("fe80:")
+            ) {
+                continue;
+            }
             const addr = addressObject.family === "IPv4"
                 ? addressObject.address
                 : addressObject.family === "IPv6"