@acarmisc/backstage-plugin-litellm-backend 0.11.0 → 0.11.1

package/dist/router.js

@@ -46,6 +46,44 @@ const bridge_1 = require("./bridge");
 const permissions_1 = require("./permissions");
 const teamAdmin_1 = require("./teamAdmin");
 const teamCreateInFlight = new Map();
+/** Backoff schedule for `withTeamFetchRetry` — 3 retries over ~1.3s total. */
+const TEAM_FETCH_RETRY_DELAYS_MS = [100, 300, 900];
+/**
+ * Only 5xx (and network/timeout) failures are treated as transient — a 4xx
+ * like "team not found" or "forbidden" is a deterministic answer from the
+ * upstream and retrying it would just add latency for the same result.
+ */
+function isRetryableTeamFetchError(err) {
+    if (err instanceof client_1.LiteLLMUpstreamError) {
+        return err.status >= 500;
+    }
+    return true;
+}
+/**
+ * Retries a LiteLLM team-info fetch with the given backoff schedule.
+ *
+ * LiteLLM proxies are commonly run behind multiple replicas; a request can
+ * land on a replica that hasn't yet caught up with a very recent write
+ * (team create/update, membership change), or on one that is transiently
+ * unhealthy. Both surface as a failed `getTeamInfo` call that would
+ * otherwise be swallowed (e.g. GET /teams silently drops the team from the
+ * response) or bubble up as a spurious 500/404 right after a write this
+ * same request just made. A short retry smooths over that window without
+ * masking a genuine, persistent failure.
+ */
+async function withTeamFetchRetry(fn, delaysMs = TEAM_FETCH_RETRY_DELAYS_MS) {
+    for (let attempt = 0;; attempt++) {
+        try {
+            return await fn();
+        }
+        catch (err) {
+            if (attempt >= delaysMs.length || !isRetryableTeamFetchError(err)) {
+                throw err;
+            }
+            await new Promise(resolve => setTimeout(resolve, delaysMs[attempt]));
+        }
+    }
+}
 async function createRouter(options) {
     const { config, logger, auth, discovery, permissions } = options;
     const baseUrl = config.getString('litellm.baseUrl');
@@ -510,8 +548,8 @@ async function createRouter(options) {
                 res.json([]);
                 return;
             }
-            const teams = await Promise.all(userInfo.teams.map(teamId => client.getTeamInfo(teamId).catch(err => {
-                logger.warn(`Failed to fetch team ${teamId}: ${err.message}`);
+            const teams = await Promise.all(userInfo.teams.map(teamId => withTeamFetchRetry(() => client.getTeamInfo(teamId)).catch(err => {
+                logger.warn(`Failed to fetch team ${teamId} after retries: ${err.message}`);
                 return null;
             })));
             res.json(teams.filter(Boolean));
@@ -773,7 +811,7 @@ async function createRouter(options) {
         }
         let existing;
         try {
-            existing = await client.getTeamInfo(teamId);
+            existing = await withTeamFetchRetry(() => client.getTeamInfo(teamId));
         }
         catch (err) {
             if (err instanceof client_1.LiteLLMUpstreamError && err.status === 404) {
@@ -885,7 +923,7 @@ async function createRouter(options) {
                 member: litellmUserId,
                 owningGroup,
             });
-            const updated = await client.getTeamInfo(teamId);
+            const updated = await withTeamFetchRetry(() => client.getTeamInfo(teamId));
             res.json(updated);
         }
         catch (err) {
@@ -920,7 +958,7 @@ async function createRouter(options) {
                 member: litellmUserId,
                 owningGroup,
             });
-            const updated = await client.getTeamInfo(teamId);
+            const updated = await withTeamFetchRetry(() => client.getTeamInfo(teamId));
             res.json(updated);
         }
         catch (err) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@acarmisc/backstage-plugin-litellm-backend",
-  "version": "0.11.0",
+  "version": "0.11.1",
   "description": "The Backstage backend plugin for LiteLLM governance",
   "backstage": {
     "role": "backend-plugin",