From 67266ba57b74d36df5c516f1a3059728639f7875 Mon Sep 17 00:00:00 2001
From: Adam Moussa <adam@seahavenind.com>
Date: Mon, 8 Jun 2026 14:58:27 -0400
Subject: [PATCH] Fix notion-sync header crash on non-ASCII titles

S3 user-metadata values are sent as HTTP headers, which must be
US-ASCII. Notion page titles routinely contain em dashes and other
non-ASCII characters, causing ERR_INVALID_CHAR and a daily sync
failure. Sanitize the notion-title metadata to printable ASCII.

Also treat a Bedrock ConflictException from StartIngestionJob as
benign (an ingestion job is already running and will pick up the
freshly uploaded files) instead of failing the whole invocation.
---
 lambda/notion-sync/index.ts | 45 ++++++++++++++++++++++++++++---------
 1 file changed, 35 insertions(+), 10 deletions(-)

diff --git a/lambda/notion-sync/index.ts b/lambda/notion-sync/index.ts
index 696e5eb..7837152 100644
--- a/lambda/notion-sync/index.ts
+++ b/lambda/notion-sync/index.ts
@@ -44,6 +44,17 @@ function getPageTitle(page: Record<string, any>): string {
   return page.id as string;
 }
 
+// S3 user-metadata values travel as HTTP headers, which must be US-ASCII with no
+// control characters. Notion titles routinely contain em dashes and other
+// non-ASCII characters (ERR_INVALID_CHAR otherwise). Strip to safe printable
+// ASCII and cap at the 256-char metadata limit.
+function sanitizeMetadataValue(value: string): string {
+  return value
+    .replace(/[^\x20-\x7E]/g, '') // drop non-printable / non-ASCII
+    .trim()
+    .slice(0, 256);
+}
+
 async function clearOldFiles(bucket: string): Promise<void> {
   let continuationToken: string | undefined;
   const toDelete: { Key: string }[] = [];
@@ -128,7 +139,7 @@ export const handler = async (): Promise<void> => {
           ContentType: 'text/markdown',
           Metadata: {
             'notion-page-id': pageId,
-            'notion-title': title.slice(0, 256), // S3 metadata value limit
+            'notion-title': sanitizeMetadataValue(title),
           },
         }),
       );
@@ -144,13 +155,27 @@ export const handler = async (): Promise<void> => {
   console.log(`Upload complete. ${synced} synced, ${failed} failed.`);
 
   console.log('Triggering Bedrock KB ingestion job...');
-  const ingestionRes = await bedrockAgent.send(
-    new StartIngestionJobCommand({
-      knowledgeBaseId: kbId,
-      dataSourceId: dsId,
-    }),
-  );
-  console.log(
-    `Ingestion job started: ${ingestionRes.ingestionJob?.ingestionJobId}`,
-  );
+  try {
+    const ingestionRes = await bedrockAgent.send(
+      new StartIngestionJobCommand({
+        knowledgeBaseId: kbId,
+        dataSourceId: dsId,
+      }),
+    );
+    console.log(
+      `Ingestion job started: ${ingestionRes.ingestionJob?.ingestionJobId}`,
+    );
+  } catch (err) {
+    // A ConflictException means an ingestion job is already running for this data
+    // source (e.g. an overlapping run). The freshly uploaded files will be picked
+    // up by that in-flight job, so this is benign — log and exit cleanly rather
+    // than failing the whole invocation.
+    if (err instanceof Error && err.name === 'ConflictException') {
+      console.log(
+        'Ingestion job already in progress; uploaded files will be picked up by the running job. Skipping.',
+      );
+      return;
+    }
+    throw err;
+  }
 };
-- 
2.50.1