OUT-3584: stream Dropbox→Assembly uploads to prevent OOM crash

src/features/sync/lib/Sync.service.ts

@@ -268,23 +268,20 @@ export class SyncService extends AuthenticatedDropboxService {
   private async uploadFileInAssembly(dbxPath: string, uploadUrl: string, copilotApi: CopilotAPI) {
     logger.info('SyncService#uploadFileInAssembly :: Uploading file to Assembly', dbxPath)
 
-    const dbx = this.dbxClient.getDropboxClient()
-    const fileMetaData = await dbx.filesDownload({ path: dbxPath }) // get metadata for the files
-    logger.info('SyncService#uploadFileInAssembly :: File metadata downloaded', dbxPath)
-
-    const downloadBody = await this.dbxClient.downloadFile({
+    // Stream the file directly from Dropbox into Assembly's S3 upload URL.
+    // `contentLength` comes from the download response headers, guaranteeing it
+    // matches the exact bytes in the stream. Avoids the Dropbox SDK's
+    // `filesDownload` which buffers the full file in memory (OOMs on videos).
+    const { body: downloadBody, contentLength } = await this.dbxClient.downloadFile({
       urlPath: DBX_URL_PATH.fileDownload,
       filePath: dbxPath,
       rootNamespaceId: z.string().parse(this.connectionToken.rootNamespaceId),
+      refreshToken: this.connectionToken.refreshToken,
     })
     logger.info('SyncService#uploadFileInAssembly :: Found downloadBody', Boolean(downloadBody))
 
     // upload file to assembly
-    const fileUploadResp = await copilotApi.uploadFile(
-      uploadUrl,
-      fileMetaData.result.size.toString(),
-      downloadBody,
-    )
+    const fileUploadResp = await copilotApi.uploadFile(uploadUrl, contentLength, downloadBody)
     logger.info('SyncService#uploadFileInAssembly :: File uploaded to Assembly', dbxPath)
 
     if (fileUploadResp.status !== httpStatus.OK) {
@@ -430,6 +427,7 @@ export class SyncService extends AuthenticatedDropboxService {
         filePath: path,
         body: resp.body,
         rootNamespaceId: z.string().parse(this.connectionToken.rootNamespaceId),
+        refreshToken: this.connectionToken.refreshToken,
       })
       logger.info('SyncService#uploadFileInDropbox :: File uploaded to', path)
       return {

src/features/workers/resync-failed-files/helper/resync-failed-files.helper.ts

@@ -88,7 +88,9 @@ const processFailedSync = async (
   if (file && file.status !== 'pending') return
 
   const fileInDropbox = await getFileFromDropbox(dbxClient, failedSync.dbxFileId ?? '')
-  if (!fileInDropbox) return
+  // Guard against non-file entries: a mapped dbxFileId could resolve to a folder
+  // or deleted entry after a Dropbox rename/delete-and-recreate. Only resync files.
+  if (!fileInDropbox || fileInDropbox['.tag'] !== 'file') return
 
   const channelSync = await db.query.channelSync.findFirst({
     where: (channelSync, { eq }) => eq(channelSync.id, failedSync.channelSyncId),

src/lib/dropbox/DropboxClient.ts

@@ -108,11 +108,18 @@ export class DropboxClient {
     urlPath,
     filePath,
     rootNamespaceId,
+    refreshToken,
   }: {
     urlPath: string
     filePath: string
     rootNamespaceId: string
-  }) {
+    refreshToken: string
+  }): Promise<{ body: NodeJS.ReadableStream | null; contentLength: string }> {
+    // Ensure a valid access token before the download. The DropboxAuth instance
+    // is created per-task with only the refresh token set; without an explicit
+    // refresh here, `getAccessToken()` returns undefined and Dropbox 400s.
+    await this.dbxAuthClient.refreshAccessToken(refreshToken)
+
     const headers = {
       Authorization: `Bearer ${this.dbxAuthClient.authInstance.getAccessToken()}`,
       'Dropbox-API-Path-Root': dropboxArgHeader({
@@ -126,7 +133,19 @@ export class DropboxClient {
       throw new DropboxResponseError(response.status, response.headers, {
         error_summary: 'DropboxClient#downloadFile. Failed to download file', // following the dropbox response error convention with snake case
       })
-    return response.body
+
+    // Use the Content-Length from the download response as the upload size.
+    // This is the exact byte count about to be streamed, guaranteed to match
+    // what the upstream (S3) sees — unlike the size in `filesListFolder` entries,
+    // which can diverge from the downloaded stream for certain file types.
+    const contentLength = response.headers.get('content-length')
+    if (!contentLength) {
+      throw new Error(
+        `DropboxClient#downloadFile. Missing Content-Length header for file: ${filePath}`,
+      )
+    }
+
+    return { body: response.body, contentLength }
   }
 
   /**
@@ -138,12 +157,20 @@ export class DropboxClient {
     filePath,
     body,
     rootNamespaceId,
+    refreshToken,
   }: {
     urlPath: string
     filePath: string
     body: NodeJS.ReadableStream | null
     rootNamespaceId: string
+    refreshToken: string
   }): Promise<DropboxFileMetadata> {
+    // Explicit token refresh — mirrors `_downloadFile`. Previously relied on
+    // an implicit refresh from a preceding SDK call (e.g. filesGetMetadata),
+    // which is a fragile contract — any future caller reaching this method
+    // without a prior SDK call would 400 on an unpopulated Bearer.
+    await this.dbxAuthClient.refreshAccessToken(refreshToken)
+
     const args = {
       path: filePath,
       autorename: false,