diff --git a/services/document-updater/app/js/ProjectFlusher.js b/services/document-updater/app/js/ProjectFlusher.js
index d6b8c5b13304..fb6687417a64 100644
--- a/services/document-updater/app/js/ProjectFlusher.js
+++ b/services/document-updater/app/js/ProjectFlusher.js
@@ -96,6 +96,7 @@ async function flushAllProjects(options) {
 }
 
 module.exports = {
+  _getKeys,
   _extractIds,
   promises: {
     flushAllProjects,
diff --git a/services/document-updater/scripts/flush_docs_with_pending_updates.js b/services/document-updater/scripts/flush_docs_with_pending_updates.js
new file mode 100644
index 000000000000..1be0f3c6acf1
--- /dev/null
+++ b/services/document-updater/scripts/flush_docs_with_pending_updates.js
@@ -0,0 +1,281 @@
+// Flush docs that were left with a stuck PendingUpdates queue in redis after
+// their project was flushed (and possibly hard deleted since).
+//
+// For each doc with a PendingUpdates queue, the project id is looked up in
+// db.docs, then the pending updates are processed and the doc is flushed and
+// removed from redis. Docs whose project cannot be found (e.g. hard deleted
+// projects) are logged and can be discarded with --discard-hard-deleted.
+
+const minimist = require('minimist')
+const Settings = require('@overleaf/settings')
+const OError = require('@overleaf/o-error')
+const DocumentManager = require('../app/js/DocumentManager')
+const HistoryManager = require('../app/js/HistoryManager')
+const ProjectFlusher = require('../app/js/ProjectFlusher')
+const { rclient } = require('../app/js/RedisManager')
+const { db, ObjectId, mongoClient } = require('../app/js/mongodb')
+
+const keys = Settings.redis.documentupdater.key_schema
+
+const summary = {
+  totalDocs: 0,
+  flushedDocs: 0,
+  missingDocs: 0,
+  discardedDocs: 0,
+  erroredDocs: 0,
+}
+
+function printSummary() {
+  console.log(
+    `Processed ${summary.totalDocs} docs with pending updates: flushed=${summary.flushedDocs} missing=${summary.missingDocs} discarded=${summary.discardedDocs} errored=${summary.erroredDocs}`
+  )
+}
+
+function usage() {
+  console.error(`
+Usage: node scripts/flush_docs_with_pending_updates.js [options]
+
+Options:
+  --estimated-docs        Required. Estimated number of documents in db.docs,
+                          used as a safety check that the right database is
+                          consulted for the doc lookups. The script aborts
+                          unless the estimated collection count is within 20%
+                          of this value.
+  --log-unflushed         When the db.docs lookup failed for a doc, log its
+                          pending updates and, if the doc has unflushed
+                          changes, all its doc keys
+                          (default: true, disable via --log-unflushed=false)
+  --discard-hard-deleted  Discard the stuck queue and the other doc keys of
+                          docs whose db.docs lookup failed (default: false)
+  --docId                 Process a single doc instead of scanning redis
+  --projectId             Override the project id for --docId, skipping the
+                          db.docs lookup
+  --dry-run               Only scan and log, without flushing, deleting or
+                          discarding anything (default: false)
+  --limit                 Maximum number of PendingUpdates keys to scan
+                          (default: 100000)
+  --help                  Show this help message
+`)
+}
+
+async function getDocIds(options) {
+  if (options.docId) {
+    return [options.docId]
+  }
+  const docKeys = await ProjectFlusher._getKeys(
+    keys.pendingUpdates({ doc_id: '*' }),
+    options.limit
+  )
+  return ProjectFlusher._extractIds(docKeys)
+}
+
+async function processDoc(docId, options) {
+  let projectId = options.projectId
+  if (!projectId) {
+    const doc = await db.docs.findOne(
+      { _id: new ObjectId(docId) },
+      { projection: { project_id: 1 } }
+    )
+    projectId = doc?.project_id?.toString()
+  }
+
+  if (projectId) {
+    await flushDoc(projectId, docId, options)
+  } else {
+    await handleMissingDoc(docId, options)
+  }
+}
+
+async function flushDoc(projectId, docId, options) {
+  const queueLength = await rclient.llen(keys.pendingUpdates({ doc_id: docId }))
+  if (options.dryRun) {
+    console.log(
+      `Would flush doc ${docId} in project ${projectId} with ${queueLength} pending updates`
+    )
+    return
+  }
+  console.log(
+    `Flushing doc ${docId} in project ${projectId} with ${queueLength} pending updates`
+  )
+  await DocumentManager.promises.flushAndDeleteDocWithLock(projectId, docId, {})
+  await HistoryManager.promises.flushProjectChanges(projectId, {})
+  const remaining = await rclient.llen(keys.pendingUpdates({ doc_id: docId }))
+  if (remaining > 0) {
+    console.error(
+      `Doc ${docId} in project ${projectId} still has ${remaining} pending updates after flushing`
+    )
+    summary.erroredDocs++
+  } else {
+    summary.flushedDocs++
+  }
+}
+
+async function handleMissingDoc(docId, options) {
+  summary.missingDocs++
+  let message = `Doc ${docId} not found in db.docs`
+
+  if (options.logUnflushed) {
+    // Log the pending updates: they are the reason this doc is processed and
+    // would be lost when discarding it. These can exist independently of any
+    // unflushed doc content.
+    const pendingUpdates = await rclient.lrange(
+      keys.pendingUpdates({ doc_id: docId }),
+      0,
+      -1
+    )
+    message += `, pending updates: ${JSON.stringify(pendingUpdates)}`
+
+    // In addition, log the unflushed doc content if there is any.
+    const unflushedTime = await rclient.get(
+      keys.unflushedTime({ doc_id: docId })
+    )
+    if (unflushedTime) {
+      const doc = {
+        unflushedTime,
+        projectId: await rclient.get(keys.projectKey({ doc_id: docId })),
+        lines: await rclient.get(keys.docLines({ doc_id: docId })),
+        version: await rclient.get(keys.docVersion({ doc_id: docId })),
+        hash: await rclient.get(keys.docHash({ doc_id: docId })),
+        ranges: await rclient.get(keys.ranges({ doc_id: docId })),
+        pathname: await rclient.get(keys.pathname({ doc_id: docId })),
+        projectHistoryId: await rclient.get(
+          keys.projectHistoryId({ doc_id: docId })
+        ),
+        lastUpdatedAt: await rclient.get(keys.lastUpdatedAt({ doc_id: docId })),
+        lastUpdatedBy: await rclient.get(keys.lastUpdatedBy({ doc_id: docId })),
+        resolvedCommentIds: await rclient.smembers(
+          keys.resolvedCommentIds({ doc_id: docId })
+        ),
+      }
+      message += `, unflushed doc keys: ${JSON.stringify(doc)}`
+    }
+  }
+  console.warn(message)
+
+  if (options.discardHardDeleted && !options.dryRun) {
+    await discardDoc(docId)
+    console.warn(`Discarded redis keys of doc ${docId}`)
+    summary.discardedDocs++
+  }
+}
+
+async function discardDoc(docId) {
+  // Purge the project-level entries first: in case the script crashes mid-way,
+  // a re-run can still resolve the project id from the ProjectId doc key.
+  const projectId = await rclient.get(keys.projectKey({ doc_id: docId }))
+  if (projectId) {
+    await rclient.srem(keys.docsInProject({ project_id: projectId }), docId)
+  }
+  await rclient.srem(keys.historyRangesSupport(), docId)
+  await rclient.del(
+    keys.pendingUpdates({ doc_id: docId }),
+    keys.docLines({ doc_id: docId }),
+    keys.projectKey({ doc_id: docId }),
+    keys.docVersion({ doc_id: docId }),
+    keys.docHash({ doc_id: docId }),
+    keys.ranges({ doc_id: docId }),
+    keys.pathname({ doc_id: docId }),
+    keys.projectHistoryId({ doc_id: docId }),
+    keys.unflushedTime({ doc_id: docId }),
+    keys.lastUpdatedBy({ doc_id: docId }),
+    keys.lastUpdatedAt({ doc_id: docId }),
+    keys.resolvedCommentIds({ doc_id: docId }),
+    keys.docOps({ doc_id: docId })
+  )
+}
+
+async function main() {
+  const argv = minimist(process.argv.slice(2), {
+    boolean: ['log-unflushed', 'discard-hard-deleted', 'dry-run', 'help'],
+    string: ['docId', 'projectId'],
+    default: { limit: 100_000, 'log-unflushed': true },
+  })
+
+  if (argv.help) {
+    usage()
+    return 0
+  }
+
+  const options = {
+    estimatedDocs: argv['estimated-docs'],
+    logUnflushed: argv['log-unflushed'],
+    discardHardDeleted: argv['discard-hard-deleted'],
+    docId: argv.docId,
+    projectId: argv.projectId,
+    dryRun: argv['dry-run'],
+    limit: argv.limit,
+  }
+
+  if (!Number.isInteger(options.estimatedDocs) || options.estimatedDocs <= 0) {
+    console.error('--estimated-docs must be set to a positive integer')
+    usage()
+    return 1
+  }
+  if (options.docId && !ObjectId.isValid(options.docId)) {
+    console.error('--docId is not a valid object id')
+    usage()
+    return 1
+  }
+  if (options.projectId && !options.docId) {
+    console.error('--projectId can only be used together with --docId')
+    usage()
+    return 1
+  }
+  if (options.projectId && !ObjectId.isValid(options.projectId)) {
+    console.error('--projectId is not a valid object id')
+    usage()
+    return 1
+  }
+
+  const actualDocs = await db.docs.estimatedDocumentCount()
+  if (
+    actualDocs < 0.8 * options.estimatedDocs ||
+    actualDocs > 1.2 * options.estimatedDocs
+  ) {
+    console.error(
+      `Aborting: the estimated db.docs count of ${actualDocs} is not within 20% of --estimated-docs=${options.estimatedDocs}, is this the right database?`
+    )
+    return 1
+  }
+
+  const docIds = await getDocIds(options)
+  const limitReached = !options.docId && docIds.length >= options.limit
+  for (const docId of docIds) {
+    summary.totalDocs++
+    try {
+      await processDoc(docId, options)
+    } catch (error) {
+      console.error(`Error processing doc ${docId}`)
+      console.error(OError.getFullStack(error))
+      summary.erroredDocs++
+    }
+  }
+
+  if (limitReached) {
+    console.error(
+      `Hit the scan limit of ${options.limit} keys; there may be more docs with pending updates. Re-run to process the rest.`
+    )
+    return 2
+  }
+  if (options.dryRun) {
+    return summary.totalDocs > 0 ? 2 : 0
+  }
+  if (summary.erroredDocs > 0 || summary.missingDocs > summary.discardedDocs) {
+    return 2
+  }
+  return 0
+}
+
+main()
+  .then(code => {
+    printSummary()
+    rclient.quit()
+    mongoClient.close()
+    process.exit(code)
+  })
+  .catch(error => {
+    printSummary()
+    console.error(OError.getFullStack(error))
+    console.error(OError.getFullInfo(error))
+    process.exit(1)
+  })