Glean Migration Deep Dive

Step-by-Step Migration

Phase 1: Prepare

Export all documents from the current search provider and map them to Glean's document schema. Each document needs id, title, url, body, and permissions.

interface GleanDocument {
  id: string;
  datasource: string;
  title: string;
  url: string;
  body: { mimeType: 'text/plain' | 'text/html'; content: string };
  permissions: { allowedUsers?: string[]; allowedGroups?: string[] };
  updatedAt: string;
}

async function exportAndTransform(sourceIndex: string): Promise<GleanDocument[]> {
  const docs: GleanDocument[] = [];
  let scrollId: string | undefined;
  do {
    const batch = scrollId
      ? await sourceClient.scroll({ scroll: '2m', scroll_id: scrollId })
      : await sourceClient.search({ index: sourceIndex, scroll: '2m', size: 500 });
    scrollId = batch._scroll_id;
    for (const hit of batch.hits.hits) {
      docs.push({
        id: hit._id, datasource: 'custom_' + sourceIndex,
        title: hit._source.title, url: hit._source.url,
        body: { mimeType: 'text/plain', content: hit._source.body },
        permissions: { allowedGroups: hit._source.acl_groups ?? ['everyone'] },
        updatedAt: hit._source.updated_at ?? new Date().toISOString(),
      });
    }
  } while (scrollId && docs.length < 1_000_000);
  return docs;
}

Phase 2: Migrate

Create the Glean datasource and bulk-index all transformed documents.

// Create datasource
await gleanClient.post('/api/index/v1/adddatasource', {
  name: 'custom_knowledge_base',
  displayName: 'Knowledge Base',
  homeUrl: 'https://kb.example.com',
  objectDefinitions: [{ name: 'Article', propertyDefinitions: [] }],
});

// Bulk index in batches of 200
const BATCH_SIZE = 200;
for (let i = 0; i < docs.length; i += BATCH_SIZE) {
  const batch = docs.slice(i, i + BATCH_SIZE);
  await gleanClient.post('/api/index/v1/bulkindexdocuments', {
    datasource: 'custom_knowledge_base',
    documents: batch.map(d => ({
      id: d.id, title: d.title, url: d.url, body: d.body,
      permissions: d.permissions, updatedAt: d.updatedAt,
    })),
  });
  console.log(`Indexed ${Math.min(i + BATCH_SIZE, docs.length)}/${docs.length}`);
}

Phase 3: Validate

Run test queries against both old and new systems and compare result quality.

const testQueries = ['onboarding process', 'security policy', 'API authentication'];
for (const query of testQueries) {
  const result = await gleanClient.post('/api/client/v1/search', {
    query, pageSize: 5, requestOptions: { datasourceFilter: 'custom_knowledge_base' },
  });
  console.log(`"${query}" → ${result.results.length} results, top: ${result.results[0]?.title}`);
}

Rollback Plan

# Delete the Glean datasource (removes all indexed documents)
curl -X DELETE "https://your-instance.glean.com/api/index/v1/deletedatasource" \
  -H "Authorization: Bearer $GLEAN_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"name": "custom_knowledge_base"}'

# Re-enable old search provider
export SEARCH_PROVIDER=elasticsearch
echo "Rollback complete — old search provider re-enabled"

Migration Checklist

• Inventory all source indices and document counts
• Map permission model (index-level ACLs to Glean datasource permissions)
• Export and transform documents to Glean schema
• Create datasource via adddatasource API
• Bulk index all documents in batches of 200
• Run validation queries and compare relevance
• Update search UI to use Glean Client API
• Decommission old search crawlers and indices
• Monitor search quality metrics for 7 days post-migration
• Remove old provider SDK dependencies from codebase

Error Handling

Resources

• Glean Developer Portal
• Indexing API
• Search API

Next Steps

After migration, configure Glean's built-in connectors for Google Drive, Confluence, and Slack to replace custom crawlers. See glean-core-workflow-a for search integration.