Vectorize now reflects your writes in under 30 seconds!
Heya, it's me, Shii-chan! Today I've got a small-but-mighty update: the vector database just got faster!
Cloudflare Changelog
developers.cloudflare.com
What was announced?
Cloudflare's Changelog brought an update to Vectorize, its vector database. The end-to-end latency for a vector change to become queryable has dropped a lot: median latency went from 2 minutes to under 30 seconds, and p99 from 5 minutes to under 2 minutes.
The story so far
On Vectorize, an insert, upsert, or delete used to take a little while before it showed up in query results. Before this change, that was about 2 minutes at the median and up to 5 minutes at p99. So a freshly added embedding wouldn't appear in search right away.
What changes
Now those changes land faster, so data you just wrote is searchable almost immediately. That's great for workloads that want the freshest data, like semantic search, recommendations, and RAG (retrieval-augmented generation). And you don't need to change any code or configuration to get it.
Dive Deep
The speedup comes from greatly improving the throughput of the Vectorize write-ahead log (WAL). The WAL records writes first and then applies them, so when it gets congested, changes take longer to appear. Widening that bottleneck made inserts, upserts, and deletes all reflect faster. For more, see the Vectorize documentation.
Wrap-up
- Vectorize cut the latency for vector changes to become queryable (Cloudflare's Changelog)
- Median 2 min to under 30 sec; p99 5 min to under 2 min
- Applies to insert, upsert, and delete; no code or config changes needed
- Perfect for anyone building RAG, semantic search, or recommendations where freshness matters!
