Paging That Moves Keys, Not Documents
Asking a cluster for a page of results used to make every shard load full documents, only to discard most of them in the final merge. 0.3.7 turns that into a two-phase exchange. The shards trade lightweight sorting keys, which is all it takes to settle the page boundaries, and the document payload is then fetched exactly once.
Warm query latency dropped by more than five times, and memory overhead fell with it.
Ingestion That Reads Your Source First
The data loader now understands a file before it commits it. Pass --report and you get a profile of your source: row counts, repeating columns and unique pairs, so you know how the data will map before a single document is written.
Composite keys are supported natively and recorded in the schema, so later loads follow the same identity and never overwrite a document by accident. The client guide shows the loader end to end.
Recovery in Milliseconds, Disk You Can Reclaim
After a crash, a node used to repair its file in a pass that could take minutes. 0.3.7 checkpoints the allocator state on every commit, so recovery now takes milliseconds.
Disk space has explicit routes back as well. Background index merges and the new rewrite command reclaim the space that deleted records were holding.
A Native Windows Service
Windows now deploys exactly as Linux does. The new MSI installs the engine as an isolated Windows service under a virtual account, restarts it automatically after a failure, rotates the log daily, and keeps its configuration intact across binary upgrades.
Get It
Everything in this post ships in CameoDB 0.3.8, the current release, which adds certificate rotation without a restart, signed artifacts and CIDR search. The v0.3.8 post has the details.
Related reading
Introducing CameoDB.ai: Your Intelligence Layer, Installed
We install the intelligence layer on your own data, then hand your team the platform to grow it.
Next postCameoDB v0.3.5: Commits on a Clock and 370,000 Docs a Second
Writes become searchable on a steady clock, and bulk ingest climbs to 370,000 documents a second.