- A c++ datastore based on leveldb
- build both desktop and server apps in one common interface
- - stored procedures via JNI
transactional, serialized operations
- - http rpc
the stored procedures are exposed via http(s)
- - cross platform
- - replication
single master, multiple slaves (only linux machines can be master)
- - hot-backup
linux-only, can be used to provision/fast-track a new slave by sending the checkpoint/backup as it's initial data
- - realtime updates
the changes/updates received by a slave can be forwarded to clients (desktop/mobile/browser) via websockets
- - portable
can be deployed as a self-contained linux executable with a bundled 12MB JRE
- - embeddable as a shared library
for migration tools and testing
- - automatic views
visitors for your declared indexes are generated from the DSL (rpc included)
- - automatic ops
simple documents can have their create/update/delete ops fully generated from the DSL (rpc included)
- 4 types of secondary indexes
- - simple (based on a single field)
- - compound (based on multiple fields)
- - unique (simple or compound)
- - clustered (simple or compound)
“A clustering index maintains a copy of the entire row, not just the primary key. As a result, when querying on a clustering key lookups in the primary index are always avoided.
A clustering index is a covering index for any query. Another way to think of a clustering index is that it is a materialization of the table sorted in another order.”
Inspired by couchdb, whose author Damien Katz mentioned about being able to "touch your data"
back in a 2009 presentation.
“I was so excited when I first started testing it out.
Because a big goal with CouchDB was that you would be able to feel like you could touch your data,
like it was right there in your hand.
There’re certain tools where you have this responsiveness and you don’t feel like
there’s all these layers between you that are obfuscating what you’re trying to get at.”
(business logic, validation, map-reduce secondary indexes) and stored your data in json format.
It had distributed multi-master sync which meant not having support for multi-document transactions.
For my usecase, having the latter was required.
And so protostuffdb was born with the goal of being able to touch your data (like couchdb) while supporting
This stack is c/c++ with java stored procedures and stores your data in protostuff binary format.
Both validation and secondary indexes are implemented declaratively using a custom DSL on top of
where the schema of your data is defined. These files are compiled by fbsgen-ds.