RefluxDB is a small timeseries database with UDP and REST interface. It uses a simple SQL dialect to query the data and recognizes a simple line protocol. Its objective is to have a set of simple timeseries primitives and pre-calculated metrics for large datasets that can be run locally and benefit of fast ssd storage.
It is built in Rust and depends on a small set of libraries to provide local storage:
- GlueDB
- Chronos
- Sled
It still in early phases so internal interfaces may change.
$ cargo build
$ cargo run
$ echo "test,host=server,region=us-east1 value=0.80 1234567890000000000"| nc -u 127.0.0.1 8089
curl -X POST -d "q=SELECT * from test" localhost:8086/query
The main components for the database are contained within the timeseries persistence manager.
The basic concepts are the following: Measurement: Unit of data tied to a time, within a timeseries, annotated with tags Timeseries: A set of immutable measurements that move forward in time Resultset: a slice of data from a timeseries within T(start) and T(end) Tags: json annotations to measurements used to filter and group resultsets
Within the databases (root directory), a new sled database will be created for each timeseries. This database is abstracted by GlueSQL to provide query language and functions support.
pros: isolation, parallelism
cons: disk space, migration
id -> UUID
time -> unix timestamp, ordered - measurement time (expanded 11-30-2021)
created_at -> unix timestamp, ordered, system time (added 11-30-2021)
name -> value name (added 11-30-2021)
value -> float (to be float, int, string and bool)
tags -> key/value tag map
"CREATE TABLE <timeseries_name> (id UUID, time TIMESTAMP, created_at TIMESTAMP, name TEXT, value FLOAT, tags MAP);",
* TODO: Immutable data: measurements can't be changed
* TODO: separated tag table: "CREATE TABLE <timeseries_name>_tags (id UUID, key TEXT, value TEXT);",
* TODO: ensure immutability is enforced through measurement id or fingerprint
* TODO: Pre-calculated stats for each series