*rqlite* is a distributed relational database, which uses [SQLite](https://www.sqlite.org/) as its storage engine. rqlite is written in [Go](http://golang.org/) and uses [Raft](http://raftconsensus.github.io/) to achieve consensus across all the instances of the SQLite databases. rqlite ensures that every change made to the database is made to a quorum of SQLite files, or none at all.
rqlite gives you the functionality of a [rock solid](http://www.sqlite.org/testing.html), fault-tolerant, replicated relational database, but with very easy installation, deployment, and operation. With it you've got a lightweight and reliable distributed store for relational data.
You could use rqlite as part of a larger system, as a central store for some critical relational data, without having to run a heavier solution like MySQL. rqlite might also be an effective way to provide a small number of SQLite read-replicas.
The quickest way to get running on OSX and Linux is to download a pre-built release binary. You can find these binaries on the [Github releases page](https://github.com/otoolep/rqlite/releases). Once installed, you can start a single rqlite node like so:
__If you want to build rqlite__, either because you want the latest code or a pre-built binary for platform is not available, take a look at [CONTRIBUTING.md](https://github.com/otoolep/rqlite/blob/master/CONTRIBUTING.md).
Under each node will be an SQLite file, which should remain in consensus. You can create clusters of any size, but clusters of 3, 5, and 7 nodes are most practical. Clusters larger than this become impractical, due to the number of nodes that must be contacted before a change can take place.
If a node needs to be restarted, perhaps because of failure, don't pass the `-join` option. Using the example nodes above, if node 2 needed to be restarted, do so as follows:
On restart it will rejoin the cluster and apply any changes to the local SQLite database that took place while it was down. Depending on the number of changes in the Raft log, restarts may take a little while.
rqlite exposes an HTTP API allowing the database to be modified such that the changes are replicated. Queries are also executed using the HTTP API, though the SQLite database could be queried directly. Modifications go through the Raft log, ensuring only changes committed by a quorum of rqlite nodes are actually executed against the SQLite database. Queries do not __necessarily__ go through the Raft log, however, since they do not change the state of the database, and therefore do not need to be captured in the log. More on this later.
To write data successfully to the database, you must create at least 1 table. To do this, perform a HTTP POST, with a `CREATE TABLE` SQL command encapsulated in a JSON array, in the body of the request. For example:
The use of the URL param `pretty` is optional, and results in pretty-printed JSON responses. Time is measured in seconds. If you do not want timings, do not pass `timings` as a URL parameter.
A bulk update is contained within a single Raft log entry, so the network round-trips between nodes in the cluster are amortized over the bulk update. This should result in better throughput, if it is possible to use this kind of update.
The behaviour of rqlite when more than 1 query is passed via `q` is undefined. If you want to execute more than one query per HTTP request, perform a POST, and place the queries in the body of the request as a JSON array. For example:
Another approach is to read the database file directly via `sqlite3`, the command-line tool that comes with SQLite. As long as you can be sure the file you access is under the leader, the records returned will be accurate and up-to-date.
**If you use the query API to execute a command that modifies the database, those changes will not be replicated**. Always use the write API for inserts and updates.
Even though serving queries does not require consensus (because the database is not changed), [queries should generally be served by the leader](https://github.com/otoolep/rqlite/issues/5). Why is this? Because without this check queries on a node could return out-of-date results. This could happen for one of two reasons:
With _none_, the node simply queries its local SQLite file, and does not even check if it is leader. This offers the fastest query response, but suffers from the problems listed above. _Weak_ instructs the node to check that it is the leader, before querying the local SQLite file. Checking leader state only involves checking local state, so is still very fast. There is, however, a very small window of time (milliseconds by default) during which the node may return stale data. This is because after the leader check, but before the local SQLite file is read, another node could be elected leader. As result the node may not be up-to-date with the rest of cluster. To avoid even this possibility, rqlite also offers _strong_. In this mode, rqlite sends the query through Raft consensus system, ensuring that the node remains the leader throughout query processing. However, this will involve the leader contacting at least a quorum of nodes, and will therefore increase query response times.
_Weak_ is probably sufficient for most applications, and is the default read consistency level. To explicitly select consistency, set the query param `level`. Examples of enabling each read consistency level for a simple query is shown below.
Transactions are supported. To execute statements within a transaction, add `transaction` to the URL. An example of the above operation executed within a transaction is shown below.
When a transaction takes place either both statements will succeed, or neither. Performance is *much, much* better if multiple SQL INSERTs or UPDATEs are executed via a transaction. Note that processing of the request ceases the moment any single query results in an error.
The behaviour of rqlite when using `BEGIN`, `COMMIT`, or `ROLLBACK` to control transactions is **not defined**. It is important to control transactions only through the query parameters shown above.
rqlite replicates SQLite for fault-tolerance. It does not replicate it for performance. In fact performance is reduced somewhat due to the network round-trips.
Depending on your machine, individual INSERT performance could be anything from 1 operation per second to more than 100 operations per second. However, by using transactions, throughput will increase significantly, often by 2 orders of magnitude. This speed-up is due to the way SQLite works. So for high throughput, execute as many operations as possible within a single transaction.
You can also try using an [in-memory database](https://www.sqlite.org/inmemorydb.html) to increase performance. In this mode no actual SQLite file is created and the entire database is stored in memory.
Using an in-memory does not put your data at risk. Since the Raft log is the authoritative store for all data, and it is written to disk, an in-memory database can be fully recreated on start-up.
A status API exists, which dumps some basic diagnostic and statistical information, as well as basic information about the underlying Raft node. Assuming rqlite is started with default settings, rqlite status is available like so:
rqlite also exports [expvar](http://godoc.org/pkg/expvar/) information. The standard, and some custom information, is exposed. This data can be retrieved like so:
The node can then be restored by loading this database file via `sqlite3` and executing `.dump`. You can then use the output of the dump to replay the entire database back into brand new node (or cluster), *with the exception* of `BEGIN TRANSACTION` and `COMMIT` commands. You should ignore those commands in the `.dump` output.
By default a backup can only be retrieved from the leader, though this check can be disabled by adding `noleader` to the URL as a query param.
rqlite automatically performs log compaction. After a fixed number of changes rqlite snapshots the SQLite database, and truncates the Raft log. This is a technical feature of the Raft consensus system, and most users of rqlite need not be concerned with this.
* Only SQL statements that are __deterministic__ are safe to use with rqlite, because statements are committed to the Raft log before they are sent to each node. For example, the following statement could result in different SQLite databases under each node:
* In case it isn't obvious, rqlite does not replicate any changes made directly to the underlying SQLite files. If you do change these files directly, you will cause rqlite to fail. Only modify the database via the HTTP API.
