All reads and writes must be performed from inside a transaction. BuntDB can have one write transaction opened at a time, but can have many concurrent read transactions. Each transaction maintains a stable view of the database. In other words, once a transaction has begun, the data for that transaction cannot be changed by other transactions.
Transactions run in a function that exposes a `Tx` object, which represents the transaction state. While inside a transaction, all database operations should be performed using this object. You should never access the origin `DB` object while inside a transaction. Doing so may have side-effects, such as blocking your application.
When a transaction fails, it will roll back, and revert all changes that occurred to the database during that transaction. There's a single return value that you can use to close the transaction. For read/write transactions, returning an error this way will force the transaction to roll back. When a read/write transaction succeeds all changes are persisted to disk.
A read-only transaction should be used when you don't need to make changes to the data. The advantage of a read-only transaction is that there can be many running concurrently.
```go
err := db.View(func(tx *buntdb.Tx) error {
...
return nil
})
```
### Read/write Transactions
A read/write transaction is used when you need to make changes to your data. There can only be one read/write transaction running at a time. So make sure you close it as soon as you are done with it.
Getting non-existent values will case an `ErrNotFound` error.
### Iterating
All keys/value pairs are ordered in the database by the key. To iterate over the keys:
```go
err := db.View(func(tx *buntdb.Tx) error {
err := tx.Ascend("", func(key, value string) bool{
fmt.Printf("key: %s, value: %s\n", key, value)
})
return err
})
```
There is also `AscendGreaterOrEqual`, `AscendLessThan`, `AscendRange`, `Descend`, `DescendLessOrEqual`, `DescendGreaterThan`, and `DescendRange`. Please see the [documentation](https://godoc.org/github.com/tidwall/buntdb) for more information on these functions.
Initially all data is stored in a single [B-tree](https://en.wikipedia.org/wiki/B-tree) with each item having one key and one value. All of these items are ordered by the key. This is great for quickly getting a value from a key or [iterating](#iterating) over the keys. Feel free to peruse the [B-tree implementation](https://github.com/tidwall/btree).
You can also create custom indexes that allow for ordering and [iterating](#iterating) over values. A custom index also uses a B-tree, but it's more flexible because it allows for custom ordering.
For example, let's say you want to create an index for ordering names:
```go
db.CreateIndex("names", "*", buntdb.IndexString)
```
This will create an index named `names` which stores and sorts all values. The second parameter is a pattern that is used to filter on keys. A `*` wildcard argument means that we want to accept all keys. `IndexString` is a built-in function that performs case-insensitive ordering on the values
Now you can add various names:
```go
db.Update(func(tx *buntdb.Tx) error {
tx.Set("user:0:name", "tom", nil)
tx.Set("user:1:name", "Randi", nil)
tx.Set("user:2:name", "jane", nil)
tx.Set("user:4:name", "Janet", nil)
tx.Set("user:5:name", "Paula", nil)
tx.Set("user:6:name", "peter", nil)
tx.Set("user:7:name", "Terri", nil)
return nil
})
```
Finally you can iterate over the index:
```go
db.View(func(tx *buntdb.Tx) error {
tx.Ascend("names", func(key, val string) bool {
fmt.Printf(buf, "%s %s\n", key, val)
return true
})
return nil
})
```
The output should be:
```
user:2:name jane
user:4:name Janet
user:5:name Paula
user:6:name peter
user:1:name Randi
user:7:name Terri
user:0:name tom
```
The pattern parameter can be used to filter on keys like this:
BuntDB has support for spatial indexes by storing rectangles in an [R-tree](https://en.wikipedia.org/wiki/R-tree). An R-tree is organized in a similar manner as a [B-tree](https://en.wikipedia.org/wiki/B-tree), and both are balanced trees. But, an R-tree is special because it can operate on data that is in multiple dimensions. This is super handy for Geospatial applications.
Then `IndexRect` is a built-in function that converts rect strings to a format that the R-tree can use. It's easy to use this function out of the box, but you might find it better to create a custom one that renders from a different format, such as [Well-known text](https://en.wikipedia.org/wiki/Well-known_text) or [GeoJSON](http://geojson.org/).
To add some lon,lat points to the `fleet` index:
```go
db.Update(func(tx *buntdb.Tx) error {
tx.Set("fleet:0:pos", "[-115.567 33.532]", nil)
tx.Set("fleet:1:pos", "[-116.671 35.735]", nil)
tx.Set("fleet:2:pos", "[-113.902 31.234]", nil)
return nil
})
```
And then you can run the `Intersects` function on the index:
```go
db.View(func(tx *buntdb.Tx) error {
tx.Intersects("fleet", "[-117 30],[-112 36]", func(key, val string) bool {
The bracket syntax `[-117 30],[-112 36]` is unique to BuntDB, and it's how the built-in rectangles are processed, but you are not limited to this syntax. Whatever Rect function you choose to use during `CreateSpatialIndex` will be used to process the parameter, in this case it's `IndexRect`.
Now `mykey` will automatically be deleted after one second. You can remove the TTL by setting the value again with the same key/value, but with the options parameter set to nil.
BuntDB uses an AOF (append-only file) which is a log of all database changes that occur from operations like `Set()` and `Delete()`.
The format of this file looks like:
```
set key:1 value1
set key:2 value2
set key:1 value3
del key:2
...
```
When the database opens again, it will read back the aof file and process each command in exact order. This read process happens one time when the database opens. From there on the file is only appended.
As you may guess this log file can grow large over time. There is a `Shrink()` function which will rewrite the aof file so that it contains only the items in the database. The shrink operation does not lock up the database so read and write transactions can continue while shrinking is in process.
Also there's the database config setting `Config.AutoShrink` which is used to allow for shrinking to be self-managed. This value is set to a multiple that represents how many more entries in the aof file versus the number of items in memory. For example; If this value is set to 10 and the number of item in memory is 150,000, then the database will automatically shrink when the aof file has 1,500,000 lines in it. Currently default value is 5, but this may change in a future release. Autoshink can be disabled by setting this value to zero.
BuntDB executes an `fsync` once every second on the [aof file](#append-only-file). Which simply means that there's a chance that up to 1 second of data might be lost. The likelihood of this happening is dependent on a lot of factors, but suffice to say this may not be good enough for environments that require very high durability.
Here are some example [benchmarks](https://github.com/tidwall/raft-buntdb#raftstore-performance-comparison) when using BuntDB in a Raft Store implementation.
