openethereum/util/src/kvdb.rs

1022 lines
31 KiB
Rust
Raw Normal View History

// Copyright 2015-2017 Parity Technologies (UK) Ltd.
2016-02-18 03:46:34 +01:00
// This file is part of Parity.
// Parity is free software: you can redistribute it and/or modify
// it under the terms of the GNU General Public License as published by
// the Free Software Foundation, either version 3 of the License, or
// (at your option) any later version.
// Parity is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
// You should have received a copy of the GNU General Public License
// along with Parity. If not, see <http://www.gnu.org/licenses/>.
2016-04-06 10:07:24 +02:00
//! Key-Value store abstraction with `RocksDB` backend.
2016-02-18 03:46:34 +01:00
use std::io::ErrorKind;
use std::marker::PhantomData;
use std::path::PathBuf;
use common::*;
use elastic_array::*;
2016-10-26 13:53:47 +02:00
use hashdb::DBValue;
use rlp::{UntrustedRlp, RlpType, Compressible};
use rocksdb::{DB, Writable, WriteBatch, WriteOptions, IteratorMode, DBIterator,
2016-10-14 14:44:11 +02:00
Options, DBCompactionStyle, BlockBasedOptions, Direction, Cache, Column, ReadOptions};
#[cfg(target_os = "linux")]
use regex::Regex;
#[cfg(target_os = "linux")]
use std::process::Command;
#[cfg(target_os = "linux")]
use std::fs::File;
2016-02-18 03:46:34 +01:00
const DB_BACKGROUND_FLUSHES: i32 = 2;
const DB_BACKGROUND_COMPACTIONS: i32 = 2;
2017-05-02 11:40:03 +02:00
const DB_WRITE_BUFFER_SIZE: usize = 2048 * 1000;
2016-06-20 21:45:24 +02:00
2017-04-19 14:58:19 +02:00
/// Required length of prefixes.
pub const PREFIX_LEN: usize = 12;
2016-02-18 03:46:34 +01:00
/// Write transaction. Batches a sequence of put/delete operations for efficiency.
#[derive(Default, Clone, PartialEq)]
2016-02-18 03:46:34 +01:00
pub struct DBTransaction {
ops: Vec<DBOp>,
}
#[derive(Clone, PartialEq)]
enum DBOp {
Insert {
col: Option<u32>,
key: ElasticArray32<u8>,
2016-10-26 13:53:47 +02:00
value: DBValue,
},
InsertCompressed {
col: Option<u32>,
key: ElasticArray32<u8>,
2016-10-26 13:53:47 +02:00
value: DBValue,
},
Delete {
col: Option<u32>,
key: ElasticArray32<u8>,
}
2016-03-11 10:57:58 +01:00
}
2016-02-18 03:46:34 +01:00
impl DBTransaction {
/// Create new transaction.
pub fn new() -> DBTransaction {
DBTransaction::with_capacity(256)
}
/// Create new transaction with capacity.
pub fn with_capacity(cap: usize) -> DBTransaction {
DBTransaction {
ops: Vec::with_capacity(cap)
}
2016-02-18 03:46:34 +01:00
}
2017-05-20 11:22:35 +02:00
/// Insert a key-value pair in the transaction. Any existing value will be overwritten upon write.
pub fn put(&mut self, col: Option<u32>, key: &[u8], value: &[u8]) {
let mut ekey = ElasticArray32::new();
ekey.append_slice(key);
self.ops.push(DBOp::Insert {
col: col,
key: ekey,
2016-10-26 13:53:47 +02:00
value: DBValue::from_slice(value),
});
}
2017-05-20 11:22:35 +02:00
/// Insert a key-value pair in the transaction. Any existing value will be overwritten upon write.
pub fn put_vec(&mut self, col: Option<u32>, key: &[u8], value: Bytes) {
let mut ekey = ElasticArray32::new();
ekey.append_slice(key);
self.ops.push(DBOp::Insert {
col: col,
key: ekey,
2016-10-26 13:53:47 +02:00
value: DBValue::from_vec(value),
});
}
2017-05-20 11:22:35 +02:00
/// Insert a key-value pair in the transaction. Any existing value will be overwritten upon write.
/// Value will be RLP-compressed on flush
pub fn put_compressed(&mut self, col: Option<u32>, key: &[u8], value: Bytes) {
let mut ekey = ElasticArray32::new();
ekey.append_slice(key);
self.ops.push(DBOp::InsertCompressed {
col: col,
key: ekey,
2016-10-26 13:53:47 +02:00
value: DBValue::from_vec(value),
});
2016-02-18 03:46:34 +01:00
}
/// Delete value by key.
pub fn delete(&mut self, col: Option<u32>, key: &[u8]) {
let mut ekey = ElasticArray32::new();
ekey.append_slice(key);
self.ops.push(DBOp::Delete {
col: col,
key: ekey,
});
2016-02-18 03:46:34 +01:00
}
}
enum KeyState {
2016-10-26 13:53:47 +02:00
Insert(DBValue),
InsertCompressed(DBValue),
Delete,
}
/// Generic key-value database.
///
/// This makes a distinction between "buffered" and "flushed" values. Values which have been
/// written can always be read, but may be present in an in-memory buffer. Values which have
/// been flushed have been moved to backing storage, like a RocksDB instance. There are certain
/// operations which are only guaranteed to operate on flushed data and not buffered,
/// although implementations may differ in this regard.
///
/// The contents of an interior buffer may be explicitly flushed using the `flush` method.
///
/// The `KeyValueDB` also deals in "column families", which can be thought of as distinct
/// stores within a database. Keys written in one column family will not be accessible from
/// any other. The number of column families must be specified at initialization, with a
/// differing interface for each database. The `None` argument in place of a column index
/// is always supported.
///
/// The API laid out here, along with the `Sync` bound implies interior synchronization for
/// implementation.
pub trait KeyValueDB: Sync + Send {
/// Helper to create a new transaction.
fn transaction(&self) -> DBTransaction { DBTransaction::new() }
/// Get a value by key.
fn get(&self, col: Option<u32>, key: &[u8]) -> Result<Option<DBValue>, String>;
/// Get a value by partial key. Only works for flushed data.
fn get_by_prefix(&self, col: Option<u32>, prefix: &[u8]) -> Option<Box<[u8]>>;
/// Write a transaction of changes to the buffer.
fn write_buffered(&self, transaction: DBTransaction);
/// Write a transaction of changes to the backing store.
fn write(&self, transaction: DBTransaction) -> Result<(), String> {
self.write_buffered(transaction);
self.flush()
}
/// Flush all buffered data.
fn flush(&self) -> Result<(), String>;
/// Iterate over flushed data for a given column.
fn iter<'a>(&'a self, col: Option<u32>) -> Box<Iterator<Item=(Box<[u8]>, Box<[u8]>)> + 'a>;
2017-04-18 15:45:15 +02:00
/// Iterate over flushed data for a given column, starting from a given prefix.
fn iter_from_prefix<'a>(&'a self, col: Option<u32>, prefix: &'a [u8])
-> Box<Iterator<Item=(Box<[u8]>, Box<[u8]>)> + 'a>;
/// Attempt to replace this database with a new one located at the given path.
fn restore(&self, new_db: &str) -> Result<(), UtilError>;
}
/// A key-value database fulfilling the `KeyValueDB` trait, living in memory.
/// This is generally intended for tests and is not particularly optimized.
pub struct InMemory {
columns: RwLock<HashMap<Option<u32>, BTreeMap<Vec<u8>, DBValue>>>,
}
/// Create an in-memory database with the given number of columns.
/// Columns will be indexable by 0..`num_cols`
pub fn in_memory(num_cols: u32) -> InMemory {
let mut cols = HashMap::new();
cols.insert(None, BTreeMap::new());
for idx in 0..num_cols {
cols.insert(Some(idx), BTreeMap::new());
}
InMemory {
columns: RwLock::new(cols)
}
}
impl KeyValueDB for InMemory {
fn get(&self, col: Option<u32>, key: &[u8]) -> Result<Option<DBValue>, String> {
let columns = self.columns.read();
match columns.get(&col) {
None => Err(format!("No such column family: {:?}", col)),
Some(map) => Ok(map.get(key).cloned()),
}
}
fn get_by_prefix(&self, col: Option<u32>, prefix: &[u8]) -> Option<Box<[u8]>> {
let columns = self.columns.read();
match columns.get(&col) {
None => None,
Some(map) =>
map.iter()
.find(|&(ref k ,_)| k.starts_with(prefix))
.map(|(_, v)| (&**v).to_vec().into_boxed_slice())
}
}
fn write_buffered(&self, transaction: DBTransaction) {
let mut columns = self.columns.write();
let ops = transaction.ops;
for op in ops {
match op {
DBOp::Insert { col, key, value } => {
if let Some(mut col) = columns.get_mut(&col) {
col.insert(key.to_vec(), value);
}
},
DBOp::InsertCompressed { col, key, value } => {
if let Some(mut col) = columns.get_mut(&col) {
let compressed = UntrustedRlp::new(&value).compress(RlpType::Blocks);
let mut value = DBValue::new();
value.append_slice(&compressed);
col.insert(key.to_vec(), value);
}
},
DBOp::Delete { col, key } => {
if let Some(mut col) = columns.get_mut(&col) {
col.remove(&*key);
}
},
}
}
}
fn flush(&self) -> Result<(), String> { Ok(()) }
fn iter<'a>(&'a self, col: Option<u32>) -> Box<Iterator<Item=(Box<[u8]>, Box<[u8]>)> + 'a> {
match self.columns.read().get(&col) {
Some(map) => Box::new( // TODO: worth optimizing at all?
map.clone()
.into_iter()
.map(|(k, v)| (k.into_boxed_slice(), v.to_vec().into_boxed_slice()))
),
2017-04-18 15:45:15 +02:00
None => Box::new(None.into_iter()),
}
}
fn iter_from_prefix<'a>(&'a self, col: Option<u32>, prefix: &'a [u8])
-> Box<Iterator<Item=(Box<[u8]>, Box<[u8]>)> + 'a>
{
match self.columns.read().get(&col) {
Some(map) => Box::new(
map.clone()
.into_iter()
.skip_while(move |&(ref k, _)| !k.starts_with(prefix))
.map(|(k, v)| (k.into_boxed_slice(), v.to_vec().into_boxed_slice()))
),
None => Box::new(None.into_iter()),
}
}
fn restore(&self, _new_db: &str) -> Result<(), UtilError> {
Err(UtilError::SimpleString("Attempted to restore in-memory database".into()))
}
}
2016-06-27 13:03:34 +02:00
/// Compaction profile for the database settings
#[derive(Clone, Copy, PartialEq, Debug)]
2016-06-27 13:03:34 +02:00
pub struct CompactionProfile {
/// L0-L1 target file size
pub initial_file_size: u64,
/// L2-LN target file size multiplier
pub file_size_multiplier: i32,
/// rate limiter for background flushes and compactions, bytes/sec, if any
pub write_rate_limit: Option<u64>,
}
impl Default for CompactionProfile {
2016-06-27 13:03:34 +02:00
/// Default profile suitable for most storage
fn default() -> CompactionProfile {
CompactionProfile::ssd()
}
}
/// Given output of df command return Linux rotational flag file path.
#[cfg(target_os = "linux")]
pub fn rotational_from_df_output(df_out: Vec<u8>) -> Option<PathBuf> {
str::from_utf8(df_out.as_slice())
.ok()
// Get the drive name.
.and_then(|df_str| Regex::new(r"/dev/(sd[:alpha:]{1,2})")
.ok()
.and_then(|re| re.captures(df_str))
.and_then(|captures| captures.get(1)))
// Generate path e.g. /sys/block/sda/queue/rotational
.map(|drive_path| {
let mut p = PathBuf::from("/sys/block");
p.push(drive_path.as_str());
p.push("queue/rotational");
p
})
}
impl CompactionProfile {
/// Attempt to determine the best profile automatically, only Linux for now.
#[cfg(target_os = "linux")]
pub fn auto(db_path: &Path) -> CompactionProfile {
let hdd_check_file = db_path
.to_str()
.and_then(|path_str| Command::new("df").arg(path_str).output().ok())
.and_then(|df_res| match df_res.status.success() {
true => Some(df_res.stdout),
false => None,
})
.and_then(rotational_from_df_output);
// Read out the file and match compaction profile.
if let Some(hdd_check) = hdd_check_file {
if let Ok(mut file) = File::open(hdd_check.as_path()) {
let mut buffer = [0; 1];
if file.read_exact(&mut buffer).is_ok() {
// 0 means not rotational.
if buffer == [48] { return Self::ssd(); }
// 1 means rotational.
if buffer == [49] { return Self::hdd(); }
}
}
}
// Fallback if drive type was not determined.
Self::default()
}
/// Just default for other platforms.
#[cfg(not(target_os = "linux"))]
pub fn auto(_db_path: &Path) -> CompactionProfile {
Self::default()
}
/// Default profile suitable for SSD storage
pub fn ssd() -> CompactionProfile {
2016-06-27 13:03:34 +02:00
CompactionProfile {
initial_file_size: 32 * 1024 * 1024,
file_size_multiplier: 2,
write_rate_limit: None,
}
}
/// Slow HDD compaction profile
2016-06-27 13:58:12 +02:00
pub fn hdd() -> CompactionProfile {
2016-06-27 13:03:34 +02:00
CompactionProfile {
initial_file_size: 192 * 1024 * 1024,
file_size_multiplier: 1,
2016-06-27 13:14:40 +02:00
write_rate_limit: Some(8 * 1024 * 1024),
2016-06-27 13:03:34 +02:00
}
}
}
2016-02-18 03:46:34 +01:00
/// Database configuration
#[derive(Clone)]
2016-02-18 03:46:34 +01:00
pub struct DatabaseConfig {
/// Max number of open files.
pub max_open_files: i32,
/// Cache sizes (in MiB) for specific columns.
pub cache_sizes: HashMap<Option<u32>, usize>,
2016-06-27 13:03:34 +02:00
/// Compaction profile
pub compaction: CompactionProfile,
/// Set number of columns
pub columns: Option<u32>,
/// Should we keep WAL enabled?
pub wal: bool,
2016-06-20 12:42:04 +02:00
}
impl DatabaseConfig {
/// Create new `DatabaseConfig` with default parameters and specified set of columns.
/// Note that cache sizes must be explicitly set.
pub fn with_columns(columns: Option<u32>) -> Self {
let mut config = Self::default();
config.columns = columns;
config
2016-06-27 13:03:34 +02:00
}
/// Set the column cache size in MiB.
pub fn set_cache(&mut self, col: Option<u32>, size: usize) {
self.cache_sizes.insert(col, size);
}
2016-06-20 12:42:04 +02:00
}
impl Default for DatabaseConfig {
fn default() -> DatabaseConfig {
DatabaseConfig {
cache_sizes: HashMap::new(),
2016-08-08 17:18:29 +02:00
max_open_files: 512,
2016-06-27 13:14:40 +02:00
compaction: CompactionProfile::default(),
columns: None,
wal: true,
2016-06-20 12:42:04 +02:00
}
}
2016-02-18 03:46:34 +01:00
}
/// Database iterator (for flushed data only)
// The compromise of holding only a virtual borrow vs. holding a lock on the
// inner DB (to prevent closing via restoration) may be re-evaluated in the future.
//
pub struct DatabaseIterator<'a> {
iter: DBIterator,
_marker: PhantomData<&'a Database>,
2016-02-18 03:46:34 +01:00
}
impl<'a> Iterator for DatabaseIterator<'a> {
2016-02-21 16:58:56 +01:00
type Item = (Box<[u8]>, Box<[u8]>);
2016-02-18 03:46:34 +01:00
fn next(&mut self) -> Option<Self::Item> {
2016-02-18 03:46:34 +01:00
self.iter.next()
}
}
struct DBAndColumns {
db: DB,
cfs: Vec<Column>,
}
// get column family configuration from database config.
fn col_config(col: u32, config: &DatabaseConfig) -> Options {
// default cache size for columns not specified.
const DEFAULT_CACHE: usize = 2;
let mut opts = Options::new();
opts.set_compaction_style(DBCompactionStyle::DBUniversalCompaction);
opts.set_target_file_size_base(config.compaction.initial_file_size);
opts.set_target_file_size_multiplier(config.compaction.file_size_multiplier);
2017-05-02 11:40:03 +02:00
opts.set_db_write_buffer_size(DB_WRITE_BUFFER_SIZE);
let col_opt = config.columns.map(|_| col);
{
let cache_size = config.cache_sizes.get(&col_opt).cloned().unwrap_or(DEFAULT_CACHE);
let mut block_opts = BlockBasedOptions::new();
// all goes to read cache.
block_opts.set_cache(Cache::new(cache_size * 1024 * 1024));
opts.set_block_based_table_factory(&block_opts);
}
opts
}
2016-02-18 03:46:34 +01:00
/// Key-Value database.
pub struct Database {
db: RwLock<Option<DBAndColumns>>,
config: DatabaseConfig,
write_opts: WriteOptions,
2016-10-14 14:44:11 +02:00
read_opts: ReadOptions,
path: String,
// Dirty values added with `write_buffered`. Cleaned on `flush`.
overlay: RwLock<Vec<HashMap<ElasticArray32<u8>, KeyState>>>,
// Values currently being flushed. Cleared when `flush` completes.
flushing: RwLock<Vec<HashMap<ElasticArray32<u8>, KeyState>>>,
// Prevents concurrent flushes.
// Value indicates if a flush is in progress.
flushing_lock: Mutex<bool>,
2016-02-18 03:46:34 +01:00
}
impl Database {
/// Open database with default settings.
pub fn open_default(path: &str) -> Result<Database, String> {
2016-06-20 12:42:04 +02:00
Database::open(&DatabaseConfig::default(), path)
2016-02-18 03:46:34 +01:00
}
/// Open database file. Creates if it does not exist.
2016-02-18 21:15:56 +01:00
pub fn open(config: &DatabaseConfig, path: &str) -> Result<Database, String> {
2016-02-18 03:46:34 +01:00
let mut opts = Options::new();
2016-06-27 13:14:40 +02:00
if let Some(rate_limit) = config.compaction.write_rate_limit {
opts.set_parsed_options(&format!("rate_limiter_bytes_per_sec={}", rate_limit))?;
2016-06-27 13:14:40 +02:00
}
opts.set_parsed_options(&format!("max_total_wal_size={}", 64 * 1024 * 1024))?;
opts.set_parsed_options("verify_checksums_in_compaction=0")?;
opts.set_max_open_files(config.max_open_files);
2016-02-18 03:46:34 +01:00
opts.create_if_missing(true);
opts.set_use_fsync(false);
2017-05-02 11:40:03 +02:00
opts.set_db_write_buffer_size(DB_WRITE_BUFFER_SIZE);
2016-06-27 13:14:40 +02:00
2016-08-18 18:03:40 +02:00
opts.set_max_background_flushes(DB_BACKGROUND_FLUSHES);
opts.set_max_background_compactions(DB_BACKGROUND_COMPACTIONS);
2016-06-27 13:14:40 +02:00
// compaction settings
2016-02-18 03:46:34 +01:00
opts.set_compaction_style(DBCompactionStyle::DBUniversalCompaction);
2016-06-27 13:14:40 +02:00
opts.set_target_file_size_base(config.compaction.initial_file_size);
opts.set_target_file_size_multiplier(config.compaction.file_size_multiplier);
2016-08-18 18:03:40 +02:00
let mut cf_options = Vec::with_capacity(config.columns.unwrap_or(0) as usize);
let cfnames: Vec<_> = (0..config.columns.unwrap_or(0)).map(|c| format!("col{}", c)).collect();
let cfnames: Vec<&str> = cfnames.iter().map(|n| n as &str).collect();
2016-08-18 18:03:40 +02:00
for col in 0 .. config.columns.unwrap_or(0) {
cf_options.push(col_config(col, &config));
2016-02-18 03:46:34 +01:00
}
let mut write_opts = WriteOptions::new();
if !config.wal {
write_opts.disable_wal(true);
}
2016-10-14 14:44:11 +02:00
let mut read_opts = ReadOptions::new();
read_opts.set_verify_checksums(false);
let mut cfs: Vec<Column> = Vec::new();
let db = match config.columns {
Some(columns) => {
2016-08-18 18:03:40 +02:00
match DB::open_cf(&opts, path, &cfnames, &cf_options) {
Ok(db) => {
cfs = cfnames.iter().map(|n| db.cf_handle(n)
.expect("rocksdb opens a cf_handle for each cfname; qed")).collect();
assert!(cfs.len() == columns as usize);
Ok(db)
}
Err(_) => {
// retry and create CFs
2016-08-18 18:03:40 +02:00
match DB::open_cf(&opts, path, &[], &[]) {
Ok(mut db) => {
cfs = cfnames.iter().enumerate().map(|(i, n)| db.create_cf(n, &cf_options[i])).collect::<Result<_, _>>()?;
Ok(db)
},
err @ Err(_) => err,
}
}
}
},
None => DB::open(&opts, path)
};
let db = match db {
2016-06-27 18:46:50 +02:00
Ok(db) => db,
Err(ref s) if s.starts_with("Corruption:") => {
info!("{}", s);
info!("Attempting DB repair for {}", path);
DB::repair(&opts, path)?;
match cfnames.is_empty() {
true => DB::open(&opts, path)?,
false => DB::open_cf(&opts, path, &cfnames, &cf_options)?
}
2016-06-27 18:46:50 +02:00
},
Err(s) => { return Err(s); }
};
let num_cols = cfs.len();
Ok(Database {
db: RwLock::new(Some(DBAndColumns{ db: db, cfs: cfs })),
config: config.clone(),
write_opts: write_opts,
overlay: RwLock::new((0..(num_cols + 1)).map(|_| HashMap::new()).collect()),
flushing: RwLock::new((0..(num_cols + 1)).map(|_| HashMap::new()).collect()),
flushing_lock: Mutex::new((false)),
path: path.to_owned(),
2016-10-14 14:44:11 +02:00
read_opts: read_opts,
})
2016-02-18 03:46:34 +01:00
}
/// Helper to create new transaction for this database.
pub fn transaction(&self) -> DBTransaction {
DBTransaction::new()
2016-02-18 03:46:34 +01:00
}
2016-08-04 23:54:26 +02:00
fn to_overlay_column(col: Option<u32>) -> usize {
col.map_or(0, |c| (c + 1) as usize)
}
/// Commit transaction to database.
pub fn write_buffered(&self, tr: DBTransaction) {
let mut overlay = self.overlay.write();
let ops = tr.ops;
for op in ops {
match op {
DBOp::Insert { col, key, value } => {
2016-08-04 23:54:26 +02:00
let c = Self::to_overlay_column(col);
overlay[c].insert(key, KeyState::Insert(value));
},
DBOp::InsertCompressed { col, key, value } => {
2016-08-04 23:54:26 +02:00
let c = Self::to_overlay_column(col);
overlay[c].insert(key, KeyState::InsertCompressed(value));
},
DBOp::Delete { col, key } => {
2016-08-04 23:54:26 +02:00
let c = Self::to_overlay_column(col);
overlay[c].insert(key, KeyState::Delete);
},
}
};
}
/// Commit buffered changes to database. Must be called under `flush_lock`
fn write_flushing_with_lock(&self, _lock: &mut MutexGuard<bool>) -> Result<(), String> {
match *self.db.read() {
Some(DBAndColumns { ref db, ref cfs }) => {
let batch = WriteBatch::new();
mem::swap(&mut *self.overlay.write(), &mut *self.flushing.write());
{
for (c, column) in self.flushing.read().iter().enumerate() {
for (ref key, ref state) in column.iter() {
match **state {
KeyState::Delete => {
if c > 0 {
batch.delete_cf(cfs[c - 1], &key)?;
} else {
batch.delete(&key)?;
}
},
KeyState::Insert(ref value) => {
if c > 0 {
batch.put_cf(cfs[c - 1], &key, value)?;
} else {
batch.put(&key, &value)?;
}
},
KeyState::InsertCompressed(ref value) => {
let compressed = UntrustedRlp::new(&value).compress(RlpType::Blocks);
if c > 0 {
batch.put_cf(cfs[c - 1], &key, &compressed)?;
} else {
batch.put(&key, &value)?;
}
}
}
}
}
}
db.write_opt(batch, &self.write_opts)?;
for column in self.flushing.write().iter_mut() {
column.clear();
2016-11-12 12:07:02 +01:00
column.shrink_to_fit();
}
Ok(())
},
None => Err("Database is closed".to_owned())
}
}
/// Commit buffered changes to database.
pub fn flush(&self) -> Result<(), String> {
let mut lock = self.flushing_lock.lock();
// If RocksDB batch allocation fails the thread gets terminated and the lock is released.
// The value inside the lock is used to detect that.
if *lock {
// This can only happen if another flushing thread is terminated unexpectedly.
return Err("Database write failure. Running low on memory perhaps?".to_owned());
}
*lock = true;
let result = self.write_flushing_with_lock(&mut lock);
*lock = false;
result
}
2016-02-18 03:46:34 +01:00
/// Commit transaction to database.
2016-02-21 16:58:56 +01:00
pub fn write(&self, tr: DBTransaction) -> Result<(), String> {
match *self.db.read() {
Some(DBAndColumns { ref db, ref cfs }) => {
let batch = WriteBatch::new();
let ops = tr.ops;
for op in ops {
match op {
DBOp::Insert { col, key, value } => {
col.map_or_else(|| batch.put(&key, &value), |c| batch.put_cf(cfs[c as usize], &key, &value))?
},
DBOp::InsertCompressed { col, key, value } => {
let compressed = UntrustedRlp::new(&value).compress(RlpType::Blocks);
col.map_or_else(|| batch.put(&key, &compressed), |c| batch.put_cf(cfs[c as usize], &key, &compressed))?
},
DBOp::Delete { col, key } => {
col.map_or_else(|| batch.delete(&key), |c| batch.delete_cf(cfs[c as usize], &key))?
},
}
}
db.write_opt(batch, &self.write_opts)
},
None => Err("Database is closed".to_owned())
}
2016-02-18 03:46:34 +01:00
}
2016-02-21 13:19:08 +01:00
2016-02-18 03:46:34 +01:00
/// Get value by key.
2016-10-26 13:53:47 +02:00
pub fn get(&self, col: Option<u32>, key: &[u8]) -> Result<Option<DBValue>, String> {
match *self.db.read() {
Some(DBAndColumns { ref db, ref cfs }) => {
let overlay = &self.overlay.read()[Self::to_overlay_column(col)];
match overlay.get(key) {
Some(&KeyState::Insert(ref value)) | Some(&KeyState::InsertCompressed(ref value)) => Ok(Some(value.clone())),
Some(&KeyState::Delete) => Ok(None),
None => {
let flushing = &self.flushing.read()[Self::to_overlay_column(col)];
match flushing.get(key) {
Some(&KeyState::Insert(ref value)) | Some(&KeyState::InsertCompressed(ref value)) => Ok(Some(value.clone())),
Some(&KeyState::Delete) => Ok(None),
None => {
col.map_or_else(
2016-10-26 13:53:47 +02:00
|| db.get_opt(key, &self.read_opts).map(|r| r.map(|v| DBValue::from_slice(&v))),
|c| db.get_cf_opt(cfs[c as usize], key, &self.read_opts).map(|r| r.map(|v| DBValue::from_slice(&v))))
},
}
},
}
},
None => Ok(None),
}
2016-02-18 03:46:34 +01:00
}
/// Get value by partial key. Prefix size should match configured prefix size. Only searches flushed values.
// TODO: support prefix seek for unflushed data
pub fn get_by_prefix(&self, col: Option<u32>, prefix: &[u8]) -> Option<Box<[u8]>> {
2017-04-18 15:45:15 +02:00
self.iter_from_prefix(col, prefix).and_then(|mut iter| {
match iter.next() {
// TODO: use prefix_same_as_start read option (not availabele in C API currently)
Some((k, v)) => if k[0 .. prefix.len()] == prefix[..] { Some(v) } else { None },
_ => None
}
})
2016-02-18 03:46:34 +01:00
}
/// Get database iterator for flushed data.
2017-04-18 15:45:15 +02:00
pub fn iter(&self, col: Option<u32>) -> Option<DatabaseIterator> {
//TODO: iterate over overlay
match *self.db.read() {
Some(DBAndColumns { ref db, ref cfs }) => {
let iter = col.map_or_else(
|| db.iterator_opt(IteratorMode::Start, &self.read_opts),
|c| db.iterator_cf_opt(cfs[c as usize], IteratorMode::Start, &self.read_opts)
.expect("iterator params are valid; qed")
);
2017-04-18 15:45:15 +02:00
Some(DatabaseIterator {
iter: iter,
_marker: PhantomData,
2017-04-18 15:45:15 +02:00
})
},
None => None,
}
}
fn iter_from_prefix(&self, col: Option<u32>, prefix: &[u8]) -> Option<DatabaseIterator> {
match *self.db.read() {
Some(DBAndColumns { ref db, ref cfs }) => {
let iter = col.map_or_else(|| db.iterator_opt(IteratorMode::From(prefix, Direction::Forward), &self.read_opts),
|c| db.iterator_cf_opt(cfs[c as usize], IteratorMode::From(prefix, Direction::Forward), &self.read_opts)
.expect("iterator params are valid; qed"));
Some(DatabaseIterator {
iter: iter,
_marker: PhantomData,
})
},
2017-04-18 15:45:15 +02:00
None => None,
}
}
/// Close the database
fn close(&self) {
*self.db.write() = None;
self.overlay.write().clear();
self.flushing.write().clear();
}
/// Restore the database from a copy at given path.
pub fn restore(&self, new_db: &str) -> Result<(), UtilError> {
self.close();
let mut backup_db = PathBuf::from(&self.path);
backup_db.pop();
backup_db.push("backup_db");
let existed = match fs::rename(&self.path, &backup_db) {
Ok(_) => true,
Err(e) => if let ErrorKind::NotFound = e.kind() {
false
} else {
return Err(e.into());
}
};
match fs::rename(&new_db, &self.path) {
Ok(_) => {
// clean up the backup.
if existed {
fs::remove_dir_all(&backup_db)?;
}
}
Err(e) => {
// restore the backup.
if existed {
fs::rename(&backup_db, &self.path)?;
}
return Err(e.into())
}
}
// reopen the database and steal handles into self
let db = Self::open(&self.config, &self.path)?;
*self.db.write() = mem::replace(&mut *db.db.write(), None);
*self.overlay.write() = mem::replace(&mut *db.overlay.write(), Vec::new());
*self.flushing.write() = mem::replace(&mut *db.flushing.write(), Vec::new());
Ok(())
2016-02-18 03:46:34 +01:00
}
/// The number of non-default column families.
pub fn num_columns(&self) -> u32 {
self.db.read().as_ref()
.and_then(|db| if db.cfs.is_empty() { None } else { Some(db.cfs.len()) } )
.map(|n| n as u32)
.unwrap_or(0)
}
/// Drop a column family.
pub fn drop_column(&self) -> Result<(), String> {
match *self.db.write() {
Some(DBAndColumns { ref mut db, ref mut cfs }) => {
if let Some(col) = cfs.pop() {
let name = format!("col{}", cfs.len());
drop(col);
db.drop_cf(&name)?;
}
Ok(())
},
None => Ok(()),
}
}
/// Add a column family.
pub fn add_column(&self) -> Result<(), String> {
match *self.db.write() {
Some(DBAndColumns { ref mut db, ref mut cfs }) => {
let col = cfs.len() as u32;
let name = format!("col{}", col);
cfs.push(db.create_cf(&name, &col_config(col, &self.config))?);
Ok(())
},
None => Ok(()),
}
}
2016-02-18 03:46:34 +01:00
}
// duplicate declaration of methods here to avoid trait import in certain existing cases
// at time of addition.
impl KeyValueDB for Database {
fn get(&self, col: Option<u32>, key: &[u8]) -> Result<Option<DBValue>, String> {
Database::get(self, col, key)
}
fn get_by_prefix(&self, col: Option<u32>, prefix: &[u8]) -> Option<Box<[u8]>> {
Database::get_by_prefix(self, col, prefix)
}
fn write_buffered(&self, transaction: DBTransaction) {
Database::write_buffered(self, transaction)
}
fn write(&self, transaction: DBTransaction) -> Result<(), String> {
Database::write(self, transaction)
}
fn flush(&self) -> Result<(), String> {
Database::flush(self)
}
fn iter<'a>(&'a self, col: Option<u32>) -> Box<Iterator<Item=(Box<[u8]>, Box<[u8]>)> + 'a> {
let unboxed = Database::iter(self, col);
2017-04-18 15:45:15 +02:00
Box::new(unboxed.into_iter().flat_map(|inner| inner))
}
fn iter_from_prefix<'a>(&'a self, col: Option<u32>, prefix: &'a [u8])
-> Box<Iterator<Item=(Box<[u8]>, Box<[u8]>)> + 'a>
{
let unboxed = Database::iter_from_prefix(self, col, prefix);
Box::new(unboxed.into_iter().flat_map(|inner| inner))
}
fn restore(&self, new_db: &str) -> Result<(), UtilError> {
Database::restore(self, new_db)
}
}
2016-10-22 15:22:34 +02:00
impl Drop for Database {
fn drop(&mut self) {
// write all buffered changes if we can.
let _ = self.flush();
}
}
2016-02-18 21:15:56 +01:00
#[cfg(test)]
mod tests {
2016-11-28 17:05:37 +01:00
use hash::H256;
2016-02-18 21:15:56 +01:00
use super::*;
2016-02-21 13:19:08 +01:00
use devtools::*;
2016-02-18 21:15:56 +01:00
use std::str::FromStr;
fn test_db(config: &DatabaseConfig) {
let path = RandomTempPath::create_dir();
let db = Database::open(config, path.as_path().to_str().unwrap()).unwrap();
let key1 = H256::from_str("02c69be41d0b7e40352fc85be1cd65eb03d40ef8427a0ca4596b1ead9a00e9fc").unwrap();
let key2 = H256::from_str("03c69be41d0b7e40352fc85be1cd65eb03d40ef8427a0ca4596b1ead9a00e9fc").unwrap();
let key3 = H256::from_str("01c69be41d0b7e40352fc85be1cd65eb03d40ef8427a0ca4596b1ead9a00e9fc").unwrap();
let mut batch = db.transaction();
batch.put(None, &key1, b"cat");
batch.put(None, &key2, b"dog");
db.write(batch).unwrap();
2016-02-18 21:15:56 +01:00
assert_eq!(&*db.get(None, &key1).unwrap().unwrap(), b"cat");
2016-02-18 21:15:56 +01:00
2017-04-20 15:04:07 +02:00
let contents: Vec<_> = db.iter(None).into_iter().flat_map(|inner| inner).collect();
2016-02-18 21:15:56 +01:00
assert_eq!(contents.len(), 2);
assert_eq!(&*contents[0].0, &*key1);
2016-02-18 21:15:56 +01:00
assert_eq!(&*contents[0].1, b"cat");
assert_eq!(&*contents[1].0, &*key2);
2016-02-18 21:15:56 +01:00
assert_eq!(&*contents[1].1, b"dog");
let mut batch = db.transaction();
batch.delete(None, &key1);
db.write(batch).unwrap();
assert!(db.get(None, &key1).unwrap().is_none());
let mut batch = db.transaction();
batch.put(None, &key1, b"cat");
db.write(batch).unwrap();
2016-02-18 21:15:56 +01:00
let mut transaction = db.transaction();
transaction.put(None, &key3, b"elephant");
transaction.delete(None, &key1);
2016-02-18 21:15:56 +01:00
db.write(transaction).unwrap();
assert!(db.get(None, &key1).unwrap().is_none());
assert_eq!(&*db.get(None, &key3).unwrap().unwrap(), b"elephant");
2016-02-21 13:19:08 +01:00
assert_eq!(&*db.get_by_prefix(None, &key3).unwrap(), b"elephant");
assert_eq!(&*db.get_by_prefix(None, &key2).unwrap(), b"dog");
let mut transaction = db.transaction();
transaction.put(None, &key1, b"horse");
transaction.delete(None, &key3);
db.write_buffered(transaction);
assert!(db.get(None, &key3).unwrap().is_none());
assert_eq!(&*db.get(None, &key1).unwrap().unwrap(), b"horse");
db.flush().unwrap();
assert!(db.get(None, &key3).unwrap().is_none());
assert_eq!(&*db.get(None, &key1).unwrap().unwrap(), b"horse");
2016-02-18 21:15:56 +01:00
}
#[test]
fn kvdb() {
let path = RandomTempPath::create_dir();
let _ = Database::open_default(path.as_path().to_str().unwrap()).unwrap();
2016-06-27 18:47:50 +02:00
test_db(&DatabaseConfig::default());
2016-02-18 21:15:56 +01:00
}
#[test]
#[cfg(target_os = "linux")]
fn df_to_rotational() {
use std::path::PathBuf;
// Example df output.
let example_df = vec![70, 105, 108, 101, 115, 121, 115, 116, 101, 109, 32, 32, 32, 32, 32, 49, 75, 45, 98, 108, 111, 99, 107, 115, 32, 32, 32, 32, 32, 85, 115, 101, 100, 32, 65, 118, 97, 105, 108, 97, 98, 108, 101, 32, 85, 115, 101, 37, 32, 77, 111, 117, 110, 116, 101, 100, 32, 111, 110, 10, 47, 100, 101, 118, 47, 115, 100, 97, 49, 32, 32, 32, 32, 32, 32, 32, 54, 49, 52, 48, 57, 51, 48, 48, 32, 51, 56, 56, 50, 50, 50, 51, 54, 32, 32, 49, 57, 52, 52, 52, 54, 49, 54, 32, 32, 54, 55, 37, 32, 47, 10];
let expected_output = Some(PathBuf::from("/sys/block/sda/queue/rotational"));
assert_eq!(rotational_from_df_output(example_df), expected_output);
}
#[test]
fn add_columns() {
let config = DatabaseConfig::default();
let config_5 = DatabaseConfig::with_columns(Some(5));
let path = RandomTempPath::create_dir();
// open empty, add 5.
{
let db = Database::open(&config, path.as_path().to_str().unwrap()).unwrap();
assert_eq!(db.num_columns(), 0);
for i in 0..5 {
db.add_column().unwrap();
assert_eq!(db.num_columns(), i + 1);
}
}
// reopen as 5.
{
let db = Database::open(&config_5, path.as_path().to_str().unwrap()).unwrap();
assert_eq!(db.num_columns(), 5);
}
}
#[test]
fn drop_columns() {
let config = DatabaseConfig::default();
let config_5 = DatabaseConfig::with_columns(Some(5));
let path = RandomTempPath::create_dir();
// open 5, remove all.
{
let db = Database::open(&config_5, path.as_path().to_str().unwrap()).unwrap();
assert_eq!(db.num_columns(), 5);
for i in (0..5).rev() {
db.drop_column().unwrap();
assert_eq!(db.num_columns(), i);
}
}
// reopen as 0.
{
let db = Database::open(&config, path.as_path().to_str().unwrap()).unwrap();
assert_eq!(db.num_columns(), 0);
}
}
2016-02-18 21:15:56 +01:00
}