Merge pull request 'TypedTree' (#1456) from krtab/garage:typed_tree into main-v2

Reviewed-on: https://git.deuxfleurs.fr/Deuxfleurs/garage/pulls/1456
Reviewed-by: Armael <armael@noreply.localhost>
Reviewed-by: Alex <lx@deuxfleurs.fr>
This commit is contained in:
Alex
2026-06-30 21:04:06 +00:00
17 changed files with 760 additions and 363 deletions
+1 -8
View File
@@ -360,14 +360,7 @@ impl Worker for BlockRcRepair {
_must_exit: &mut watch::Receiver<bool>,
) -> Result<WorkerState, GarageError> {
for _i in 0..RC_REPAIR_ITER_COUNT {
let next1 = self
.block_manager
.rc
.rc_table
.range(self.cursor.as_slice()..)?
.next()
.transpose()?
.map(|(k, _)| Hash::try_from(k.as_slice()).unwrap());
let next1 = self.block_manager.rc.get_first_hash_from(self.cursor)?;
let next2 = self
.block_ref_table
.data
+5 -6
View File
@@ -144,7 +144,7 @@ impl BlockManager {
// Open metadata tables
let rc = db
.open_tree("block_local_rc")
.open_typed_tree("block_local_rc")
.expect("Unable to open block_local_rc tree");
let rc = BlockRc::new(rc);
@@ -158,9 +158,9 @@ impl BlockManager {
let metrics = BlockManagerMetrics::new(
config.compression_level,
rc.rc_table.clone(),
resync.queue.clone(),
resync.errors.clone(),
rc.rc_table.untyped().clone(),
resync.queue.untyped().clone(),
resync.errors.untyped().clone(),
buffer_kb_semaphore.clone(),
);
@@ -449,9 +449,8 @@ impl BlockManager {
let mut blocks = Vec::with_capacity(self.resync.errors.approximate_len()?);
for ent in self.resync.errors.iter()? {
let (hash, cnt) = ent?;
let cnt = ErrorCounter::decode(&cnt);
blocks.push(BlockResyncErrorInfo {
hash: Hash::try_from(&hash).unwrap(),
hash,
refcount: 0,
error_count: cnt.errors,
last_try: cnt.last_try,
+103 -98
View File
@@ -1,4 +1,5 @@
use std::convert::TryInto;
use std::num::NonZeroU64;
use arc_swap::ArcSwapOption;
@@ -14,12 +15,12 @@ pub type CalculateRefcount =
Box<dyn Fn(&db::Transaction, &Hash) -> db::TxResult<usize, Error> + Send + Sync>;
pub struct BlockRc {
pub rc_table: db::Tree,
pub(crate) rc_table: db::TypedTree<Hash, RcEntry>,
pub(crate) recalc_rc: ArcSwapOption<Vec<CalculateRefcount>>,
}
impl BlockRc {
pub(crate) fn new(rc: db::Tree) -> Self {
pub(crate) fn new(rc: db::TypedTree<Hash, RcEntry>) -> Self {
Self {
rc_table: rc,
recalc_rc: ArcSwapOption::new(None),
@@ -33,11 +34,8 @@ impl BlockRc {
tx: &mut db::Transaction,
hash: &Hash,
) -> db::TxOpResult<bool> {
let old_rc = RcEntry::parse_opt(tx.get(&self.rc_table, hash)?);
match old_rc.increment().serialize() {
Some(x) => tx.insert(&self.rc_table, hash, x)?,
None => unreachable!(),
}
let old_rc = RcState(self.rc_table.tx_get(tx, hash)?);
self.rc_table.tx_insert(tx, hash, &old_rc.increment())?;
Ok(old_rc.is_zero())
}
@@ -48,17 +46,27 @@ impl BlockRc {
tx: &mut db::Transaction,
hash: &Hash,
) -> db::TxOpResult<bool> {
let new_rc = RcEntry::parse_opt(tx.get(&self.rc_table, hash)?).decrement();
match new_rc.serialize() {
Some(x) => tx.insert(&self.rc_table, hash, x)?,
None => tx.remove(&self.rc_table, hash)?,
let new_rc = RcState(self.rc_table.tx_get(tx, hash)?).decrement();
match &new_rc.0 {
None => self.rc_table.tx_remove(tx, hash)?,
Some(rc) => self.rc_table.tx_insert(tx, hash, rc)?,
}
Ok(matches!(new_rc, RcEntry::Deletable { .. }))
Ok(matches!(new_rc.0, Some(RcEntry::Deletable { .. })))
}
/// Read a block's reference count
pub(crate) fn get_block_rc(&self, hash: &Hash) -> Result<RcEntry, Error> {
Ok(RcEntry::parse_opt(self.rc_table.get(hash.as_ref())?))
/// Read a block's reference counting state
pub(crate) fn get_block_rc(&self, hash: &Hash) -> Result<RcState, Error> {
Ok(RcState(self.rc_table.get(hash)?))
}
/// Return the first hash stored in the RC table at or after `cursor`
pub fn get_first_hash_from(&self, cursor: Hash) -> Result<Option<Hash>, Error> {
Ok(self
.rc_table
.range(cursor..)?
.next()
.transpose()?
.map(|(k, _)| k))
}
/// Delete an entry in the RC table if it is deletable and the
@@ -66,12 +74,11 @@ impl BlockRc {
pub(crate) fn clear_deleted_block_rc(&self, hash: &Hash) -> Result<(), Error> {
let now = now_msec();
self.rc_table.db().transaction(|tx| {
let rcval = RcEntry::parse_opt(tx.get(&self.rc_table, hash)?);
match rcval {
RcEntry::Deletable { at_time } if now > at_time => {
tx.remove(&self.rc_table, hash)?;
let rcval = self.rc_table.tx_get(tx, hash)?;
if let Some(RcEntry::Deletable { at_time }) = rcval {
if now > at_time {
self.rc_table.tx_remove(tx, hash)?;
}
_ => (),
}
Ok(())
})?;
@@ -91,28 +98,25 @@ impl BlockRc {
for f in recalc_fns.iter() {
cnt += f(tx, hash)?;
}
let old_rc = RcEntry::parse_opt(tx.get(&self.rc_table, hash)?);
let old_count = RcState(self.rc_table.tx_get(tx, hash)?).as_u64();
trace!(
"Block RC for {:?}: stored={}, calculated={}",
hash,
old_rc.as_u64(),
old_count,
cnt
);
if cnt as u64 != old_rc.as_u64() {
if cnt as u64 != old_count {
warn!(
"Fixing inconsistent block RC for {:?}: was {}, should be {}",
hash,
old_rc.as_u64(),
cnt
hash, old_count, cnt
);
let new_rc = if cnt > 0 {
RcEntry::Present { count: cnt as u64 }
} else {
RcEntry::Deletable {
let new_rc = match NonZeroU64::new(cnt as u64) {
Some(count) => RcEntry::Present { count },
None => RcEntry::Deletable {
at_time: now_msec() + BLOCK_GC_DELAY.as_millis() as u64,
}
},
};
tx.insert(&self.rc_table, hash, new_rc.serialize().unwrap())?;
self.rc_table.tx_insert(tx, hash, &new_rc)?;
Ok((cnt, true))
} else {
Ok((cnt, false))
@@ -131,13 +135,47 @@ impl BlockRc {
}
}
/// Describes the state of the reference counter for a block
impl db::DbBytes for RcEntry {
fn encode(&self) -> Vec<u8> {
match self {
RcEntry::Present { count } => u64::to_be_bytes(count.get()).to_vec(),
RcEntry::Deletable { at_time } => {
[u64::to_be_bytes(0), u64::to_be_bytes(*at_time)].concat()
}
}
}
fn decode(bytes: &[u8]) -> std::result::Result<Self, db::DecodeError> {
if bytes.len() == 8 {
let count = NonZeroU64::new(u64::from_be_bytes(bytes.try_into().unwrap()))
.ok_or(db::DecodeError("invalid RC entry: zero count".into()))?;
Ok(RcEntry::Present { count })
} else if bytes.len() == 16 {
Ok(RcEntry::Deletable {
at_time: u64::from_be_bytes(bytes[8..16].try_into().unwrap()),
})
} else {
Err(db::DecodeError(
format!(
"invalid RC entry: expected 8 or 16 bytes, got {}",
bytes.len()
)
.into(),
))
}
}
}
/// A block's entry in the RC table.
///
/// A block with zero references and no pending deletion has no entry
/// in the RC table at all: see [`RcState`].
#[derive(Clone, Copy, Debug)]
pub(crate) enum RcEntry {
/// Present: the block has `count` references, with `count` > 0.
/// Present: the block has `count` references.
///
/// This is stored as `u64::to_be_bytes(count)`
Present { count: u64 },
Present { count: NonZeroU64 },
/// Deletable: the block has zero references, and can be deleted
/// once time (returned by `now_msec`) is larger than `at_time`
@@ -147,72 +185,39 @@ pub(crate) enum RcEntry {
/// (this allows for the data format to be backwards compatible with
/// previous Garage versions that didn't have this intermediate state)
Deletable { at_time: u64 },
/// Absent: the block has zero references, and can be deleted
/// immediately
Absent,
}
impl RcEntry {
fn parse(bytes: &[u8]) -> Self {
if bytes.len() == 8 {
RcEntry::Present {
count: u64::from_be_bytes(bytes.try_into().unwrap()),
}
} else if bytes.len() == 16 {
RcEntry::Deletable {
at_time: u64::from_be_bytes(bytes[8..16].try_into().unwrap()),
}
} else {
panic!("Invalid RC entry: {:?}, database is corrupted. This is an error Garage is currently unable to recover from. Sorry, and also please report a bug.",
bytes
)
}
}
/// Describes the state of the reference counter for a block: the block's
/// entry in the RC table, or `None` if it has none, meaning the block has
/// zero references and can be deleted immediately.
#[derive(Clone, Copy, Debug)]
pub(crate) struct RcState(Option<RcEntry>);
fn parse_opt<V: AsRef<[u8]>>(bytes: Option<V>) -> Self {
bytes
.map(|b| Self::parse(b.as_ref()))
.unwrap_or(Self::Absent)
}
fn serialize(self) -> Option<Vec<u8>> {
match self {
RcEntry::Present { count } => Some(u64::to_be_bytes(count).to_vec()),
RcEntry::Deletable { at_time } => {
Some([u64::to_be_bytes(0), u64::to_be_bytes(at_time)].concat())
}
RcEntry::Absent => None,
}
}
fn increment(self) -> Self {
let old_count = match self {
RcEntry::Present { count } => count,
_ => 0,
impl RcState {
/// The new RC table entry after a reference is taken on the block
fn increment(&self) -> RcEntry {
let count = match self.0 {
Some(RcEntry::Present { count }) => count.saturating_add(1),
_ => NonZeroU64::new(1).unwrap(),
};
RcEntry::Present {
count: old_count + 1,
}
RcEntry::Present { count }
}
fn decrement(self) -> Self {
match self {
RcEntry::Present { count } => {
if count > 1 {
RcEntry::Present { count: count - 1 }
} else {
RcEntry::Deletable {
at_time: now_msec() + BLOCK_GC_DELAY.as_millis() as u64,
}
}
}
del => del,
}
/// The new state after a reference to the block is dropped
fn decrement(&self) -> Self {
RcState(match self.0 {
Some(RcEntry::Present { count }) => Some(match NonZeroU64::new(count.get() - 1) {
Some(count) => RcEntry::Present { count },
None => RcEntry::Deletable {
at_time: now_msec() + BLOCK_GC_DELAY.as_millis() as u64,
},
}),
unchanged => unchanged,
})
}
pub(crate) fn is_zero(&self) -> bool {
matches!(self, RcEntry::Deletable { .. } | RcEntry::Absent)
matches!(self.0, None | Some(RcEntry::Deletable { .. }))
}
pub(crate) fn is_nonzero(&self) -> bool {
@@ -220,10 +225,10 @@ impl RcEntry {
}
pub(crate) fn is_deletable(&self) -> bool {
match self {
RcEntry::Present { .. } => false,
RcEntry::Deletable { at_time } => now_msec() > *at_time,
RcEntry::Absent => true,
match self.0 {
Some(RcEntry::Present { .. }) => false,
Some(RcEntry::Deletable { at_time }) => now_msec() > at_time,
None => true,
}
}
@@ -232,8 +237,8 @@ impl RcEntry {
}
pub(crate) fn as_u64(&self) -> u64 {
match self {
RcEntry::Present { count } => *count,
match self.0 {
Some(RcEntry::Present { count }) => count.get(),
_ => 0,
}
}
+2 -3
View File
@@ -102,16 +102,15 @@ impl Worker for RepairWorker {
let mut batch_of_hashes = vec![];
let start_bound = match self.next_start.as_ref() {
None => Bound::Unbounded,
Some(x) => Bound::Excluded(x.as_slice()),
Some(x) => Bound::Excluded(*x),
};
for entry in self
.manager
.rc
.rc_table
.range::<&[u8], _>((start_bound, Bound::Unbounded))?
.range((start_bound, Bound::Unbounded))?
{
let (hash, _) = entry?;
let hash = Hash::try_from(&hash[..]).unwrap();
batch_of_hashes.push(hash);
if batch_of_hashes.len() >= 1000 {
break;
+106 -45
View File
@@ -44,15 +44,58 @@ pub(crate) const MAX_RESYNC_WORKERS: usize = 8;
const INITIAL_RESYNC_TRANQUILITY: u32 = 2;
pub struct BlockResyncManager {
pub(crate) queue: db::Tree,
pub(crate) queue: db::TypedTree<ResyncQueueKey, Hash>,
pub(crate) notify: Arc<Notify>,
pub(crate) errors: db::Tree,
pub(crate) errors: db::TypedTree<Hash, ErrorCounter>,
busy_set: BusySet,
persister: PersisterShared<ResyncPersistedConfig>,
}
/// Key of the resync queue tree: blocks are resynced in order of increasing
/// `when` (msec timestamp of the next try), with the block hash as tie-breaker.
///
// CAREFUL: this type implements `DbOrdKey`, so its byte encoding must be
// order-preserving.
// The derived `Ord` compares fields in declaration order, which must match
// the order in which `encode()` writes them; and `when` must remain an
// *unsigned* integer, as the big-endian encoding is only order-preserving
// for unsigned types.
#[derive(Clone, Copy, Debug, PartialEq, Eq, PartialOrd, Ord, Hash)]
pub(crate) struct ResyncQueueKey {
pub(crate) when: u64,
pub(crate) hash: Hash,
}
impl db::DbBytes for ResyncQueueKey {
fn encode(&self) -> Vec<u8> {
let mut v = Vec::with_capacity(40);
v.extend_from_slice(&u64::to_be_bytes(self.when));
v.extend_from_slice(self.hash.as_slice());
v
}
fn decode(bytes: &[u8]) -> std::result::Result<Self, db::DecodeError> {
if bytes.len() != 40 {
return Err(db::DecodeError(
format!(
"invalid resync queue key: expected 40 bytes, got {}",
bytes.len()
)
.into(),
));
}
Ok(ResyncQueueKey {
when: u64::from_be_bytes(bytes[..8].try_into().unwrap()),
hash: Hash::try_from(&bytes[8..])
.ok_or_else(|| db::DecodeError("invalid resync queue key: bad hash".into()))?,
})
}
}
impl db::DbOrdKey for ResyncQueueKey {}
#[derive(Serialize, Deserialize, Clone, Copy)]
struct ResyncPersistedConfig {
n_workers: usize,
@@ -74,22 +117,21 @@ enum ResyncIterResult {
IdleFor(Duration),
}
type BusySet = Arc<Mutex<HashSet<Vec<u8>>>>;
type BusySet = Arc<Mutex<HashSet<ResyncQueueKey>>>;
struct BusyBlock {
time_bytes: Vec<u8>,
hash_bytes: Vec<u8>,
key: ResyncQueueKey,
busy_set: BusySet,
}
impl BlockResyncManager {
pub(crate) fn new(db: &db::Db, system: &System) -> Self {
let queue = db
.open_tree("block_local_resync_queue")
.open_typed_tree("block_local_resync_queue")
.expect("Unable to open block_local_resync_queue tree");
let errors = db
.open_tree("block_local_resync_errors")
.open_typed_tree("block_local_resync_errors")
.expect("Unable to open block_local_resync_errors tree");
let persister = PersisterShared::new(&system.metadata_dir, "resync_cfg");
@@ -116,11 +158,10 @@ impl BlockResyncManager {
/// Clear the error counter for a block and put it in queue immediately
pub fn clear_backoff(&self, hash: &Hash) -> Result<(), Error> {
let now = now_msec();
if let Some(ec) = self.errors.get(hash)? {
let mut ec = ErrorCounter::decode(&ec);
if let Some(mut ec) = self.errors.get(hash)? {
if ec.errors > 0 {
ec.last_try = now - ec.delay_msec();
self.errors.insert(hash, ec.encode())?;
self.errors.insert(hash, &ec)?;
self.put_to_resync_at(hash, now)?;
return Ok(());
}
@@ -251,23 +292,21 @@ impl BlockResyncManager {
pub(crate) fn put_to_resync_at(&self, hash: &Hash, when: u64) -> db::Result<()> {
trace!("Put resync_queue: {} {:?}", when, hash);
let mut key = u64::to_be_bytes(when).to_vec();
key.extend(hash.as_ref());
self.queue.insert(key, hash.as_ref())?;
let qkey = ResyncQueueKey { when, hash: *hash };
self.queue.insert(&qkey, hash)?;
self.notify.notify_waiters();
Ok(())
}
async fn resync_iter(&self, manager: &BlockManager) -> Result<ResyncIterResult, db::Error> {
if let Some(block) = self.get_block_to_resync()? {
let time_msec = u64::from_be_bytes(block.time_bytes[0..8].try_into().unwrap());
let time_msec = block.key.when;
let now = now_msec();
if now >= time_msec {
let hash = Hash::try_from(&block.hash_bytes[..]).unwrap();
let hash = block.key.hash;
if let Some(ec) = self.errors.get(hash.as_slice())? {
let ec = ErrorCounter::decode(&ec);
if let Some(ec) = self.errors.get(&hash)? {
if now < ec.next_try() {
// if next retry after an error is not yet,
// don't do resync and return early, but still
@@ -277,7 +316,7 @@ impl BlockResyncManager {
// is not removing the one we added just above
// (we want to do the remove after the insert to ensure
// that the item is not lost if we crash in-between)
self.queue.remove(&block.time_bytes)?;
self.queue.remove(&block.key)?;
return Ok(ResyncIterResult::BusyDidNothing);
}
}
@@ -307,21 +346,21 @@ impl BlockResyncManager {
manager.metrics.resync_error_counter.add(1);
error!("Error when resyncing {:?}: {}", hash, e);
let err_counter = match self.errors.get(hash.as_slice())? {
Some(ec) => ErrorCounter::decode(&ec).add1(now + 1),
let err_counter = match self.errors.get(&hash)? {
Some(ec) => ec.add1(now + 1),
None => ErrorCounter::new(now + 1),
};
self.errors.insert(hash.as_slice(), err_counter.encode())?;
self.errors.insert(&hash, &err_counter)?;
self.put_to_resync_at(&hash, err_counter.next_try())?;
// err_counter.next_try() >= now + 1 > now,
// the entry we remove from the queue is not
// the entry we inserted with put_to_resync_at
self.queue.remove(&block.time_bytes)?;
self.queue.remove(&block.key)?;
} else {
self.errors.remove(hash.as_slice())?;
self.queue.remove(&block.time_bytes)?;
self.errors.remove(&hash)?;
self.queue.remove(&block.key)?;
}
Ok(ResyncIterResult::BusyDidSomething)
@@ -344,12 +383,11 @@ impl BlockResyncManager {
fn get_block_to_resync(&self) -> Result<Option<BusyBlock>, db::Error> {
let mut busy = self.busy_set.lock().unwrap();
for it in self.queue.iter()? {
let (time_bytes, hash_bytes) = it?;
if !busy.contains(&time_bytes) {
busy.insert(time_bytes.clone());
let (key, _) = it?;
if !busy.contains(&key) {
busy.insert(key);
return Ok(Some(BusyBlock {
time_bytes,
hash_bytes,
key,
busy_set: self.busy_set.clone(),
}));
}
@@ -506,7 +544,7 @@ impl BlockResyncManager {
impl Drop for BusyBlock {
fn drop(&mut self) {
let mut busy = self.busy_set.lock().unwrap();
busy.remove(&self.time_bytes);
busy.remove(&self.key);
}
}
@@ -516,6 +554,7 @@ pub(crate) struct ResyncWorker {
tranquilizer: Tranquilizer,
next_delay: Duration,
persister: PersisterShared<ResyncPersistedConfig>,
had_decode_error: bool,
}
impl ResyncWorker {
@@ -527,6 +566,7 @@ impl ResyncWorker {
tranquilizer: Tranquilizer::new(30),
next_delay: Duration::from_secs(10),
persister,
had_decode_error: false,
}
}
}
@@ -574,6 +614,15 @@ impl Worker for ResyncWorker {
self.next_delay = delay;
Ok(WorkerState::Idle)
}
Err(db::Error::Decode(e)) => {
// We give it one second chance in the very unlikely case that the bytes would somehow
// have been corrupted during read and that a new read would lead to a correct decoding.
if self.had_decode_error {
panic!("An error has happened when decoding something stored in the local k/v store: {}.", e);
}
self.had_decode_error = true;
Ok(WorkerState::Busy)
}
Err(e) => {
// The errors that we have here are only db errors
// We don't really know how to handle them so just ¯\_(ツ)_/¯
@@ -581,6 +630,7 @@ impl Worker for ResyncWorker {
// if it does there is not much we can do -- TODO should we just panic?)
// Here we just give the error to the worker manager,
// it will print it to the logs and increment a counter
self.had_decode_error = false;
Err(e.into())
}
}
@@ -602,6 +652,7 @@ impl Worker for ResyncWorker {
/// Counts the number of errors when resyncing a block,
/// and the time of the last try.
///
/// Used to implement exponential backoff.
#[derive(Clone, Copy, Debug)]
pub(crate) struct ErrorCounter {
@@ -609,6 +660,31 @@ pub(crate) struct ErrorCounter {
pub(crate) last_try: u64,
}
impl db::DbBytes for ErrorCounter {
fn encode(&self) -> Vec<u8> {
let mut v = Vec::with_capacity(16);
v.extend_from_slice(&u64::to_be_bytes(self.errors));
v.extend_from_slice(&u64::to_be_bytes(self.last_try));
v
}
fn decode(bytes: &[u8]) -> std::result::Result<Self, db::DecodeError> {
if bytes.len() != 16 {
return Err(db::DecodeError(
format!(
"invalid error counter: expected 16 bytes, got {}",
bytes.len()
)
.into(),
));
}
Ok(Self {
errors: u64::from_be_bytes(bytes[..8].try_into().unwrap()),
last_try: u64::from_be_bytes(bytes[8..].try_into().unwrap()),
})
}
}
impl ErrorCounter {
fn new(now: u64) -> Self {
Self {
@@ -617,21 +693,6 @@ impl ErrorCounter {
}
}
pub(crate) fn decode(data: &[u8]) -> Self {
Self {
errors: u64::from_be_bytes(data[0..8].try_into().unwrap()),
last_try: u64::from_be_bytes(data[8..16].try_into().unwrap()),
}
}
fn encode(&self) -> Vec<u8> {
[
u64::to_be_bytes(self.errors),
u64::to_be_bytes(self.last_try),
]
.concat()
}
fn add1(self, now: u64) -> Self {
Self {
errors: self.errors + 1,
+56 -50
View File
@@ -12,7 +12,7 @@ use fjall::{
use crate::{
open::{Engine, OpenOpt},
Db, Error, IDb, ITx, ITxFn, OnCommit, Result, TxError, TxFnResult, TxOpError, TxOpResult,
Db, DbError, DbResult, Error, IDb, ITx, ITxFn, OnCommit, TxError, TxFnResult, TxOpError,
TxResult, TxValueIter, Value, ValueIter,
};
@@ -20,10 +20,10 @@ pub use fjall;
// --
pub(crate) fn open_db(path: &Path, opt: &OpenOpt) -> Result<Db> {
pub(crate) fn open_db(path: &Path, opt: &OpenOpt) -> DbResult<Db> {
info!("Opening Fjall database at: {}", path.display());
if opt.fsync {
return Err(Error(
return Err(DbError(
"metadata_fsync is not supported with the Fjall database engine".into(),
));
}
@@ -37,21 +37,27 @@ pub(crate) fn open_db(path: &Path, opt: &OpenOpt) -> Result<Db> {
// -- err
impl From<fjall::Error> for Error {
fn from(e: fjall::Error) -> Error {
Error(format!("fjall: {}", e).into())
impl From<fjall::Error> for DbError {
fn from(e: fjall::Error) -> DbError {
DbError(format!("fjall: {}", e).into())
}
}
impl From<fjall::LsmError> for Error {
fn from(e: fjall::LsmError) -> Error {
Error(format!("fjall lsm_tree: {}", e).into())
impl From<fjall::LsmError> for DbError {
fn from(e: fjall::LsmError) -> DbError {
DbError(format!("fjall lsm_tree: {}", e).into())
}
}
impl From<fjall::Error> for Error {
fn from(e: fjall::Error) -> Error {
Error::Db(DbError::from(e))
}
}
impl From<fjall::Error> for TxOpError {
fn from(e: fjall::Error) -> TxOpError {
TxOpError(e.into())
DbError::from(e).into()
}
}
@@ -76,11 +82,11 @@ impl FjallDb {
fn get_tree(
&self,
i: usize,
) -> Result<MappedRwLockReadGuard<'_, TransactionalPartitionHandle>> {
) -> DbResult<MappedRwLockReadGuard<'_, TransactionalPartitionHandle>> {
RwLockReadGuard::try_map(self.trees.read(), |trees: &Vec<_>| {
trees.get(i).map(|tup| &tup.1)
})
.map_err(|_| Error("invalid tree id".into()))
.map_err(|_| DbError("invalid tree id".into()))
}
}
@@ -89,7 +95,7 @@ impl IDb for FjallDb {
"Fjall (EXPERIMENTAL!)".into()
}
fn open_tree(&self, name: &str) -> Result<usize> {
fn open_tree(&self, name: &str) -> DbResult<usize> {
let mut trees = self.trees.write();
let safe_name = encode_name(name)?;
if let Some(i) = trees.iter().position(|(name, _)| *name == safe_name) {
@@ -104,15 +110,15 @@ impl IDb for FjallDb {
}
}
fn list_trees(&self) -> Result<Vec<String>> {
fn list_trees(&self) -> DbResult<Vec<String>> {
self.keyspace
.list_partitions()
.iter()
.map(|n| decode_name(n))
.collect::<Result<Vec<_>>>()
.collect::<DbResult<Vec<_>>>()
}
fn snapshot(&self, base_path: &Path) -> Result<()> {
fn snapshot(&self, base_path: &Path) -> DbResult<()> {
std::fs::create_dir_all(base_path)?;
let path = Engine::Fjall.db_path(base_path);
@@ -138,7 +144,7 @@ impl IDb for FjallDb {
// ----
fn get(&self, tree_idx: usize, key: &[u8]) -> Result<Option<Value>> {
fn get(&self, tree_idx: usize, key: &[u8]) -> DbResult<Option<Value>> {
let tree = self.get_tree(tree_idx)?;
let tx = self.keyspace.read_tx();
let val = tx.get(&tree, key)?;
@@ -148,17 +154,17 @@ impl IDb for FjallDb {
}
}
fn approximate_len(&self, tree_idx: usize) -> Result<usize> {
fn approximate_len(&self, tree_idx: usize) -> DbResult<usize> {
let tree = self.get_tree(tree_idx)?;
Ok(tree.approximate_len())
}
fn is_empty(&self, tree_idx: usize) -> Result<bool> {
fn is_empty(&self, tree_idx: usize) -> DbResult<bool> {
let tree = self.get_tree(tree_idx)?;
let tx = self.keyspace.read_tx();
Ok(tx.is_empty(&tree)?)
}
fn insert(&self, tree_idx: usize, key: &[u8], value: &[u8]) -> Result<()> {
fn insert(&self, tree_idx: usize, key: &[u8], value: &[u8]) -> DbResult<()> {
let tree = self.get_tree(tree_idx)?;
let mut tx = self.keyspace.write_tx();
tx.insert(&tree, key, value);
@@ -166,7 +172,7 @@ impl IDb for FjallDb {
Ok(())
}
fn remove(&self, tree_idx: usize, key: &[u8]) -> Result<()> {
fn remove(&self, tree_idx: usize, key: &[u8]) -> DbResult<()> {
let tree = self.get_tree(tree_idx)?;
let mut tx = self.keyspace.write_tx();
tx.remove(&tree, key);
@@ -174,11 +180,11 @@ impl IDb for FjallDb {
Ok(())
}
fn clear(&self, tree_idx: usize) -> Result<()> {
fn clear(&self, tree_idx: usize) -> DbResult<()> {
let mut trees = self.trees.write();
if tree_idx >= trees.len() {
return Err(Error("invalid tree id".into()));
return Err(DbError("invalid tree id".into()));
}
let (name, tree) = trees.remove(tree_idx);
@@ -191,13 +197,13 @@ impl IDb for FjallDb {
Ok(())
}
fn iter(&self, tree_idx: usize) -> Result<ValueIter<'_>> {
fn iter(&self, tree_idx: usize) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree_idx)?;
let tx = self.keyspace.read_tx();
Ok(Box::new(tx.iter(&tree).map(iterator_remap)))
}
fn iter_rev(&self, tree_idx: usize) -> Result<ValueIter<'_>> {
fn iter_rev(&self, tree_idx: usize) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree_idx)?;
let tx = self.keyspace.read_tx();
Ok(Box::new(tx.iter(&tree).rev().map(iterator_remap)))
@@ -208,7 +214,7 @@ impl IDb for FjallDb {
tree_idx: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> Result<ValueIter<'_>> {
) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree_idx)?;
let tx = self.keyspace.read_tx();
Ok(Box::new(
@@ -221,7 +227,7 @@ impl IDb for FjallDb {
tree_idx: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> Result<ValueIter<'_>> {
) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree_idx)?;
let tx = self.keyspace.read_tx();
Ok(Box::new(
@@ -252,9 +258,9 @@ impl IDb for FjallDb {
}
TxFnResult::DbErr => {
tx.tx.rollback();
Err(TxError::Db(Error(
"(this message will be discarded)".into(),
)))
Err(TxError::Db(
DbError("(this message will be discarded)".into()).into(),
))
}
}
}
@@ -268,47 +274,47 @@ struct FjallTx<'a> {
}
impl<'a> FjallTx<'a> {
fn get_tree(&self, i: usize) -> TxOpResult<&TransactionalPartitionHandle> {
fn get_tree(&self, i: usize) -> DbResult<&TransactionalPartitionHandle> {
self.trees.get(i).map(|tup| &tup.1).ok_or_else(|| {
TxOpError(Error(
DbError(
"invalid tree id (it might have been opened after the transaction started)".into(),
))
)
})
}
}
impl<'a> ITx for FjallTx<'a> {
fn get(&self, tree_idx: usize, key: &[u8]) -> TxOpResult<Option<Value>> {
fn get(&self, tree_idx: usize, key: &[u8]) -> DbResult<Option<Value>> {
let tree = self.get_tree(tree_idx)?;
match self.tx.get(tree, key)? {
Some(v) => Ok(Some(v.to_vec())),
None => Ok(None),
}
}
fn len(&self, tree_idx: usize) -> TxOpResult<usize> {
fn len(&self, tree_idx: usize) -> DbResult<usize> {
let tree = self.get_tree(tree_idx)?;
Ok(self.tx.len(tree)?)
}
fn insert(&mut self, tree_idx: usize, key: &[u8], value: &[u8]) -> TxOpResult<()> {
fn insert(&mut self, tree_idx: usize, key: &[u8], value: &[u8]) -> DbResult<()> {
let tree = self.get_tree(tree_idx)?.clone();
self.tx.insert(&tree, key, value);
Ok(())
}
fn remove(&mut self, tree_idx: usize, key: &[u8]) -> TxOpResult<()> {
fn remove(&mut self, tree_idx: usize, key: &[u8]) -> DbResult<()> {
let tree = self.get_tree(tree_idx)?.clone();
self.tx.remove(&tree, key);
Ok(())
}
fn clear(&mut self, _tree_idx: usize) -> TxOpResult<()> {
fn clear(&mut self, _tree_idx: usize) -> DbResult<()> {
unimplemented!("LSM tree clearing in cross-partition transaction is not supported")
}
fn iter(&self, tree_idx: usize) -> TxOpResult<TxValueIter<'_>> {
fn iter(&self, tree_idx: usize) -> DbResult<TxValueIter<'_>> {
let tree = self.get_tree(tree_idx)?.clone();
Ok(Box::new(self.tx.iter(&tree).map(iterator_remap_tx)))
}
fn iter_rev(&self, tree_idx: usize) -> TxOpResult<TxValueIter<'_>> {
fn iter_rev(&self, tree_idx: usize) -> DbResult<TxValueIter<'_>> {
let tree = self.get_tree(tree_idx)?.clone();
Ok(Box::new(self.tx.iter(&tree).rev().map(iterator_remap_tx)))
}
@@ -318,7 +324,7 @@ impl<'a> ITx for FjallTx<'a> {
tree_idx: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> TxOpResult<TxValueIter<'_>> {
) -> DbResult<TxValueIter<'_>> {
let tree = self.get_tree(tree_idx)?;
let low = clone_bound(low);
let high = clone_bound(high);
@@ -333,7 +339,7 @@ impl<'a> ITx for FjallTx<'a> {
tree_idx: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> TxOpResult<TxValueIter<'_>> {
) -> DbResult<TxValueIter<'_>> {
let tree = self.get_tree(tree_idx)?;
let low = clone_bound(low);
let high = clone_bound(high);
@@ -348,14 +354,14 @@ impl<'a> ITx for FjallTx<'a> {
// -- maps fjall's (k, v) to ours
fn iterator_remap(r: fjall::Result<(fjall::Slice, fjall::Slice)>) -> Result<(Value, Value)> {
fn iterator_remap(r: fjall::Result<(fjall::Slice, fjall::Slice)>) -> DbResult<(Value, Value)> {
r.map(|(k, v)| (k.to_vec(), v.to_vec()))
.map_err(|e| e.into())
.map_err(DbError::from)
}
fn iterator_remap_tx(r: fjall::Result<(fjall::Slice, fjall::Slice)>) -> TxOpResult<(Value, Value)> {
fn iterator_remap_tx(r: fjall::Result<(fjall::Slice, fjall::Slice)>) -> DbResult<(Value, Value)> {
r.map(|(k, v)| (k.to_vec(), v.to_vec()))
.map_err(|e| e.into())
.map_err(DbError::from)
}
// -- utils to deal with Garage's tightness on Bound lifetimes
@@ -378,7 +384,7 @@ fn clone_bound(bound: Bound<&[u8]>) -> ByteVecBound {
// -- utils to encode table names --
fn encode_name(s: &str) -> Result<String> {
fn encode_name(s: &str) -> DbResult<String> {
let base = 'A' as u32;
let mut ret = String::with_capacity(s.len() + 10);
@@ -392,7 +398,7 @@ fn encode_name(s: &str) -> Result<String> {
ret.push(char::from_u32(base + c_hi).unwrap());
ret.push(char::from_u32(base + c_lo).unwrap());
} else {
return Err(Error(
return Err(DbError(
format!("table name {} could not be safely encoded", s).into(),
));
}
@@ -400,10 +406,10 @@ fn encode_name(s: &str) -> Result<String> {
Ok(ret)
}
fn decode_name(s: &str) -> Result<String> {
fn decode_name(s: &str) -> DbResult<String> {
use std::convert::TryFrom;
let errfn = || Error(format!("encoded table name {} is invalid", s).into());
let errfn = || DbError(format!("encoded table name {} is invalid", s).into());
let c_map = |c: char| {
let c = c as u32;
let base = 'A' as u32;
+97 -57
View File
@@ -9,6 +9,7 @@ pub mod lmdb_adapter;
pub mod sqlite_adapter;
pub mod open;
pub mod typed;
#[cfg(test)]
pub mod test;
@@ -23,6 +24,7 @@ use std::sync::Arc;
use thiserror::Error;
pub use open::*;
pub use typed::{DbBytes, DbOrdKey, TypedIter, TypedTree, TypedTxIter};
pub(crate) type OnCommit = Vec<Box<dyn FnOnce()>>;
@@ -38,21 +40,34 @@ pub struct Transaction<'a> {
pub struct Tree(Arc<dyn IDb>, usize);
pub type Value = Vec<u8>;
pub type ValueIter<'a> = Box<dyn std::iter::Iterator<Item = Result<(Value, Value)>> + 'a>;
pub type TxValueIter<'a> = Box<dyn std::iter::Iterator<Item = TxOpResult<(Value, Value)>> + 'a>;
pub type ValueIter<'a> = Box<dyn std::iter::Iterator<Item = DbResult<(Value, Value)>> + 'a>;
pub type TxValueIter<'a> = Box<dyn std::iter::Iterator<Item = DbResult<(Value, Value)>> + 'a>;
// ----
#[derive(Debug, Error)]
#[error("{0}")]
pub struct Error(pub Cow<'static, str>);
#[error("database error: {0}")]
pub struct DbError(pub Cow<'static, str>);
impl From<std::io::Error> for Error {
fn from(e: std::io::Error) -> Error {
Error(format!("IO: {}", e).into())
#[derive(Debug, Error)]
#[error("decode error: {0}")]
pub struct DecodeError(pub Cow<'static, str>);
#[derive(Debug, Error)]
pub enum Error {
#[error(transparent)]
Db(#[from] DbError),
#[error(transparent)]
Decode(#[from] DecodeError),
}
impl From<std::io::Error> for DbError {
fn from(e: std::io::Error) -> DbError {
DbError(format!("IO: {}", e).into())
}
}
pub type DbResult<T> = std::result::Result<T, DbError>;
pub type Result<T> = std::result::Result<T, Error>;
#[derive(Debug, Error)]
@@ -60,6 +75,18 @@ pub type Result<T> = std::result::Result<T, Error>;
pub struct TxOpError(pub(crate) Error);
pub type TxOpResult<T> = std::result::Result<T, TxOpError>;
impl From<DbError> for TxOpError {
fn from(e: DbError) -> TxOpError {
TxOpError(e.into())
}
}
impl From<DecodeError> for TxOpError {
fn from(e: DecodeError) -> TxOpError {
TxOpError(e.into())
}
}
#[derive(Debug)]
pub enum TxError<E> {
Abort(E),
@@ -73,6 +100,12 @@ impl<E> From<TxOpError> for TxError<E> {
}
}
impl<E> From<DbError> for TxError<E> {
fn from(e: DbError) -> TxError<E> {
TxError::Db(e.into())
}
}
pub fn unabort<R, E>(res: TxResult<R, E>) -> TxOpResult<std::result::Result<R, E>> {
match res {
Ok(v) => Ok(Ok(v)),
@@ -88,12 +121,12 @@ impl Db {
self.0.engine()
}
pub fn open_tree<S: AsRef<str>>(&self, name: S) -> Result<Tree> {
pub fn open_tree<S: AsRef<str>>(&self, name: S) -> DbResult<Tree> {
let tree_id = self.0.open_tree(name.as_ref())?;
Ok(Tree(self.0.clone(), tree_id))
}
pub fn list_trees(&self) -> Result<Vec<String>> {
pub fn list_trees(&self) -> DbResult<Vec<String>> {
self.0.list_trees()
}
@@ -147,27 +180,28 @@ impl Db {
}
}
pub fn snapshot(&self, path: &Path) -> Result<()> {
pub fn snapshot(&self, path: &Path) -> DbResult<()> {
self.0.snapshot(path)
}
pub fn import(&self, other: &Db) -> Result<()> {
let existing_trees = self.list_trees()?;
if !existing_trees.is_empty() {
return Err(Error(
return Err(DbError(
format!(
"destination database already contains data: {:?}",
existing_trees
)
.into(),
));
)
.into());
}
let tree_names = other.list_trees()?;
for name in tree_names {
let tree = self.open_tree(&name)?;
if !tree.is_empty()? {
return Err(Error(format!("tree {} already contains data", name).into()));
return Err(DbError(format!("tree {} already contains data", name).into()).into());
}
let ex_tree = other.open_tree(&name)?;
@@ -186,7 +220,7 @@ impl Db {
});
let total = match tx_res {
Err(TxError::Db(e)) => return Err(e),
Err(TxError::Abort(e)) => return Err(e),
Err(TxError::Abort(e)) => return Err(e.into()),
Ok(x) => x,
};
@@ -204,24 +238,24 @@ impl Tree {
}
#[inline]
pub fn get<T: AsRef<[u8]>>(&self, key: T) -> Result<Option<Value>> {
pub fn get<T: AsRef<[u8]>>(&self, key: T) -> DbResult<Option<Value>> {
self.0.get(self.1, key.as_ref())
}
#[inline]
pub fn approximate_len(&self) -> Result<usize> {
pub fn approximate_len(&self) -> DbResult<usize> {
self.0.approximate_len(self.1)
}
#[inline]
pub fn is_empty(&self) -> Result<bool> {
pub fn is_empty(&self) -> DbResult<bool> {
self.0.is_empty(self.1)
}
#[inline]
pub fn first(&self) -> Result<Option<(Value, Value)>> {
pub fn first(&self) -> DbResult<Option<(Value, Value)>> {
self.iter()?.next().transpose()
}
#[inline]
pub fn get_gt<T: AsRef<[u8]>>(&self, from: T) -> Result<Option<(Value, Value)>> {
pub fn get_gt<T: AsRef<[u8]>>(&self, from: T) -> DbResult<Option<(Value, Value)>> {
if from.as_ref().is_empty() {
self.iter()?.next().transpose()
} else {
@@ -233,31 +267,31 @@ impl Tree {
/// Returns the old value if there was one
#[inline]
pub fn insert<T: AsRef<[u8]>, U: AsRef<[u8]>>(&self, key: T, value: U) -> Result<()> {
pub fn insert<T: AsRef<[u8]>, U: AsRef<[u8]>>(&self, key: T, value: U) -> DbResult<()> {
self.0.insert(self.1, key.as_ref(), value.as_ref())
}
/// Returns the old value if there was one
#[inline]
pub fn remove<T: AsRef<[u8]>>(&self, key: T) -> Result<()> {
pub fn remove<T: AsRef<[u8]>>(&self, key: T) -> DbResult<()> {
self.0.remove(self.1, key.as_ref())
}
/// Clears all values from the tree
#[inline]
pub fn clear(&self) -> Result<()> {
pub fn clear(&self) -> DbResult<()> {
self.0.clear(self.1)
}
#[inline]
pub fn iter(&self) -> Result<ValueIter<'_>> {
pub fn iter(&self) -> DbResult<ValueIter<'_>> {
self.0.iter(self.1)
}
#[inline]
pub fn iter_rev(&self) -> Result<ValueIter<'_>> {
pub fn iter_rev(&self) -> DbResult<ValueIter<'_>> {
self.0.iter_rev(self.1)
}
#[inline]
pub fn range<K, R>(&self, range: R) -> Result<ValueIter<'_>>
pub fn range<K, R>(&self, range: R) -> DbResult<ValueIter<'_>>
where
K: AsRef<[u8]>,
R: RangeBounds<K>,
@@ -267,7 +301,7 @@ impl Tree {
self.0.range(self.1, get_bound(sb), get_bound(eb))
}
#[inline]
pub fn range_rev<K, R>(&self, range: R) -> Result<ValueIter<'_>>
pub fn range_rev<K, R>(&self, range: R) -> DbResult<ValueIter<'_>>
where
K: AsRef<[u8]>,
R: RangeBounds<K>,
@@ -282,11 +316,11 @@ impl Tree {
impl<'a> Transaction<'a> {
#[inline]
pub fn get<T: AsRef<[u8]>>(&self, tree: &Tree, key: T) -> TxOpResult<Option<Value>> {
self.tx.get(tree.1, key.as_ref())
self.tx.get(tree.1, key.as_ref()).map_err(Into::into)
}
#[inline]
pub fn len(&self, tree: &Tree) -> TxOpResult<usize> {
self.tx.len(tree.1)
self.tx.len(tree.1).map_err(Into::into)
}
/// Returns the old value if there was one
@@ -297,26 +331,28 @@ impl<'a> Transaction<'a> {
key: T,
value: U,
) -> TxOpResult<()> {
self.tx.insert(tree.1, key.as_ref(), value.as_ref())
self.tx
.insert(tree.1, key.as_ref(), value.as_ref())
.map_err(Into::into)
}
/// Returns the old value if there was one
#[inline]
pub fn remove<T: AsRef<[u8]>>(&mut self, tree: &Tree, key: T) -> TxOpResult<()> {
self.tx.remove(tree.1, key.as_ref())
self.tx.remove(tree.1, key.as_ref()).map_err(Into::into)
}
/// Clears all values in a tree
#[inline]
pub fn clear(&mut self, tree: &Tree) -> TxOpResult<()> {
self.tx.clear(tree.1)
self.tx.clear(tree.1).map_err(Into::into)
}
#[inline]
pub fn iter(&self, tree: &Tree) -> TxOpResult<TxValueIter<'_>> {
self.tx.iter(tree.1)
self.tx.iter(tree.1).map_err(Into::into)
}
#[inline]
pub fn iter_rev(&self, tree: &Tree) -> TxOpResult<TxValueIter<'_>> {
self.tx.iter_rev(tree.1)
self.tx.iter_rev(tree.1).map_err(Into::into)
}
#[inline]
@@ -327,7 +363,9 @@ impl<'a> Transaction<'a> {
{
let sb = range.start_bound();
let eb = range.end_bound();
self.tx.range(tree.1, get_bound(sb), get_bound(eb))
self.tx
.range(tree.1, get_bound(sb), get_bound(eb))
.map_err(Into::into)
}
#[inline]
pub fn range_rev<K, R>(&self, tree: &Tree, range: R) -> TxOpResult<TxValueIter<'_>>
@@ -337,7 +375,9 @@ impl<'a> Transaction<'a> {
{
let sb = range.start_bound();
let eb = range.end_bound();
self.tx.range_rev(tree.1, get_bound(sb), get_bound(eb))
self.tx
.range_rev(tree.1, get_bound(sb), get_bound(eb))
.map_err(Into::into)
}
#[inline]
@@ -350,60 +390,60 @@ impl<'a> Transaction<'a> {
pub(crate) trait IDb: Send + Sync {
fn engine(&self) -> String;
fn open_tree(&self, name: &str) -> Result<usize>;
fn list_trees(&self) -> Result<Vec<String>>;
fn snapshot(&self, path: &Path) -> Result<()>;
fn open_tree(&self, name: &str) -> DbResult<usize>;
fn list_trees(&self) -> DbResult<Vec<String>>;
fn snapshot(&self, path: &Path) -> DbResult<()>;
fn get(&self, tree: usize, key: &[u8]) -> Result<Option<Value>>;
fn approximate_len(&self, tree: usize) -> Result<usize>;
fn is_empty(&self, tree: usize) -> Result<bool>;
fn get(&self, tree: usize, key: &[u8]) -> DbResult<Option<Value>>;
fn approximate_len(&self, tree: usize) -> DbResult<usize>;
fn is_empty(&self, tree: usize) -> DbResult<bool>;
fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> Result<()>;
fn remove(&self, tree: usize, key: &[u8]) -> Result<()>;
fn clear(&self, tree: usize) -> Result<()>;
fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> DbResult<()>;
fn remove(&self, tree: usize, key: &[u8]) -> DbResult<()>;
fn clear(&self, tree: usize) -> DbResult<()>;
fn iter(&self, tree: usize) -> Result<ValueIter<'_>>;
fn iter_rev(&self, tree: usize) -> Result<ValueIter<'_>>;
fn iter(&self, tree: usize) -> DbResult<ValueIter<'_>>;
fn iter_rev(&self, tree: usize) -> DbResult<ValueIter<'_>>;
fn range<'r>(
&self,
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> Result<ValueIter<'_>>;
) -> DbResult<ValueIter<'_>>;
fn range_rev<'r>(
&self,
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> Result<ValueIter<'_>>;
) -> DbResult<ValueIter<'_>>;
fn transaction(&self, f: &dyn ITxFn) -> TxResult<OnCommit, ()>;
}
pub(crate) trait ITx {
fn get(&self, tree: usize, key: &[u8]) -> TxOpResult<Option<Value>>;
fn len(&self, tree: usize) -> TxOpResult<usize>;
fn get(&self, tree: usize, key: &[u8]) -> DbResult<Option<Value>>;
fn len(&self, tree: usize) -> DbResult<usize>;
fn insert(&mut self, tree: usize, key: &[u8], value: &[u8]) -> TxOpResult<()>;
fn remove(&mut self, tree: usize, key: &[u8]) -> TxOpResult<()>;
fn clear(&mut self, tree: usize) -> TxOpResult<()>;
fn insert(&mut self, tree: usize, key: &[u8], value: &[u8]) -> DbResult<()>;
fn remove(&mut self, tree: usize, key: &[u8]) -> DbResult<()>;
fn clear(&mut self, tree: usize) -> DbResult<()>;
fn iter(&self, tree: usize) -> TxOpResult<TxValueIter<'_>>;
fn iter_rev(&self, tree: usize) -> TxOpResult<TxValueIter<'_>>;
fn iter(&self, tree: usize) -> DbResult<TxValueIter<'_>>;
fn iter_rev(&self, tree: usize) -> DbResult<TxValueIter<'_>>;
fn range<'r>(
&self,
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> TxOpResult<TxValueIter<'_>>;
) -> DbResult<TxValueIter<'_>>;
fn range_rev<'r>(
&self,
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> TxOpResult<TxValueIter<'_>>;
) -> DbResult<TxValueIter<'_>>;
}
pub(crate) trait ITxFn {
+49 -43
View File
@@ -14,7 +14,7 @@ type Database = heed::Database<Bytes, Bytes>;
use crate::{
open::{Engine, OpenOpt},
Db, Error, IDb, ITx, ITxFn, OnCommit, Result, TxError, TxFnResult, TxOpError, TxOpResult,
Db, DbError, DbResult, Error, IDb, ITx, ITxFn, OnCommit, TxError, TxFnResult, TxOpError,
TxResult, TxValueIter, Value, ValueIter,
};
@@ -22,10 +22,10 @@ pub use heed;
// ---- top-level open function
pub(crate) fn open_db(path: &PathBuf, opt: &OpenOpt) -> Result<Db> {
pub(crate) fn open_db(path: &PathBuf, opt: &OpenOpt) -> DbResult<Db> {
info!("Opening LMDB database at: {}", path.display());
if let Err(e) = std::fs::create_dir_all(path) {
return Err(Error(
return Err(DbError(
format!("Unable to create LMDB data directory: {}", e).into(),
));
}
@@ -48,7 +48,7 @@ pub(crate) fn open_db(path: &PathBuf, opt: &OpenOpt) -> Result<Db> {
env_builder.open(path)
};
match open_res {
Err(heed::Error::Io(e)) if e.kind() == std::io::ErrorKind::OutOfMemory => Err(Error(
Err(heed::Error::Io(e)) if e.kind() == std::io::ErrorKind::OutOfMemory => Err(DbError(
"OutOfMemory error while trying to open LMDB database. This can happen \
if your operating system is not allowing you to use sufficient virtual \
memory address space. Please check that no limit is set (ulimit -v). \
@@ -56,22 +56,28 @@ pub(crate) fn open_db(path: &PathBuf, opt: &OpenOpt) -> Result<Db> {
On 32-bit machines, you should probably switch to another database engine."
.into(),
)),
Err(e) => Err(Error(format!("Cannot open LMDB database: {}", e).into())),
Err(e) => Err(DbError(format!("Cannot open LMDB database: {}", e).into())),
Ok(db) => Ok(LmdbDb::init(db)),
}
}
// -- err
impl From<heed::Error> for DbError {
fn from(e: heed::Error) -> DbError {
DbError(format!("LMDB: {}", e).into())
}
}
impl From<heed::Error> for Error {
fn from(e: heed::Error) -> Error {
Error(format!("LMDB: {}", e).into())
DbError::from(e).into()
}
}
impl From<heed::Error> for TxOpError {
fn from(e: heed::Error) -> TxOpError {
TxOpError(e.into())
DbError::from(e).into()
}
}
@@ -91,14 +97,14 @@ impl LmdbDb {
Db(Arc::new(s))
}
fn get_tree(&self, i: usize) -> Result<Database> {
fn get_tree(&self, i: usize) -> DbResult<Database> {
self.trees
.read()
.unwrap()
.0
.get(i)
.cloned()
.ok_or_else(|| Error("invalid tree id".into()))
.ok_or_else(|| DbError("invalid tree id".into()))
}
}
@@ -107,7 +113,7 @@ impl IDb for LmdbDb {
"LMDB (using Heed crate)".into()
}
fn open_tree(&self, name: &str) -> Result<usize> {
fn open_tree(&self, name: &str) -> DbResult<usize> {
let mut trees = self.trees.write().unwrap();
if let Some(i) = trees.1.get(name) {
Ok(*i)
@@ -122,7 +128,7 @@ impl IDb for LmdbDb {
}
}
fn list_trees(&self) -> Result<Vec<String>> {
fn list_trees(&self) -> DbResult<Vec<String>> {
let rtxn = self.db.read_txn()?;
let tree0 = match self
.db
@@ -153,7 +159,7 @@ impl IDb for LmdbDb {
Ok(ret2)
}
fn snapshot(&self, base_path: &Path) -> Result<()> {
fn snapshot(&self, base_path: &Path) -> DbResult<()> {
std::fs::create_dir_all(base_path)?;
let path = Engine::Lmdb.db_path(base_path);
self.db
@@ -163,7 +169,7 @@ impl IDb for LmdbDb {
// ----
fn get(&self, tree: usize, key: &[u8]) -> Result<Option<Value>> {
fn get(&self, tree: usize, key: &[u8]) -> DbResult<Option<Value>> {
let tree = self.get_tree(tree)?;
let tx = self.db.read_txn()?;
@@ -174,18 +180,18 @@ impl IDb for LmdbDb {
}
}
fn approximate_len(&self, tree: usize) -> Result<usize> {
fn approximate_len(&self, tree: usize) -> DbResult<usize> {
let tree = self.get_tree(tree)?;
let tx = self.db.read_txn()?;
Ok(tree.len(&tx)?.try_into().unwrap())
}
fn is_empty(&self, tree: usize) -> Result<bool> {
fn is_empty(&self, tree: usize) -> DbResult<bool> {
let tree = self.get_tree(tree)?;
let tx = self.db.read_txn()?;
Ok(tree.is_empty(&tx)?)
}
fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> Result<()> {
fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> DbResult<()> {
let tree = self.get_tree(tree)?;
let mut tx = self.db.write_txn()?;
tree.put(&mut tx, key, value)?;
@@ -193,7 +199,7 @@ impl IDb for LmdbDb {
Ok(())
}
fn remove(&self, tree: usize, key: &[u8]) -> Result<()> {
fn remove(&self, tree: usize, key: &[u8]) -> DbResult<()> {
let tree = self.get_tree(tree)?;
let mut tx = self.db.write_txn()?;
tree.delete(&mut tx, key)?;
@@ -201,7 +207,7 @@ impl IDb for LmdbDb {
Ok(())
}
fn clear(&self, tree: usize) -> Result<()> {
fn clear(&self, tree: usize) -> DbResult<()> {
let tree = self.get_tree(tree)?;
let mut tx = self.db.write_txn()?;
tree.clear(&mut tx)?;
@@ -209,14 +215,14 @@ impl IDb for LmdbDb {
Ok(())
}
fn iter(&self, tree: usize) -> Result<ValueIter<'_>> {
fn iter(&self, tree: usize) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree)?;
let tx = self.db.read_txn()?;
// Safety: the cloture does not store its argument anywhere,
unsafe { TxAndIterator::make(tx, |tx| Ok(tree.iter(tx)?)) }
}
fn iter_rev(&self, tree: usize) -> Result<ValueIter<'_>> {
fn iter_rev(&self, tree: usize) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree)?;
let tx = self.db.read_txn()?;
// Safety: the cloture does not store its argument anywhere,
@@ -228,7 +234,7 @@ impl IDb for LmdbDb {
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> Result<ValueIter<'_>> {
) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree)?;
let tx = self.db.read_txn()?;
// Safety: the cloture does not store its argument anywhere,
@@ -239,7 +245,7 @@ impl IDb for LmdbDb {
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> Result<ValueIter<'_>> {
) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree)?;
let tx = self.db.read_txn()?;
// Safety: the cloture does not store its argument anywhere,
@@ -271,9 +277,9 @@ impl IDb for LmdbDb {
}
TxFnResult::DbErr => {
tx.tx.abort();
Err(TxError::Db(Error(
"(this message will be discarded)".into(),
)))
Err(TxError::Db(
DbError("(this message will be discarded)".into()).into(),
))
}
}
}
@@ -287,49 +293,49 @@ struct LmdbTx<'a> {
}
impl<'a> LmdbTx<'a> {
fn get_tree(&self, i: usize) -> TxOpResult<&Database> {
fn get_tree(&self, i: usize) -> DbResult<&Database> {
self.trees.get(i).ok_or_else(|| {
TxOpError(Error(
DbError(
"invalid tree id (it might have been opened after the transaction started)".into(),
))
)
})
}
}
impl<'a> ITx for LmdbTx<'a> {
fn get(&self, tree: usize, key: &[u8]) -> TxOpResult<Option<Value>> {
fn get(&self, tree: usize, key: &[u8]) -> DbResult<Option<Value>> {
let tree = self.get_tree(tree)?;
match tree.get(&self.tx, key)? {
Some(v) => Ok(Some(v.to_vec())),
None => Ok(None),
}
}
fn len(&self, tree: usize) -> TxOpResult<usize> {
fn len(&self, tree: usize) -> DbResult<usize> {
let tree = self.get_tree(tree)?;
Ok(tree.len(&self.tx)? as usize)
}
fn insert(&mut self, tree: usize, key: &[u8], value: &[u8]) -> TxOpResult<()> {
fn insert(&mut self, tree: usize, key: &[u8], value: &[u8]) -> DbResult<()> {
let tree = *self.get_tree(tree)?;
tree.put(&mut self.tx, key, value)?;
Ok(())
}
fn remove(&mut self, tree: usize, key: &[u8]) -> TxOpResult<()> {
fn remove(&mut self, tree: usize, key: &[u8]) -> DbResult<()> {
let tree = *self.get_tree(tree)?;
tree.delete(&mut self.tx, key)?;
Ok(())
}
fn clear(&mut self, tree: usize) -> TxOpResult<()> {
fn clear(&mut self, tree: usize) -> DbResult<()> {
let tree = *self.get_tree(tree)?;
tree.clear(&mut self.tx)?;
Ok(())
}
fn iter(&self, tree: usize) -> TxOpResult<TxValueIter<'_>> {
fn iter(&self, tree: usize) -> DbResult<TxValueIter<'_>> {
let tree = *self.get_tree(tree)?;
Ok(Box::new(tree.iter(&self.tx)?.map(tx_iter_item)))
}
fn iter_rev(&self, tree: usize) -> TxOpResult<TxValueIter<'_>> {
fn iter_rev(&self, tree: usize) -> DbResult<TxValueIter<'_>> {
let tree = *self.get_tree(tree)?;
Ok(Box::new(tree.rev_iter(&self.tx)?.map(tx_iter_item)))
}
@@ -339,7 +345,7 @@ impl<'a> ITx for LmdbTx<'a> {
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> TxOpResult<TxValueIter<'_>> {
) -> DbResult<TxValueIter<'_>> {
let tree = *self.get_tree(tree)?;
Ok(Box::new(
tree.range(&self.tx, &(low, high))?.map(tx_iter_item),
@@ -350,7 +356,7 @@ impl<'a> ITx for LmdbTx<'a> {
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> TxOpResult<TxValueIter<'_>> {
) -> DbResult<TxValueIter<'_>> {
let tree = *self.get_tree(tree)?;
Ok(Box::new(
tree.rev_range(&self.tx, &(low, high))?.map(tx_iter_item),
@@ -386,9 +392,9 @@ where
}
/// Safety: iterfun must not store its argument anywhere but in its result.
unsafe fn make<F>(tx: RoTxn<'a, WithTls>, iterfun: F) -> Result<ValueIter<'a>>
unsafe fn make<F>(tx: RoTxn<'a, WithTls>, iterfun: F) -> DbResult<ValueIter<'a>>
where
F: FnOnce(&'a RoTxn<'a>) -> Result<I>,
F: FnOnce(&'a RoTxn<'a>) -> DbResult<I>,
{
let res = TxAndIterator {
tx,
@@ -436,13 +442,13 @@ impl<'a, I> Iterator for TxAndIteratorPin<'a, I>
where
I: Iterator<Item = IteratorItem<'a>> + 'a,
{
type Item = Result<(Value, Value)>;
type Item = DbResult<(Value, Value)>;
fn next(&mut self) -> Option<Self::Item> {
let mut_ref = Pin::as_mut(&mut self.0);
let next = mut_ref.iter().as_mut()?.next()?;
let res = match next {
Err(e) => Err(e.into()),
Err(e) => Err(DbError::from(e)),
Ok((k, v)) => Ok((k.to_vec(), v.to_vec())),
};
Some(res)
@@ -453,9 +459,9 @@ where
fn tx_iter_item<'a>(
item: std::result::Result<(&'a [u8], &'a [u8]), heed::Error>,
) -> TxOpResult<(Vec<u8>, Vec<u8>)> {
) -> DbResult<(Vec<u8>, Vec<u8>)> {
item.map(|(k, v)| (k.to_vec(), v.to_vec()))
.map_err(|e| TxOpError(Error::from(e)))
.map_err(DbError::from)
}
// ---- utility ----
+10 -9
View File
@@ -1,6 +1,6 @@
use std::path::{Path, PathBuf};
use crate::{Db, Error, Result};
use crate::{Db, DbError, Error, Result};
/// List of supported database engine types
///
@@ -49,14 +49,14 @@ impl std::str::FromStr for Engine {
"lmdb" | "heed" => Ok(Self::Lmdb),
"sqlite" | "sqlite3" | "rusqlite" => Ok(Self::Sqlite),
"fjall" => Ok(Self::Fjall),
"sled" => Err(Error("Sled is no longer supported as a database engine. Converting your old metadata db can be done using an older Garage binary (e.g. v0.9.4).".into())),
kind => Err(Error(
"sled" => Err(DbError("Sled is no longer supported as a database engine. Converting your old metadata db can be done using an older Garage binary (e.g. v0.9.4).".into()).into()),
kind => Err(DbError(
format!(
"Invalid DB engine: {} (options are: lmdb, sqlite, fjall)",
kind
)
.into(),
)),
).into()),
}
}
}
@@ -72,22 +72,23 @@ pub fn open_db(path: &PathBuf, engine: Engine, opt: &OpenOpt) -> Result<Db> {
match engine {
// ---- Sqlite DB ----
#[cfg(feature = "sqlite")]
Engine::Sqlite => crate::sqlite_adapter::open_db(path, opt),
Engine::Sqlite => Ok(crate::sqlite_adapter::open_db(path, opt)?),
// ---- LMDB DB ----
#[cfg(feature = "lmdb")]
Engine::Lmdb => crate::lmdb_adapter::open_db(path, opt),
Engine::Lmdb => Ok(crate::lmdb_adapter::open_db(path, opt)?),
// ---- Fjall DB ----
#[cfg(feature = "fjall")]
Engine::Fjall => crate::fjall_adapter::open_db(path, opt),
Engine::Fjall => Ok(crate::fjall_adapter::open_db(path, opt)?),
// Pattern is unreachable when all supported DB engines are compiled into binary. The allow
// attribute is added so that we won't have to change this match in case stop building
// support for one or more engines by default.
#[allow(unreachable_patterns)]
engine => Err(Error(
engine => Err(DbError(
format!("DB engine support not available in this build: {}", engine).into(),
)),
)
.into()),
}
}
+55 -43
View File
@@ -12,7 +12,7 @@ use rusqlite::{params, Rows, Statement, Transaction};
use crate::{
open::{Engine, OpenOpt},
Db, Error, IDb, ITx, ITxFn, OnCommit, Result, TxError, TxFnResult, TxOpError, TxOpResult,
Db, DbError, DbResult, Error, IDb, ITx, ITxFn, OnCommit, TxError, TxFnResult, TxOpError,
TxResult, TxValueIter, Value, ValueIter,
};
@@ -20,7 +20,7 @@ pub use rusqlite;
// ---- top-level open function
pub(crate) fn open_db(path: &PathBuf, opt: &OpenOpt) -> Result<Db> {
pub(crate) fn open_db(path: &PathBuf, opt: &OpenOpt) -> DbResult<Db> {
info!("Opening Sqlite database at: {}", path.display());
let manager = r2d2_sqlite::SqliteConnectionManager::file(path);
SqliteDb::open(manager, opt.fsync)
@@ -32,21 +32,33 @@ type Connection = r2d2::PooledConnection<SqliteConnectionManager>;
// --- err
impl From<rusqlite::Error> for DbError {
fn from(e: rusqlite::Error) -> DbError {
DbError(format!("Sqlite: {}", e).into())
}
}
impl From<rusqlite::Error> for Error {
fn from(e: rusqlite::Error) -> Error {
Error(format!("Sqlite: {}", e).into())
DbError::from(e).into()
}
}
impl From<r2d2::Error> for DbError {
fn from(e: r2d2::Error) -> DbError {
DbError(format!("Sqlite: {}", e).into())
}
}
impl From<r2d2::Error> for Error {
fn from(e: r2d2::Error) -> Error {
Error(format!("Sqlite: {}", e).into())
DbError::from(e).into()
}
}
impl From<rusqlite::Error> for TxOpError {
fn from(e: rusqlite::Error) -> TxOpError {
TxOpError(e.into())
DbError::from(e).into()
}
}
@@ -62,7 +74,7 @@ pub struct SqliteDb {
}
impl SqliteDb {
pub fn open(manager: SqliteConnectionManager, sync_mode: bool) -> Result<Db> {
pub fn open(manager: SqliteConnectionManager, sync_mode: bool) -> DbResult<Db> {
let manager = manager.with_init(move |db| {
db.pragma_update(None, "journal_mode", "WAL")?;
if sync_mode {
@@ -82,16 +94,16 @@ impl SqliteDb {
}
impl SqliteDb {
fn get_tree(&self, i: usize) -> Result<Arc<str>> {
fn get_tree(&self, i: usize) -> DbResult<Arc<str>> {
self.trees
.read()
.unwrap()
.get(i)
.cloned()
.ok_or_else(|| Error("invalid tree id".into()))
.ok_or_else(|| DbError("invalid tree id".into()))
}
fn internal_get(&self, db: &Connection, tree: &str, key: &[u8]) -> Result<Option<Value>> {
fn internal_get(&self, db: &Connection, tree: &str, key: &[u8]) -> DbResult<Option<Value>> {
let mut stmt = db.prepare(&format!("SELECT v FROM {} WHERE k = ?1", tree))?;
let mut res_iter = stmt.query([key])?;
match res_iter.next()? {
@@ -106,7 +118,7 @@ impl IDb for SqliteDb {
format!("sqlite3 v{} (using rusqlite crate)", rusqlite::version())
}
fn open_tree(&self, name: &str) -> Result<usize> {
fn open_tree(&self, name: &str) -> DbResult<usize> {
let name = format!("tree_{}", name.replace(':', "_COLON_"));
let mut trees = self.trees.write().unwrap();
@@ -133,7 +145,7 @@ impl IDb for SqliteDb {
}
}
fn list_trees(&self) -> Result<Vec<String>> {
fn list_trees(&self) -> DbResult<Vec<String>> {
let mut trees = vec![];
let db = self.db.get()?;
@@ -150,13 +162,13 @@ impl IDb for SqliteDb {
Ok(trees)
}
fn snapshot(&self, base_path: &Path) -> Result<()> {
fn snapshot(&self, base_path: &Path) -> DbResult<()> {
std::fs::create_dir_all(base_path)?;
let path = Engine::Sqlite
.db_path(base_path)
.into_os_string()
.into_string()
.map_err(|_| Error("invalid sqlite path string".into()))?;
.map_err(|_| DbError("invalid sqlite path string".into()))?;
info!("Start sqlite VACUUM INTO `{}`", path);
self.db.get()?.execute("VACUUM INTO ?1", params![path])?;
@@ -167,12 +179,12 @@ impl IDb for SqliteDb {
// ----
fn get(&self, tree: usize, key: &[u8]) -> Result<Option<Value>> {
fn get(&self, tree: usize, key: &[u8]) -> DbResult<Option<Value>> {
let tree = self.get_tree(tree)?;
self.internal_get(&self.db.get()?, &tree, key)
}
fn approximate_len(&self, tree: usize) -> Result<usize> {
fn approximate_len(&self, tree: usize) -> DbResult<usize> {
let tree = self.get_tree(tree)?;
let db = self.db.get()?;
@@ -184,11 +196,11 @@ impl IDb for SqliteDb {
}
}
fn is_empty(&self, tree: usize) -> Result<bool> {
fn is_empty(&self, tree: usize) -> DbResult<bool> {
Ok(self.approximate_len(tree)? == 0)
}
fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> Result<()> {
fn insert(&self, tree: usize, key: &[u8], value: &[u8]) -> DbResult<()> {
let tree = self.get_tree(tree)?;
let db = self.db.get()?;
let lock = self.write_lock.lock();
@@ -206,7 +218,7 @@ impl IDb for SqliteDb {
Ok(())
}
fn remove(&self, tree: usize, key: &[u8]) -> Result<()> {
fn remove(&self, tree: usize, key: &[u8]) -> DbResult<()> {
let tree = self.get_tree(tree)?;
let db = self.db.get()?;
let lock = self.write_lock.lock();
@@ -217,7 +229,7 @@ impl IDb for SqliteDb {
Ok(())
}
fn clear(&self, tree: usize) -> Result<()> {
fn clear(&self, tree: usize) -> DbResult<()> {
let tree = self.get_tree(tree)?;
let db = self.db.get()?;
let lock = self.write_lock.lock();
@@ -228,13 +240,13 @@ impl IDb for SqliteDb {
Ok(())
}
fn iter(&self, tree: usize) -> Result<ValueIter<'_>> {
fn iter(&self, tree: usize) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree)?;
let sql = format!("SELECT k, v FROM {} ORDER BY k ASC", tree);
DbValueIterator::make(self.db.get()?, &sql, [])
}
fn iter_rev(&self, tree: usize) -> Result<ValueIter<'_>> {
fn iter_rev(&self, tree: usize) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree)?;
let sql = format!("SELECT k, v FROM {} ORDER BY k DESC", tree);
DbValueIterator::make(self.db.get()?, &sql, [])
@@ -245,7 +257,7 @@ impl IDb for SqliteDb {
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> Result<ValueIter<'_>> {
) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree)?;
let (bounds_sql, params) = bounds_sql(low, high);
@@ -263,7 +275,7 @@ impl IDb for SqliteDb {
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> Result<ValueIter<'_>> {
) -> DbResult<ValueIter<'_>> {
let tree = self.get_tree(tree)?;
let (bounds_sql, params) = bounds_sql(low, high);
@@ -300,9 +312,9 @@ impl IDb for SqliteDb {
}
TxFnResult::DbErr => {
tx.tx.rollback().map_err(Error::from).map_err(TxError::Db)?;
Err(TxError::Db(Error(
"(this message will be discarded)".into(),
)))
Err(TxError::Db(
DbError("(this message will be discarded)".into()).into(),
))
}
};
@@ -320,15 +332,15 @@ struct SqliteTx<'a> {
}
impl<'a> SqliteTx<'a> {
fn get_tree(&self, i: usize) -> TxOpResult<&'_ str> {
fn get_tree(&self, i: usize) -> DbResult<&'_ str> {
self.trees.get(i).map(Arc::as_ref).ok_or_else(|| {
TxOpError(Error(
DbError(
"invalid tree id (it might have been opened after the transaction started)".into(),
))
)
})
}
fn internal_get(&self, tree: &str, key: &[u8]) -> TxOpResult<Option<Value>> {
fn internal_get(&self, tree: &str, key: &[u8]) -> DbResult<Option<Value>> {
let mut stmt = self
.tx
.prepare(&format!("SELECT v FROM {} WHERE k = ?1", tree))?;
@@ -341,11 +353,11 @@ impl<'a> SqliteTx<'a> {
}
impl<'a> ITx for SqliteTx<'a> {
fn get(&self, tree: usize, key: &[u8]) -> TxOpResult<Option<Value>> {
fn get(&self, tree: usize, key: &[u8]) -> DbResult<Option<Value>> {
let tree = self.get_tree(tree)?;
self.internal_get(tree, key)
}
fn len(&self, tree: usize) -> TxOpResult<usize> {
fn len(&self, tree: usize) -> DbResult<usize> {
let tree = self.get_tree(tree)?;
let mut stmt = self.tx.prepare(&format!("SELECT COUNT(*) FROM {}", tree))?;
let mut res_iter = stmt.query([])?;
@@ -355,30 +367,30 @@ impl<'a> ITx for SqliteTx<'a> {
}
}
fn insert(&mut self, tree: usize, key: &[u8], value: &[u8]) -> TxOpResult<()> {
fn insert(&mut self, tree: usize, key: &[u8], value: &[u8]) -> DbResult<()> {
let tree = self.get_tree(tree)?;
let sql = format!("INSERT OR REPLACE INTO {} (k, v) VALUES (?1, ?2)", tree);
self.tx.execute(&sql, params![key, value])?;
Ok(())
}
fn remove(&mut self, tree: usize, key: &[u8]) -> TxOpResult<()> {
fn remove(&mut self, tree: usize, key: &[u8]) -> DbResult<()> {
let tree = self.get_tree(tree)?;
self.tx
.execute(&format!("DELETE FROM {} WHERE k = ?1", tree), params![key])?;
Ok(())
}
fn clear(&mut self, tree: usize) -> TxOpResult<()> {
fn clear(&mut self, tree: usize) -> DbResult<()> {
let tree = self.get_tree(tree)?;
self.tx.execute(&format!("DELETE FROM {}", tree), [])?;
Ok(())
}
fn iter(&self, tree: usize) -> TxOpResult<TxValueIter<'_>> {
fn iter(&self, tree: usize) -> DbResult<TxValueIter<'_>> {
let tree = self.get_tree(tree)?;
let sql = format!("SELECT k, v FROM {} ORDER BY k ASC", tree);
TxValueIterator::make(self, &sql, [])
}
fn iter_rev(&self, tree: usize) -> TxOpResult<TxValueIter<'_>> {
fn iter_rev(&self, tree: usize) -> DbResult<TxValueIter<'_>> {
let tree = self.get_tree(tree)?;
let sql = format!("SELECT k, v FROM {} ORDER BY k DESC", tree);
TxValueIterator::make(self, &sql, [])
@@ -389,7 +401,7 @@ impl<'a> ITx for SqliteTx<'a> {
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> TxOpResult<TxValueIter<'_>> {
) -> DbResult<TxValueIter<'_>> {
let tree = self.get_tree(tree)?;
let (bounds_sql, params) = bounds_sql(low, high);
@@ -407,7 +419,7 @@ impl<'a> ITx for SqliteTx<'a> {
tree: usize,
low: Bound<&'r [u8]>,
high: Bound<&'r [u8]>,
) -> TxOpResult<TxValueIter<'_>> {
) -> DbResult<TxValueIter<'_>> {
let tree = self.get_tree(tree)?;
let (bounds_sql, params) = bounds_sql(low, high);
@@ -439,7 +451,7 @@ impl DbValueIterator {
db: Connection,
sql: &str,
args: P,
) -> Result<ValueIter<'res>> {
) -> DbResult<ValueIter<'res>> {
let res = DbValueIterator {
db,
stmt: None,
@@ -484,7 +496,7 @@ impl Drop for DbValueIterator {
struct DbValueIteratorPin(Pin<Box<DbValueIterator>>);
impl Iterator for DbValueIteratorPin {
type Item = Result<(Value, Value)>;
type Item = DbResult<(Value, Value)>;
fn next(&mut self) -> Option<Self::Item> {
let mut_ref = Pin::as_mut(&mut self.0);
@@ -509,7 +521,7 @@ impl<'a> TxValueIterator<'a> {
tx: &'a SqliteTx<'a>,
sql: &str,
args: P,
) -> TxOpResult<TxValueIter<'a>> {
) -> DbResult<TxValueIter<'a>> {
let stmt = tx.tx.prepare(sql)?;
let res = TxValueIterator {
stmt,
@@ -543,7 +555,7 @@ impl<'a> Drop for TxValueIterator<'a> {
struct TxValueIteratorPin<'a>(Pin<Box<TxValueIterator<'a>>>);
impl<'a> Iterator for TxValueIteratorPin<'a> {
type Item = TxOpResult<(Value, Value)>;
type Item = DbResult<(Value, Value)>;
fn next(&mut self) -> Option<Self::Item> {
let mut_ref = Pin::as_mut(&mut self.0);
+241
View File
@@ -0,0 +1,241 @@
use std::marker::PhantomData;
use std::ops::{Bound, RangeBounds};
// Todo: some parts of this code, notably around ranges are never used but are here to prepare
// the migration of the parts of the codebase that still use untyped trees. At one point, this
// migration should be done or these functions deleted.
use super::{
DbResult, DecodeError, Error, Result, Transaction, Tree, TxOpError, TxOpResult, TxValueIter,
ValueIter,
};
pub use super::Db;
pub trait DbBytes: Sized {
fn encode(&self) -> Vec<u8>;
fn decode(bytes: &[u8]) -> std::result::Result<Self, DecodeError>;
}
/// Subtrait of [`DbBytes`] for types used as tree keys with operations where order matters
/// (`get_gt`, range, etc...).
///
/// Implementors must guarantee that the byte encoding is order-preserving:
/// for any `a, b: Self`, `a.cmp(&b) == a.encode().cmp(&b.encode())`.
pub trait DbOrdKey: DbBytes + Ord {}
#[derive(Clone)]
pub struct TypedTree<K, V> {
inner: Tree,
_phantom: PhantomData<[(K, V)]>,
}
impl<K: DbBytes, V: DbBytes> TypedTree<K, V> {
pub fn new(tree: Tree) -> Self {
Self {
inner: tree,
_phantom: PhantomData,
}
}
pub fn db(&self) -> Db {
self.inner.db()
}
pub fn untyped(&self) -> &Tree {
&self.inner
}
pub fn get(&self, key: &K) -> Result<Option<V>> {
self.inner
.get(key.encode())?
.map(|v| V::decode(&v).map_err(Error::from))
.transpose()
}
pub fn approximate_len(&self) -> DbResult<usize> {
self.inner.approximate_len()
}
pub fn is_empty(&self) -> DbResult<bool> {
self.inner.is_empty()
}
pub fn insert(&self, key: &K, value: &V) -> DbResult<()> {
self.inner.insert(key.encode(), value.encode())
}
pub fn remove(&self, key: &K) -> DbResult<()> {
self.inner.remove(key.encode())
}
pub fn clear(&self) -> DbResult<()> {
self.inner.clear()
}
pub fn tx_get(&self, tx: &Transaction<'_>, key: &K) -> TxOpResult<Option<V>> {
tx.get(&self.inner, key.encode())?
.map(|v| V::decode(&v).map_err(TxOpError::from))
.transpose()
}
pub fn tx_insert(&self, tx: &mut Transaction<'_>, key: &K, value: &V) -> TxOpResult<()> {
tx.insert(&self.inner, key.encode(), value.encode())
}
pub fn tx_remove(&self, tx: &mut Transaction<'_>, key: &K) -> TxOpResult<()> {
tx.remove(&self.inner, key.encode())
}
pub fn tx_clear(&self, tx: &mut Transaction<'_>) -> TxOpResult<()> {
tx.clear(&self.inner)
}
}
impl<K: DbOrdKey, V: DbBytes> TypedTree<K, V> {
pub fn first(&self) -> Result<Option<(K, V)>> {
self.iter()?.next().transpose()
}
pub fn get_gt(&self, from: &K) -> Result<Option<(K, V)>> {
self.inner
.get_gt(from.encode())?
.map(|(k, v)| {
Ok((
K::decode(&k).map_err(Error::from)?,
V::decode(&v).map_err(Error::from)?,
))
})
.transpose()
}
pub fn iter(&self) -> Result<TypedIter<'_, K, V>> {
Ok(TypedIter::new(self.inner.iter()?))
}
pub fn iter_rev(&self) -> Result<TypedIter<'_, K, V>> {
Ok(TypedIter::new(self.inner.iter_rev()?))
}
pub fn range<R: RangeBounds<K>>(&self, range: R) -> Result<TypedIter<'_, K, V>> {
Ok(TypedIter::new(self.inner.range(encode_range(range))?))
}
pub fn range_rev<R: RangeBounds<K>>(&self, range: R) -> Result<TypedIter<'_, K, V>> {
Ok(TypedIter::new(self.inner.range_rev(encode_range(range))?))
}
pub fn tx_iter<'t>(&self, tx: &'t Transaction<'_>) -> TxOpResult<TypedTxIter<'t, K, V>> {
Ok(TypedTxIter::new(tx.iter(&self.inner)?))
}
pub fn tx_iter_rev<'t>(&self, tx: &'t Transaction<'_>) -> TxOpResult<TypedTxIter<'t, K, V>> {
Ok(TypedTxIter::new(tx.iter_rev(&self.inner)?))
}
pub fn tx_range<'t, R: RangeBounds<K>>(
&self,
tx: &'t Transaction<'_>,
range: R,
) -> TxOpResult<TypedTxIter<'t, K, V>> {
Ok(TypedTxIter::new(
tx.range(&self.inner, encode_range(range))?,
))
}
pub fn tx_range_rev<'t, R: RangeBounds<K>>(
&self,
tx: &'t Transaction<'_>,
range: R,
) -> TxOpResult<TypedTxIter<'t, K, V>> {
Ok(TypedTxIter::new(
tx.range_rev(&self.inner, encode_range(range))?,
))
}
}
impl<K: DbBytes, V: DbBytes> From<Tree> for TypedTree<K, V> {
fn from(tree: Tree) -> Self {
Self::new(tree)
}
}
impl Db {
pub fn open_typed_tree<K: DbBytes, V: DbBytes, S: AsRef<str>>(
&self,
name: S,
) -> DbResult<TypedTree<K, V>> {
Ok(TypedTree::new(self.open_tree(name)?))
}
}
pub struct TypedIter<'a, K, V> {
inner: ValueIter<'a>,
_phantom: PhantomData<(K, V)>,
}
impl<'a, K, V> TypedIter<'a, K, V> {
fn new(inner: ValueIter<'a>) -> Self {
Self {
inner,
_phantom: PhantomData,
}
}
}
impl<K: DbOrdKey, V: DbBytes> Iterator for TypedIter<'_, K, V> {
type Item = Result<(K, V)>;
fn next(&mut self) -> Option<Self::Item> {
self.inner.next().map(|res| {
let (k, v) = res?;
Ok((
K::decode(&k).map_err(Error::from)?,
V::decode(&v).map_err(Error::from)?,
))
})
}
}
pub struct TypedTxIter<'a, K, V> {
inner: TxValueIter<'a>,
_phantom: PhantomData<(K, V)>,
}
impl<'a, K, V> TypedTxIter<'a, K, V> {
fn new(inner: TxValueIter<'a>) -> Self {
Self {
inner,
_phantom: PhantomData,
}
}
}
impl<K: DbOrdKey, V: DbBytes> Iterator for TypedTxIter<'_, K, V> {
type Item = TxOpResult<(K, V)>;
fn next(&mut self) -> Option<Self::Item> {
self.inner.next().map(|res| {
let (k, v) = res?;
Ok((
K::decode(&k).map_err(TxOpError::from)?,
V::decode(&v).map_err(TxOpError::from)?,
))
})
}
}
fn encode_range<K: DbOrdKey, R: RangeBounds<K>>(range: R) -> (Bound<Vec<u8>>, Bound<Vec<u8>>) {
(
encode_bound(range.start_bound()),
encode_bound(range.end_bound()),
)
}
fn encode_bound<K: DbOrdKey>(bound: Bound<&K>) -> Bound<Vec<u8>> {
match bound {
Bound::Included(k) => Bound::Included(k.encode()),
Bound::Excluded(k) => Bound::Excluded(k.encode()),
Bound::Unbounded => Bound::Unbounded,
}
}
+1 -1
View File
@@ -49,7 +49,7 @@ pub struct OpenLmdbOpt {
pub(crate) fn do_conversion(args: ConvertDbOpt) -> Result<()> {
if args.input_engine == args.output_engine {
return Err(Error("input and output database engine must differ".into()));
return Err(DbError("input and output database engine must differ".into()).into());
}
let opt = OpenOpt {
+1
View File
@@ -163,6 +163,7 @@ impl<T: CountedItem> TableSchema for CounterTable<T> {
pub struct IndexCounter<T: CountedItem> {
this_node: Uuid,
// TODO: migrate to TypedTree
local_counter: db::Tree,
pub table: Arc<Table<CounterTable<T>, TableShardedReplication>>,
}
+1
View File
@@ -84,6 +84,7 @@ pub struct K2VRpcHandler {
// Using a mutex on the local_timestamp_tree is not strictly necessary,
// but it helps to not try to do several inserts at the same time,
// which would create transaction conflicts and force many useless retries.
// TODO: migrate to TypedTree
local_timestamp_tree: Mutex<db::Tree>,
endpoint: Arc<Endpoint<K2VRpc, Self>>,
+5
View File
@@ -26,15 +26,20 @@ pub struct TableData<F: TableSchema, R: TableReplication> {
pub instance: F,
pub replication: R,
// TODO: migrate to TypedTree
pub store: db::Tree,
// TODO: migrate to TypedTree
pub(crate) merkle_tree: db::Tree,
// TODO: migrate to TypedTree
pub(crate) merkle_todo: db::Tree,
pub(crate) merkle_todo_notify: Notify,
// TODO: migrate to TypedTree
pub(crate) insert_queue: db::Tree,
pub(crate) insert_queue_notify: Arc<Notify>,
// TODO: migrate to TypedTree
pub(crate) gc_todo: db::Tree,
pub(crate) metrics: TableMetrics,
+15
View File
@@ -155,6 +155,21 @@ pub fn gen_uuid() -> Uuid {
rand::rng().random::<[u8; 32]>().into()
}
impl garage_db::DbBytes for FixedBytes32 {
fn encode(&self) -> Vec<u8> {
self.0.into()
}
fn decode(bytes: &[u8]) -> std::result::Result<Self, garage_db::DecodeError> {
Self::try_from(bytes).ok_or_else(|| {
garage_db::DecodeError(
format!("invalid hash: expected 32 bytes, got {}", bytes.len()).into(),
)
})
}
}
impl garage_db::DbOrdKey for FixedBytes32 {}
#[cfg(test)]
mod test {
use super::*;
+12
View File
@@ -79,6 +79,18 @@ impl Error {
}
}
impl From<garage_db::DbError> for Error {
fn from(e: garage_db::DbError) -> Error {
Error::Db(e.into())
}
}
impl From<garage_db::DecodeError> for Error {
fn from(e: garage_db::DecodeError) -> Error {
Error::Db(e.into())
}
}
impl From<garage_db::TxError<Error>> for Error {
fn from(e: garage_db::TxError<Error>) -> Error {
match e {