mirror of
https://github.com/aljazceru/turso.git
synced 2025-12-29 14:04:22 +01:00
455 lines
16 KiB
Rust
455 lines
16 KiB
Rust
use std::sync::Arc;
|
|
|
|
use crate::{
|
|
database_sync_operations::{
|
|
checkpoint_wal_file, connect, connect_untracked, db_bootstrap, push_logical_changes,
|
|
reset_wal_file, transfer_logical_changes, transfer_physical_changes, wait_full_body,
|
|
wal_pull, wal_push, WalPullResult,
|
|
},
|
|
database_tape::DatabaseTape,
|
|
errors::Error,
|
|
io_operations::IoOperations,
|
|
protocol_io::ProtocolIO,
|
|
types::{Coro, DatabaseMetadata},
|
|
wal_session::WalSession,
|
|
Result,
|
|
};
|
|
|
|
#[derive(Debug, Clone)]
|
|
pub struct DatabaseSyncEngineOpts {
|
|
pub client_name: String,
|
|
pub wal_pull_batch_size: u64,
|
|
}
|
|
|
|
pub struct DatabaseSyncEngine<P: ProtocolIO> {
|
|
io: Arc<dyn turso_core::IO>,
|
|
protocol: Arc<P>,
|
|
draft_tape: DatabaseTape,
|
|
draft_path: String,
|
|
synced_path: String,
|
|
meta_path: String,
|
|
opts: DatabaseSyncEngineOpts,
|
|
meta: Option<DatabaseMetadata>,
|
|
// we remember information if Synced DB is dirty - which will make Database to reset it in case of any sync attempt
|
|
// this bit is set to false when we properly reset Synced DB
|
|
// this bit is set to true when we transfer changes from Draft to Synced or on initialization
|
|
synced_is_dirty: bool,
|
|
}
|
|
|
|
async fn update_meta<IO: ProtocolIO>(
|
|
coro: &Coro,
|
|
io: &IO,
|
|
meta_path: &str,
|
|
orig: &mut Option<DatabaseMetadata>,
|
|
update: impl FnOnce(&mut DatabaseMetadata),
|
|
) -> Result<()> {
|
|
let mut meta = orig.as_ref().unwrap().clone();
|
|
update(&mut meta);
|
|
tracing::info!("update_meta: {meta:?}");
|
|
let completion = io.full_write(meta_path, meta.dump()?)?;
|
|
// todo: what happen if we will actually update the metadata on disk but fail and so in memory state will not be updated
|
|
wait_full_body(coro, &completion).await?;
|
|
*orig = Some(meta);
|
|
Ok(())
|
|
}
|
|
|
|
async fn set_meta<IO: ProtocolIO>(
|
|
coro: &Coro,
|
|
io: &IO,
|
|
meta_path: &str,
|
|
orig: &mut Option<DatabaseMetadata>,
|
|
meta: DatabaseMetadata,
|
|
) -> Result<()> {
|
|
tracing::info!("set_meta: {meta:?}");
|
|
let completion = io.full_write(meta_path, meta.dump()?)?;
|
|
// todo: what happen if we will actually update the metadata on disk but fail and so in memory state will not be updated
|
|
wait_full_body(coro, &completion).await?;
|
|
*orig = Some(meta);
|
|
Ok(())
|
|
}
|
|
|
|
impl<C: ProtocolIO> DatabaseSyncEngine<C> {
|
|
/// Creates new instance of SyncEngine and initialize it immediately if no consistent local data exists
|
|
pub async fn new(
|
|
coro: &Coro,
|
|
io: Arc<dyn turso_core::IO>,
|
|
protocol: Arc<C>,
|
|
path: &str,
|
|
opts: DatabaseSyncEngineOpts,
|
|
) -> Result<Self> {
|
|
let draft_path = format!("{path}-draft");
|
|
let draft_tape = io.open_tape(&draft_path, true)?;
|
|
let mut db = Self {
|
|
io,
|
|
protocol,
|
|
draft_tape,
|
|
draft_path,
|
|
synced_path: format!("{path}-synced"),
|
|
meta_path: format!("{path}-info"),
|
|
opts,
|
|
meta: None,
|
|
synced_is_dirty: true,
|
|
};
|
|
db.init(coro).await?;
|
|
Ok(db)
|
|
}
|
|
|
|
/// Create database connection and appropriately configure it before use
|
|
pub async fn connect(&self, coro: &Coro) -> Result<Arc<turso_core::Connection>> {
|
|
connect(coro, &self.draft_tape).await
|
|
}
|
|
|
|
/// Sync all new changes from remote DB and apply them locally
|
|
/// This method will **not** send local changed to the remote
|
|
/// This method will block writes for the period of pull
|
|
pub async fn pull(&mut self, coro: &Coro) -> Result<()> {
|
|
tracing::info!(
|
|
"pull: draft={}, synced={}",
|
|
self.draft_path,
|
|
self.synced_path
|
|
);
|
|
|
|
// reset Synced DB if it wasn't properly cleaned-up on previous "sync-method" attempt
|
|
self.reset_synced_if_dirty(coro).await?;
|
|
|
|
loop {
|
|
// update Synced DB with fresh changes from remote
|
|
let pull_result = self.pull_synced_from_remote(coro).await?;
|
|
|
|
{
|
|
// we will "replay" Synced WAL to the Draft WAL later without pushing it to the remote
|
|
// so, we pass 'capture: true' as we need to preserve all changes for future push of WAL
|
|
let synced = self.io.open_tape(&self.synced_path, true)?;
|
|
|
|
// we will start wal write session for Draft DB in order to hold write lock during transfer of changes
|
|
let mut draft_session = WalSession::new(connect(coro, &self.draft_tape).await?);
|
|
draft_session.begin()?;
|
|
|
|
// mark Synced as dirty as we will start transfer of logical changes there and if we will fail in the middle - we will need to cleanup Synced db
|
|
self.synced_is_dirty = true;
|
|
|
|
// transfer logical changes to the Synced DB in order to later execute physical "rebase" operation
|
|
let client_id = &self.meta().client_unique_id;
|
|
transfer_logical_changes(coro, &self.draft_tape, &synced, client_id, true).await?;
|
|
|
|
// now we are ready to do the rebase: let's transfer physical changes from Synced to Draft
|
|
let synced_wal_watermark = self.meta().synced_wal_match_watermark;
|
|
let synced_sync_watermark = self.meta().synced_frame_no.expect(
|
|
"synced_frame_no must be set as we call pull_synced_from_remote before that",
|
|
);
|
|
let draft_wal_watermark = self.meta().draft_wal_match_watermark;
|
|
let draft_sync_watermark = transfer_physical_changes(
|
|
coro,
|
|
&synced,
|
|
draft_session,
|
|
synced_wal_watermark,
|
|
synced_sync_watermark,
|
|
draft_wal_watermark,
|
|
)
|
|
.await?;
|
|
update_meta(
|
|
coro,
|
|
self.protocol.as_ref(),
|
|
&self.meta_path,
|
|
&mut self.meta,
|
|
|m| {
|
|
m.draft_wal_match_watermark = draft_sync_watermark;
|
|
m.synced_wal_match_watermark = synced_sync_watermark;
|
|
},
|
|
)
|
|
.await?;
|
|
}
|
|
|
|
// Synced DB is 100% dirty now - let's reset it
|
|
assert!(self.synced_is_dirty);
|
|
self.reset_synced_if_dirty(coro).await?;
|
|
|
|
let WalPullResult::NeedCheckpoint = pull_result else {
|
|
break;
|
|
};
|
|
tracing::info!(
|
|
"ready to checkpoint synced db file at {:?}, generation={}",
|
|
self.synced_path,
|
|
self.meta().synced_generation
|
|
);
|
|
{
|
|
let synced = self.io.open_tape(&self.synced_path, false)?;
|
|
checkpoint_wal_file(coro, &connect_untracked(&synced)?).await?;
|
|
update_meta(
|
|
coro,
|
|
self.protocol.as_ref(),
|
|
&self.meta_path,
|
|
&mut self.meta,
|
|
|m| {
|
|
m.synced_generation += 1;
|
|
m.synced_frame_no = Some(0);
|
|
m.synced_wal_match_watermark = 0;
|
|
},
|
|
)
|
|
.await?;
|
|
}
|
|
}
|
|
|
|
Ok(())
|
|
}
|
|
|
|
/// Sync local changes to remote DB
|
|
/// This method will **not** pull remote changes to the local DB
|
|
/// This method will **not** block writes for the period of sync
|
|
pub async fn push(&mut self, coro: &Coro) -> Result<()> {
|
|
tracing::info!(
|
|
"push: draft={}, synced={}",
|
|
self.draft_path,
|
|
self.synced_path
|
|
);
|
|
|
|
// reset Synced DB if it wasn't properly cleaned-up on previous "sync-method" attempt
|
|
self.reset_synced_if_dirty(coro).await?;
|
|
|
|
// update Synced DB with fresh changes from remote in order to avoid WAL frame conflicts
|
|
self.pull_synced_from_remote(coro).await?;
|
|
|
|
// we will push Synced WAL to the remote
|
|
// so, we pass 'capture: false' as we don't need to preserve changes made to Synced WAL in turso_cdc
|
|
let synced = self.io.open_tape(&self.synced_path, false)?;
|
|
|
|
self.synced_is_dirty = true;
|
|
let client_id = &self.meta().client_unique_id;
|
|
push_logical_changes(
|
|
coro,
|
|
self.protocol.as_ref(),
|
|
&self.draft_tape,
|
|
&synced,
|
|
client_id,
|
|
)
|
|
.await?;
|
|
|
|
self.reset_synced_if_dirty(coro).await?;
|
|
|
|
Ok(())
|
|
}
|
|
|
|
/// Sync local changes to remote DB and bring new changes from remote to local
|
|
/// This method will block writes for the period of sync
|
|
pub async fn sync(&mut self, coro: &Coro) -> Result<()> {
|
|
// todo(sivukhin): this is bit suboptimal as both 'push' and 'pull' will call pull_synced_from_remote
|
|
// but for now - keep it simple
|
|
self.push(coro).await?;
|
|
self.pull(coro).await?;
|
|
Ok(())
|
|
}
|
|
|
|
async fn init(&mut self, coro: &Coro) -> Result<()> {
|
|
tracing::info!(
|
|
"initialize sync engine: draft={}, synced={}, opts={:?}",
|
|
self.draft_path,
|
|
self.synced_path,
|
|
self.opts,
|
|
);
|
|
|
|
let completion = self.protocol.full_read(&self.meta_path)?;
|
|
let data = wait_full_body(coro, &completion).await?;
|
|
let meta = if data.is_empty() {
|
|
None
|
|
} else {
|
|
Some(DatabaseMetadata::load(&data)?)
|
|
};
|
|
|
|
match meta {
|
|
Some(meta) => {
|
|
self.meta = Some(meta);
|
|
}
|
|
None => {
|
|
let meta = self.bootstrap_db_files(coro).await?;
|
|
tracing::info!("write meta after successful bootstrap: meta={meta:?}");
|
|
set_meta(
|
|
coro,
|
|
self.protocol.as_ref(),
|
|
&self.meta_path,
|
|
&mut self.meta,
|
|
meta,
|
|
)
|
|
.await?;
|
|
}
|
|
};
|
|
|
|
let draft_exists = self.io.try_open(&self.draft_path)?.is_some();
|
|
let synced_exists = self.io.try_open(&self.synced_path)?.is_some();
|
|
if !draft_exists || !synced_exists {
|
|
let error = "Draft or Synced files doesn't exists, but metadata is".to_string();
|
|
return Err(Error::DatabaseSyncEngineError(error));
|
|
}
|
|
|
|
if self.meta().synced_frame_no.is_none() {
|
|
// sync WAL from the remote in case of bootstrap - all subsequent initializations will be fast
|
|
self.pull(coro).await?;
|
|
}
|
|
Ok(())
|
|
}
|
|
|
|
async fn pull_synced_from_remote(&mut self, coro: &Coro) -> Result<WalPullResult> {
|
|
tracing::info!(
|
|
"pull_synced_from_remote: draft={:?}, synced={:?}",
|
|
self.draft_path,
|
|
self.synced_path,
|
|
);
|
|
let synced = self.io.open_tape(&self.synced_path, false)?;
|
|
let synced_conn = connect(coro, &synced).await?;
|
|
let mut wal = WalSession::new(synced_conn);
|
|
|
|
let generation = self.meta().synced_generation;
|
|
let mut start_frame = self.meta().synced_frame_no.unwrap_or(0) + 1;
|
|
loop {
|
|
let end_frame = start_frame + self.opts.wal_pull_batch_size;
|
|
let update = async |coro, frame_no| {
|
|
update_meta(
|
|
coro,
|
|
self.protocol.as_ref(),
|
|
&self.meta_path,
|
|
&mut self.meta,
|
|
|m| m.synced_frame_no = Some(frame_no),
|
|
)
|
|
.await
|
|
};
|
|
match wal_pull(
|
|
coro,
|
|
self.protocol.as_ref(),
|
|
&mut wal,
|
|
generation,
|
|
start_frame,
|
|
end_frame,
|
|
update,
|
|
)
|
|
.await?
|
|
{
|
|
WalPullResult::Done => return Ok(WalPullResult::Done),
|
|
WalPullResult::NeedCheckpoint => return Ok(WalPullResult::NeedCheckpoint),
|
|
WalPullResult::PullMore => {
|
|
start_frame = end_frame;
|
|
continue;
|
|
}
|
|
}
|
|
}
|
|
}
|
|
|
|
#[allow(dead_code)]
|
|
async fn push_synced_to_remote(&mut self, coro: &Coro) -> Result<()> {
|
|
tracing::info!(
|
|
"push_synced_to_remote: draft={}, synced={}, id={}",
|
|
self.draft_path,
|
|
self.synced_path,
|
|
self.meta().client_unique_id
|
|
);
|
|
let synced = self.io.open_tape(&self.synced_path, false)?;
|
|
let synced_conn = connect(coro, &synced).await?;
|
|
|
|
let mut wal = WalSession::new(synced_conn);
|
|
wal.begin()?;
|
|
|
|
// todo(sivukhin): push frames in multiple batches
|
|
let generation = self.meta().synced_generation;
|
|
let start_frame = self.meta().synced_frame_no.unwrap_or(0) + 1;
|
|
let end_frame = wal.conn().wal_state()?.max_frame + 1;
|
|
match wal_push(
|
|
coro,
|
|
self.protocol.as_ref(),
|
|
&mut wal,
|
|
None,
|
|
generation,
|
|
start_frame,
|
|
end_frame,
|
|
)
|
|
.await
|
|
{
|
|
Ok(_) => {
|
|
update_meta(
|
|
coro,
|
|
self.protocol.as_ref(),
|
|
&self.meta_path,
|
|
&mut self.meta,
|
|
|m| m.synced_frame_no = Some(end_frame - 1),
|
|
)
|
|
.await?;
|
|
self.synced_is_dirty = false;
|
|
Ok(())
|
|
}
|
|
Err(err) => {
|
|
tracing::info!("push_synced_to_remote: failed: err={err}");
|
|
Err(err)
|
|
}
|
|
}
|
|
}
|
|
|
|
async fn bootstrap_db_files(&mut self, coro: &Coro) -> Result<DatabaseMetadata> {
|
|
assert!(
|
|
self.meta.is_none(),
|
|
"bootstrap_db_files must be called only when meta is not set"
|
|
);
|
|
tracing::info!(
|
|
"bootstrap_db_files: draft={}, synced={}",
|
|
self.draft_path,
|
|
self.synced_path,
|
|
);
|
|
|
|
let start_time = std::time::Instant::now();
|
|
// cleanup all files left from previous attempt to bootstrap
|
|
// we shouldn't write any WAL files - but let's truncate them too for safety
|
|
if let Some(file) = self.io.try_open(&self.draft_path)? {
|
|
self.io.truncate(coro, file, 0).await?;
|
|
}
|
|
if let Some(file) = self.io.try_open(&self.synced_path)? {
|
|
self.io.truncate(coro, file, 0).await?;
|
|
}
|
|
if let Some(file) = self.io.try_open(&format!("{}-wal", self.draft_path))? {
|
|
self.io.truncate(coro, file, 0).await?;
|
|
}
|
|
if let Some(file) = self.io.try_open(&format!("{}-wal", self.synced_path))? {
|
|
self.io.truncate(coro, file, 0).await?;
|
|
}
|
|
|
|
let files = &[
|
|
self.io.create(&self.draft_path)?,
|
|
self.io.create(&self.synced_path)?,
|
|
];
|
|
let db_info = db_bootstrap(coro, self.protocol.as_ref(), files).await?;
|
|
|
|
let elapsed = std::time::Instant::now().duration_since(start_time);
|
|
tracing::info!(
|
|
"bootstrap_db_files: finished draft={:?}, synced={:?}: elapsed={:?}",
|
|
self.draft_path,
|
|
self.synced_path,
|
|
elapsed
|
|
);
|
|
|
|
Ok(DatabaseMetadata {
|
|
client_unique_id: format!("{}-{}", self.opts.client_name, uuid::Uuid::new_v4()),
|
|
synced_generation: db_info.current_generation,
|
|
synced_frame_no: None,
|
|
draft_wal_match_watermark: 0,
|
|
synced_wal_match_watermark: 0,
|
|
})
|
|
}
|
|
|
|
/// Reset WAL of Synced database which potentially can have some local changes
|
|
async fn reset_synced_if_dirty(&mut self, coro: &Coro) -> Result<()> {
|
|
tracing::info!(
|
|
"reset_synced: synced_path={:?}, synced_is_dirty={}",
|
|
self.synced_path,
|
|
self.synced_is_dirty
|
|
);
|
|
// if we know that Synced DB is not dirty - let's skip this phase completely
|
|
if !self.synced_is_dirty {
|
|
return Ok(());
|
|
}
|
|
if let Some(synced_wal) = self.io.try_open(&format!("{}-wal", self.synced_path))? {
|
|
reset_wal_file(coro, synced_wal, self.meta().synced_frame_no.unwrap_or(0)).await?;
|
|
}
|
|
self.synced_is_dirty = false;
|
|
Ok(())
|
|
}
|
|
|
|
fn meta(&self) -> &DatabaseMetadata {
|
|
self.meta.as_ref().expect("metadata must be set")
|
|
}
|
|
}
|