reifydb-transaction 0.4.10

// SPDX-License-Identifier: Apache-2.0
// Copyright (c) 2025 ReifyDB

// This file includes and modifies code from the skipdb project (https://github.com/al8n/skipdb),
// originally licensed under the Apache License, Version 2.0.
// Original copyright:
//   Copyright (c) 2024 Al Liu
//
// The original Apache License can be found at:
//   http://www.apache.org/licenses/LICENSE-2.0

use std::ops::RangeBounds;

use reifydb_core::{
	common::CommitVersion,
	encoded::{
		key::{EncodedKey, EncodedKeyRange},
		row::EncodedRow,
	},
	event::transaction::PostCommitEvent,
	interface::store::{MultiVersionBatch, MultiVersionCommit, MultiVersionContains, MultiVersionGet},
};
use reifydb_type::{
	Result,
	util::{cowvec::CowVec, hex},
};
use tracing::instrument;

use super::{MultiTransaction, TransactionManagerCommand, version::StandardVersionProvider};
use crate::{delta::optimize_deltas, multi::types::TransactionValue};

/// Snapshot of write transaction state for savepoint/restore.
pub struct WriteSavepoint {
	pub(crate) pending_writes: PendingWrites,
	pub(crate) count: u64,
	pub(crate) size: u64,
	pub(crate) duplicates: Vec<Pending>,
}

pub struct MultiWriteTransaction {
	engine: MultiTransaction,
	pub(crate) tm: TransactionManagerCommand<StandardVersionProvider>,
}

impl MultiWriteTransaction {
	#[instrument(name = "transaction::command::new", level = "debug", skip(engine))]
	pub fn new(engine: MultiTransaction) -> Result<Self> {
		let tm = engine.tm.write()?;
		Ok(Self {
			engine,
			tm,
		})
	}
}

impl MultiWriteTransaction {
	/// Snapshot pending writes for later restore.
	pub fn savepoint(&self) -> WriteSavepoint {
		WriteSavepoint {
			pending_writes: self.tm.pending_writes.clone(),
			count: self.tm.count,
			size: self.tm.size,
			duplicates: self.tm.duplicates.clone(),
		}
	}

	/// Restore pending writes from a savepoint.
	pub fn restore_savepoint(&mut self, sp: WriteSavepoint) {
		self.tm.pending_writes = sp.pending_writes;
		self.tm.count = sp.count;
		self.tm.size = sp.size;
		self.tm.duplicates = sp.duplicates;
	}
}

impl MultiWriteTransaction {
	#[instrument(name = "transaction::command::commit", level = "debug", skip(self), fields(pending_count = self.tm.pending_writes().len()))]
	pub fn commit(&mut self) -> Result<CommitVersion> {
		// For read-only transactions (no pending writes), skip conflict detection
		if self.tm.pending_writes().is_empty() {
			self.tm.discard();
			return Ok(CommitVersion(0));
		}

		// Use commit_pending to allocate the commit version via oracle BEFORE writing to storage
		// This ensures entries have the correct commit version
		let (commit_version, entries) = self.tm.commit_pending()?;

		if entries.is_empty() {
			self.tm.discard();
			return Ok(CommitVersion(0));
		}

		// Collect and optimize deltas for storage commit
		let mut raw_deltas = CowVec::with_capacity(entries.len());
		for pending in &entries {
			raw_deltas.push(pending.delta.clone());
		}
		let optimized = optimize_deltas(raw_deltas.iter().cloned());
		let deltas = CowVec::new(optimized);

		MultiVersionCommit::commit(&self.engine.store, deltas.clone(), commit_version)?;

		self.tm.discard();

		// Emit PostCommitEvent BEFORE marking the watermark as done.
		// The CDC poll actor uses done_until() as the safe upper bound for
		// fetching CDC entries from the store. If done_commit runs first,
		// there is a window where done_until >= V but V's CDC has not been
		// written to the store yet (because the CDC producer processes
		// PostCommitEvents asynchronously). A concurrent commit on another
		// thread could then produce a CDC entry at V+1 that the poll actor
		// sees, causing it to advance its checkpoint past V and permanently
		// skip V's CDC.
		self.engine.event_bus.emit(PostCommitEvent::new(deltas, commit_version));
		self.tm.oracle.done_commit(commit_version);

		Ok(commit_version)
	}
}

impl MultiWriteTransaction {
	pub fn version(&self) -> CommitVersion {
		self.tm.version()
	}

	pub fn pending_writes(&self) -> &PendingWrites {
		self.tm.pending_writes()
	}

	pub fn read_as_of_version_exclusive(&mut self, version: CommitVersion) {
		self.tm.read_as_of_version_exclusive(version);
	}

	pub fn read_as_of_version_inclusive(&mut self, version: CommitVersion) -> Result<()> {
		self.read_as_of_version_exclusive(CommitVersion(version.0 + 1));
		Ok(())
	}

	#[instrument(name = "transaction::command::rollback", level = "debug", skip(self))]
	pub fn rollback(&mut self) -> Result<()> {
		self.tm.rollback()
	}

	#[instrument(name = "transaction::command::contains_key", level = "trace", skip(self), fields(key_hex = %hex::display(key.as_ref())))]
	pub fn contains_key(&mut self, key: &EncodedKey) -> Result<bool> {
		let version = self.tm.version();
		match self.tm.contains_key(key)? {
			Some(true) => Ok(true),
			Some(false) => Ok(false),
			None => MultiVersionContains::contains(&self.engine.store, key, version),
		}
	}

	#[instrument(name = "transaction::command::get", level = "trace", skip(self), fields(key_hex = %hex::display(key.as_ref())))]
	pub fn get(&mut self, key: &EncodedKey) -> Result<Option<TransactionValue>> {
		let version = self.tm.version();
		match self.tm.get(key)? {
			Some(v) => {
				if v.row().is_some() {
					Ok(Some(v.into()))
				} else {
					Ok(None)
				}
			}
			None => Ok(MultiVersionGet::get(&self.engine.store, key, version)?.map(Into::into)),
		}
	}

	#[instrument(name = "transaction::command::set", level = "trace", skip(self, row), fields(key_hex = %hex::display(key.as_ref()), value_len = row.len()))]
	pub fn set(&mut self, key: &EncodedKey, row: EncodedRow) -> Result<()> {
		self.tm.set(key, row)
	}

	#[instrument(name = "transaction::command::unset", level = "trace", skip(self, row), fields(key_hex = %hex::display(key.as_ref()), value_len = row.len()))]
	pub fn unset(&mut self, key: &EncodedKey, row: EncodedRow) -> Result<()> {
		self.tm.unset(key, row)
	}

	#[instrument(name = "transaction::command::remove", level = "trace", skip(self), fields(key_hex = %hex::display(key.as_ref())))]
	pub fn remove(&mut self, key: &EncodedKey) -> Result<()> {
		self.tm.remove(key)
	}

	pub fn prefix(&mut self, prefix: &EncodedKey) -> Result<MultiVersionBatch> {
		let items: Vec<_> = self.range(EncodedKeyRange::prefix(prefix), 1024).collect::<Result<Vec<_>>>()?;
		Ok(MultiVersionBatch {
			items,
			has_more: false,
		})
	}

	pub fn prefix_rev(&mut self, prefix: &EncodedKey) -> Result<MultiVersionBatch> {
		let items: Vec<_> =
			self.range_rev(EncodedKeyRange::prefix(prefix), 1024).collect::<Result<Vec<_>>>()?;
		Ok(MultiVersionBatch {
			items,
			has_more: false,
		})
	}

	/// Create a streaming iterator for forward range queries, merging pending writes.
	///
	/// This properly handles high version density by scanning until batch_size
	/// unique logical keys are collected. The stream yields individual entries
	/// and maintains cursor state internally. Pending writes are merged with
	/// committed storage data.
	pub fn range(
		&mut self,
		range: EncodedKeyRange,
		batch_size: usize,
	) -> Box<dyn Iterator<Item = Result<MultiVersionRow>> + Send + '_> {
		let version = self.tm.version();
		let (mut marker, pw) = self.tm.marker_with_pending_writes();
		let start = range.start_bound();
		let end = range.end_bound();

		marker.mark_range(range.clone());

		// Collect pending writes in range as owned data
		let pending: Vec<(EncodedKey, Pending)> =
			pw.range((start, end)).map(|(k, v)| (k.clone(), v.clone())).collect();

		let storage_iter = self.engine.store.range(range, version, batch_size);

		Box::new(MergePendingIterator::new(pending, storage_iter, false))
	}

	/// Create a streaming iterator for reverse range queries, merging pending writes.
	///
	/// This properly handles high version density by scanning until batch_size
	/// unique logical keys are collected. The stream yields individual entries
	/// in reverse key order and maintains cursor state internally.
	pub fn range_rev(
		&mut self,
		range: EncodedKeyRange,
		batch_size: usize,
	) -> Box<dyn Iterator<Item = Result<MultiVersionRow>> + Send + '_> {
		let version = self.tm.version();
		let (mut marker, pw) = self.tm.marker_with_pending_writes();
		let start = range.start_bound();
		let end = range.end_bound();

		marker.mark_range(range.clone());

		// Collect pending writes in range as owned data (reversed)
		let pending: Vec<(EncodedKey, Pending)> =
			pw.range((start, end)).rev().map(|(k, v)| (k.clone(), v.clone())).collect();

		let storage_iter = self.engine.store.range_rev(range, version, batch_size);

		Box::new(MergePendingIterator::new(pending, storage_iter, true))
	}
}

use std::{cmp::Ordering, iter, vec};

use reifydb_core::interface::store::MultiVersionRow;

use crate::multi::{pending::PendingWrites, types::Pending};

/// Iterator that merges pending writes with storage iterator.
pub(crate) struct MergePendingIterator<I> {
	pending_iter: iter::Peekable<vec::IntoIter<(EncodedKey, Pending)>>,
	storage_iter: I,
	next_storage: Option<MultiVersionRow>,
	reverse: bool,
}

impl<I> MergePendingIterator<I>
where
	I: Iterator<Item = Result<MultiVersionRow>>,
{
	pub(crate) fn new(pending: Vec<(EncodedKey, Pending)>, storage_iter: I, reverse: bool) -> Self {
		Self {
			pending_iter: pending.into_iter().peekable(),
			storage_iter,
			next_storage: None,
			reverse,
		}
	}
}

impl<I> Iterator for MergePendingIterator<I>
where
	I: Iterator<Item = Result<MultiVersionRow>>,
{
	type Item = Result<MultiVersionRow>;

	fn next(&mut self) -> Option<Self::Item> {
		loop {
			// Fetch next storage item if needed
			if self.next_storage.is_none() {
				self.next_storage = match self.storage_iter.next() {
					Some(Ok(v)) => Some(v),
					Some(Err(e)) => return Some(Err(e)),
					None => None,
				};
			}

			match (self.pending_iter.peek(), &self.next_storage) {
				(Some((pending_key, _)), Some(storage_val)) => {
					let cmp = pending_key.cmp(&storage_val.key);
					let should_yield_pending = if self.reverse {
						// Reverse: larger keys first
						matches!(cmp, Ordering::Greater)
					} else {
						// Forward: smaller keys first
						matches!(cmp, Ordering::Less)
					};

					if should_yield_pending {
						// Pending key comes first
						let (key, value) = self.pending_iter.next().unwrap();
						if let Some(row) = value.row() {
							return Some(Ok(MultiVersionRow {
								key,
								row: row.clone(),
								version: value.version,
							}));
						}
						// Tombstone: skip (continue loop)
					} else if matches!(cmp, Ordering::Equal) {
						// Same key - pending shadows storage
						let (key, value) = self.pending_iter.next().unwrap();
						self.next_storage = None; // Consume storage entry
						if let Some(row) = value.row() {
							return Some(Ok(MultiVersionRow {
								key,
								row: row.clone(),
								version: value.version,
							}));
						}
						// Tombstone: skip (continue loop)
					} else {
						// Storage key comes first
						return Some(Ok(self.next_storage.take().unwrap()));
					}
				}
				(Some(_), None) => {
					// Only pending left
					let (key, value) = self.pending_iter.next().unwrap();
					if let Some(row) = value.row() {
						return Some(Ok(MultiVersionRow {
							key,
							row: row.clone(),
							version: value.version,
						}));
					}
					// Tombstone: skip (continue loop)
				}
				(None, Some(_)) => {
					// Only storage left
					return Some(Ok(self.next_storage.take().unwrap()));
				}
				(None, None) => return None,
			}
		}
	}
}