feat: optimistic transaction protocol

rust/Cargo.toml

@@ -13,7 +13,7 @@ readme = "README.md"
 edition = "2021"
 
 [dependencies]
-arrow = { version = "33.0.0", optional = true }
+arrow = { version = "33", optional = true }
 async-trait = "0.1"
 bytes = "1"
 chrono = { version = "0.4.22", default-features = false, features = ["clock"] }
@@ -26,7 +26,7 @@ log = "0"
 libc = ">=0.2.90, <1"
 num-bigint = "0.4"
 num-traits = "0.2.15"
-object_store = "0.5.3"
+object_store = "0.5.6"
 once_cell = "1.16.0"
 parking_lot = "0.12"
 parquet = { version = "33", features = [
@@ -57,6 +57,8 @@ datafusion = { version = "19", optional = true }
 datafusion-expr = { version = "19", optional = true }
 datafusion-common = { version = "19", optional = true }
 datafusion-proto = { version = "19", optional = true }
+datafusion-sql = { version = "19", optional = true }
+sqlparser = { version = "0.30", optional = true }
 
 # NOTE dependencies only for integration tests
 fs_extra = { version = "1.2.0", optional = true }
@@ -91,6 +93,8 @@ datafusion = [
     "datafusion-expr",
     "datafusion-common",
     "datafusion-proto",
+    "datafusion-sql",
+    "sqlparser",
     "arrow",
     "parquet",
 ]

rust/src/action/mod.rs

@@ -544,6 +544,13 @@ pub enum DeltaOperation {
         /// The predicate used during the write.
         predicate: Option<String>,
     },
+
+    /// Delete data matching predicate from delta table
+    Delete {
+        /// The condition the to be deleted data must match
+        predicate: Option<String>,
+    },
+
     /// Represents a Delta `StreamingUpdate` operation.
     #[serde(rename_all = "camelCase")]
     StreamingUpdate {
@@ -580,6 +587,7 @@ impl DeltaOperation {
             }
             DeltaOperation::Create { .. } => "CREATE TABLE",
             DeltaOperation::Write { .. } => "WRITE",
+            DeltaOperation::Delete { .. } => "DELETE",
             DeltaOperation::StreamingUpdate { .. } => "STREAMING UPDATE",
             DeltaOperation::Optimize { .. } => "OPTIMIZE",
             DeltaOperation::FileSystemCheck { .. } => "FSCK",
@@ -622,7 +630,8 @@ impl DeltaOperation {
             Self::Create { .. }
             | Self::FileSystemCheck {}
             | Self::StreamingUpdate { .. }
-            | Self::Write { .. } => true,
+            | Self::Write { .. }
+            | Self::Delete { .. } => true,
         }
     }
 
@@ -641,9 +650,20 @@ impl DeltaOperation {
         match self {
             // TODO add more operations
             Self::Write { predicate, .. } => predicate.clone(),
+            Self::Delete { predicate, .. } => predicate.clone(),
             _ => None,
         }
     }
+
+    /// Denotes if the operation reads the entire table
+    pub fn read_whole_table(&self) -> bool {
+        match self {
+            // TODO just adding one operation example, as currently none of the
+            // implemented operations scan the entire table.
+            Self::Write { predicate, .. } if predicate.is_none() => false,
+            _ => false,
+        }
+    }
 }
 
 /// The SaveMode used when performing a DeltaOperation

rust/src/delta.rs

@@ -17,6 +17,7 @@ use super::schema::*;
 use super::table_state::DeltaTableState;
 use crate::action::{Add, Stats};
 use crate::delta_config::DeltaConfigError;
+use crate::operations::transaction::TransactionError;
 use crate::operations::vacuum::VacuumBuilder;
 use crate::storage::{commit_uri_from_version, ObjectStoreRef};
 
@@ -216,6 +217,12 @@ pub enum DeltaTableError {
         #[from]
         source: std::io::Error,
     },
+    /// Error raised while commititng transaction
+    #[error("Transaction failed: {source}")]
+    Transaction {
+        /// The source error
+        source: TransactionError,
+    },
     /// Error returned when transaction is failed to be committed because given version already exists.
     #[error("Delta transaction failed, version {0} already exists.")]
     VersionAlreadyExists(DeltaDataTypeVersion),

rust/src/delta_datafusion.rs

@@ -390,7 +390,7 @@ impl TableProvider for DeltaTable {
             (!filters.is_empty()).then_some(conjunction(filters.iter().cloned()))
         {
             let pruning_predicate = PruningPredicate::try_new(predicate, schema.clone())?;
-            let files_to_prune = pruning_predicate.prune(self)?;
+            let files_to_prune = pruning_predicate.prune(&self.state)?;
             self.get_state()
                 .files()
                 .iter()
@@ -630,7 +630,7 @@ fn to_scalar_value(stat_val: &serde_json::Value) -> Option<ScalarValue> {
     }
 }
 
-fn to_correct_scalar_value(
+pub(crate) fn to_correct_scalar_value(
     stat_val: &serde_json::Value,
     field_dt: &ArrowDataType,
 ) -> Option<ScalarValue> {

rust/src/operations/create.rs

@@ -285,7 +285,6 @@ impl std::future::IntoFuture for CreateBuilder {
 
         Box::pin(async move {
             let mode = this.mode.clone();
-            let metadata = this.metadata.clone();
             let (mut table, actions, operation) = this.into_table_and_actions()?;
             if table.object_store().is_delta_table_location().await? {
                 match mode {
@@ -302,10 +301,10 @@ impl std::future::IntoFuture for CreateBuilder {
             }
             let version = commit(
                 table.object_store().as_ref(),
-                0,
                 &actions,
                 operation,
-                metadata,
+                &table.state,
+                None,
             )
             .await?;
             table.load_version(version).await?;

rust/src/operations/filesystem_check.rs

@@ -15,7 +15,6 @@ use crate::action::{Action, Add, DeltaOperation, Remove};
 use crate::operations::transaction::commit;
 use crate::storage::DeltaObjectStore;
 use crate::table_state::DeltaTableState;
-use crate::DeltaDataTypeVersion;
 use crate::{DeltaDataTypeLong, DeltaResult, DeltaTable, DeltaTableError};
 use futures::future::BoxFuture;
 use futures::StreamExt;
@@ -50,8 +49,6 @@ pub struct FileSystemCheckMetrics {
 }
 
 struct FileSystemCheckPlan {
-    /// Version of the snapshot provided
-    version: DeltaDataTypeVersion,
     /// Delta object store for handling data files
     store: Arc<DeltaObjectStore>,
     /// Files that no longer exists in undlying ObjectStore but have active add actions
@@ -88,7 +85,6 @@ impl FileSystemCheckBuilder {
     async fn create_fsck_plan(&self) -> DeltaResult<FileSystemCheckPlan> {
         let mut files_relative: HashMap<&str, &Add> =
             HashMap::with_capacity(self.state.files().len());
-        let version = self.state.version();
         let store = self.store.clone();
 
         for active in self.state.files() {
@@ -118,14 +114,13 @@ impl FileSystemCheckBuilder {
 
         Ok(FileSystemCheckPlan {
             files_to_remove,
-            version,
             store,
         })
     }
 }
 
 impl FileSystemCheckPlan {
-    pub async fn execute(self) -> DeltaResult<FileSystemCheckMetrics> {
+    pub async fn execute(self, snapshot: &DeltaTableState) -> DeltaResult<FileSystemCheckMetrics> {
         if self.files_to_remove.is_empty() {
             return Ok(FileSystemCheckMetrics {
                 dry_run: false,
@@ -135,8 +130,6 @@ impl FileSystemCheckPlan {
 
         let mut actions = Vec::with_capacity(self.files_to_remove.len());
         let mut removed_file_paths = Vec::with_capacity(self.files_to_remove.len());
-        let version = self.version;
-        let store = &self.store;
 
         for file in self.files_to_remove {
             let deletion_time = SystemTime::now().duration_since(UNIX_EPOCH).unwrap();
@@ -154,10 +147,11 @@ impl FileSystemCheckPlan {
         }
 
         commit(
-            store,
-            version + 1,
+            self.store.as_ref(),
             &actions,
             DeltaOperation::FileSystemCheck {},
+            snapshot,
+            // TODO pass through metadata
             None,
         )
         .await?;
@@ -188,7 +182,7 @@ impl std::future::IntoFuture for FileSystemCheckBuilder {
                 ));
             }
 
-            let metrics = plan.execute().await?;
+            let metrics = plan.execute(&this.state).await?;
             let mut table = DeltaTable::new_with_state(this.store, this.state);
             table.update().await?;
             Ok((table, metrics))

rust/src/operations/optimize.rs

@@ -173,7 +173,7 @@ impl<'a> std::future::IntoFuture for OptimizeBuilder<'a> {
                 this.target_size.to_owned(),
                 writer_properties,
             )?;
-            let metrics = plan.execute(this.store.clone()).await?;
+            let metrics = plan.execute(this.store.clone(), &this.snapshot).await?;
             let mut table = DeltaTable::new_with_state(this.store, this.snapshot);
             table.update().await?;
             Ok((table, metrics))
@@ -270,7 +270,11 @@ pub struct MergePlan {
 
 impl MergePlan {
     /// Peform the operations outlined in the plan.
-    pub async fn execute(self, object_store: ObjectStoreRef) -> Result<Metrics, DeltaTableError> {
+    pub async fn execute(
+        self,
+        object_store: ObjectStoreRef,
+        snapshot: &DeltaTableState,
+    ) -> Result<Metrics, DeltaTableError> {
         let mut actions = vec![];
         let mut metrics = self.metrics;
 
@@ -368,9 +372,9 @@ impl MergePlan {
 
             commit(
                 object_store.as_ref(),
-                self.read_table_version + 1,
                 &actions,
                 self.input_parameters.into(),
+                snapshot,
                 Some(metadata),
             )
             .await?;