feat: memtable filter push down (#2539)

* feat: memtable support filter pushdown to prune primary keys

* fix: switch to next time series when pk not selected

* fix: allow predicate evaluation failure

* fix: some clippy warnings

* fix: panic when no primary key in schema

* feat: cache decoded record batch for primary key

* refactor: use arcswap instead of rwlock

* fix: format toml

src/mito2/Cargo.toml

@@ -12,6 +12,7 @@ test = ["common-test-util"]
 anymap = "1.0.0-beta.2"
 api.workspace = true
 aquamarine.workspace = true
+arc-swap = "1.6"
 async-channel = "1.9"
 async-compat = "0.2"
 async-stream.workspace = true

src/mito2/src/flush.rs

@@ -307,7 +307,7 @@ impl RegionFlushTask {
             }
 
             let file_id = FileId::random();
-            let iter = mem.iter(None, &[]);
+            let iter = mem.iter(None, None);
             let source = Source::Iter(iter);
             let mut writer = self
                 .access_layer

src/mito2/src/memtable.rs

@@ -23,11 +23,11 @@ use std::fmt;
 use std::sync::atomic::{AtomicBool, AtomicUsize, Ordering};
 use std::sync::Arc;
 
-use common_query::logical_plan::Expr;
 use common_time::Timestamp;
 use metrics::{decrement_gauge, increment_gauge};
 use store_api::metadata::RegionMetadataRef;
 use store_api::storage::ColumnId;
+use table::predicate::Predicate;
 
 use crate::error::Result;
 use crate::flush::WriteBufferManagerRef;
@@ -73,7 +73,11 @@ pub trait Memtable: Send + Sync + fmt::Debug {
     /// Scans the memtable.
     /// `projection` selects columns to read, `None` means reading all columns.
     /// `filters` are the predicates to be pushed down to memtable.
-    fn iter(&self, projection: Option<&[ColumnId]>, filters: &[Expr]) -> BoxedBatchIterator;
+    fn iter(
+        &self,
+        projection: Option<&[ColumnId]>,
+        predicate: Option<Predicate>,
+    ) -> BoxedBatchIterator;
 
     /// Returns true if the memtable is empty.
     fn is_empty(&self) -> bool;