diff --git a/Cargo.lock b/Cargo.lock
index d60af01a9f91..a827be32fd1f 100644
--- a/Cargo.lock
+++ b/Cargo.lock
@@ -5070,8 +5070,12 @@ name = "promql"
 version = "0.1.0"
 dependencies = [
  "common-error",
+ "datafusion",
+ "datatypes",
+ "futures",
  "promql-parser",
  "snafu",
+ "tokio",
 ]
 
 [[package]]
diff --git a/src/datatypes/src/schema.rs b/src/datatypes/src/schema.rs
index 4952e36cc0fd..cd738448f0c9 100644
--- a/src/datatypes/src/schema.rs
+++ b/src/datatypes/src/schema.rs
@@ -20,6 +20,7 @@ use std::collections::HashMap;
 use std::sync::Arc;
 
 use arrow::datatypes::{Field, Schema as ArrowSchema};
+pub use column_schema::TIME_INDEX_KEY;
 use datafusion_common::DFSchemaRef;
 use snafu::{ensure, ResultExt};
 
@@ -30,7 +31,7 @@ pub use crate::schema::constraint::ColumnDefaultConstraint;
 pub use crate::schema::raw::RawSchema;
 
 /// Key used to store version number of the schema in metadata.
-const VERSION_KEY: &str = "greptime:version";
+pub const VERSION_KEY: &str = "greptime:version";
 
 /// A common schema, should be immutable.
 #[derive(Debug, Clone, PartialEq, Eq)]
diff --git a/src/datatypes/src/schema/column_schema.rs b/src/datatypes/src/schema/column_schema.rs
index 94fbba87eccb..7fc76fdda95a 100644
--- a/src/datatypes/src/schema/column_schema.rs
+++ b/src/datatypes/src/schema/column_schema.rs
@@ -26,7 +26,7 @@ use crate::vectors::VectorRef;
 pub type Metadata = HashMap<String, String>;
 
 /// Key used to store whether the column is time index in arrow field's metadata.
-const TIME_INDEX_KEY: &str = "greptime:time_index";
+pub const TIME_INDEX_KEY: &str = "greptime:time_index";
 /// Key used to store default constraint in arrow field's metadata.
 const DEFAULT_CONSTRAINT_KEY: &str = "greptime:default_constraint";
 
diff --git a/src/promql/Cargo.toml b/src/promql/Cargo.toml
index 7bc49ea2138c..48b988f92357 100644
--- a/src/promql/Cargo.toml
+++ b/src/promql/Cargo.toml
@@ -6,5 +6,11 @@ license.workspace = true
 
 [dependencies]
 common-error = { path = "../common/error" }
+datafusion.workspace = true
+datatypes = { path = "../datatypes" }
+futures = "0.3"
 promql-parser = { git = "https://github.com/GreptimeTeam/promql-parser.git", rev = "71d8a90" }
 snafu = { version = "0.7", features = ["backtraces"] }
+
+[dev-dependencies]
+tokio = { version = "1.23", features = ["full"] }
diff --git a/src/promql/src/extension_plan.rs b/src/promql/src/extension_plan.rs
new file mode 100644
index 000000000000..c314ce041cec
--- /dev/null
+++ b/src/promql/src/extension_plan.rs
@@ -0,0 +1,17 @@
+// Copyright 2022 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+mod normalize;
+
+pub use normalize::{SeriesNormalize, SeriesNormalizeExec, SeriesNormalizeStream};
diff --git a/src/promql/src/extension_plan/normalize.rs b/src/promql/src/extension_plan/normalize.rs
new file mode 100644
index 000000000000..ae08dc62c4e7
--- /dev/null
+++ b/src/promql/src/extension_plan/normalize.rs
@@ -0,0 +1,355 @@
+// Copyright 2022 Greptime Team
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+use std::any::Any;
+use std::pin::Pin;
+use std::sync::Arc;
+use std::task::{Context, Poll};
+use std::time::Duration;
+
+use datafusion::arrow::compute;
+use datafusion::common::{DFSchemaRef, Result as DataFusionResult, Statistics};
+use datafusion::execution::context::TaskContext;
+use datafusion::logical_expr::{LogicalPlan, UserDefinedLogicalNode};
+use datafusion::physical_expr::PhysicalSortExpr;
+use datafusion::physical_plan::metrics::{BaselineMetrics, ExecutionPlanMetricsSet, MetricsSet};
+use datafusion::physical_plan::{
+    DisplayFormatType, ExecutionPlan, Partitioning, RecordBatchStream, SendableRecordBatchStream,
+};
+use datatypes::arrow::array::{ArrowPrimitiveType, TimestampMillisecondArray};
+use datatypes::arrow::datatypes::{SchemaRef, TimestampMillisecondType};
+use datatypes::arrow::error::Result as ArrowResult;
+use datatypes::arrow::record_batch::RecordBatch;
+use futures::{Stream, StreamExt};
+
+type Millisecond = <TimestampMillisecondType as ArrowPrimitiveType>::Native;
+
+/// Normalize the input record batch. Notice that for simplicity, this method assumes
+/// the input batch only contains sample points from one time series.
+///
+/// Roughly speaking, this method does these things:
+/// - bias sample's timestamp by offset
+/// - sort the record batch based on timestamp column
+#[derive(Debug)]
+pub struct SeriesNormalize {
+    offset: Millisecond,
+    time_index_column_name: String,
+
+    input: LogicalPlan,
+}
+
+impl UserDefinedLogicalNode for SeriesNormalize {
+    fn as_any(&self) -> &dyn Any {
+        self as _
+    }
+
+    fn inputs(&self) -> Vec<&LogicalPlan> {
+        vec![&self.input]
+    }
+
+    fn schema(&self) -> &DFSchemaRef {
+        self.input.schema()
+    }
+
+    fn expressions(&self) -> Vec<datafusion::logical_expr::Expr> {
+        vec![]
+    }
+
+    fn fmt_for_explain(&self, f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        write!(f, "PromSeriesNormalize: offset=[{}]", self.offset)
+    }
+
+    fn from_template(
+        &self,
+        _exprs: &[datafusion::logical_expr::Expr],
+        inputs: &[LogicalPlan],
+    ) -> Arc<dyn UserDefinedLogicalNode> {
+        assert!(!inputs.is_empty());
+
+        Arc::new(Self {
+            offset: self.offset,
+            time_index_column_name: self.time_index_column_name.clone(),
+            input: inputs[0].clone(),
+        })
+    }
+}
+
+impl SeriesNormalize {
+    pub fn new<N: AsRef<str>>(
+        offset: Duration,
+        time_index_column_name: N,
+        input: LogicalPlan,
+    ) -> Self {
+        Self {
+            offset: offset.as_millis() as i64,
+            time_index_column_name: time_index_column_name.as_ref().to_string(),
+            input,
+        }
+    }
+
+    pub fn to_execution_plan(&self, exec_input: Arc<dyn ExecutionPlan>) -> Arc<dyn ExecutionPlan> {
+        Arc::new(SeriesNormalizeExec {
+            offset: self.offset,
+            time_index_column_name: self.time_index_column_name.clone(),
+            input: exec_input,
+            metric: ExecutionPlanMetricsSet::new(),
+        })
+    }
+}
+
+#[derive(Debug)]
+pub struct SeriesNormalizeExec {
+    offset: Millisecond,
+    time_index_column_name: String,
+
+    input: Arc<dyn ExecutionPlan>,
+    metric: ExecutionPlanMetricsSet,
+}
+
+impl ExecutionPlan for SeriesNormalizeExec {
+    fn as_any(&self) -> &dyn Any {
+        self
+    }
+
+    fn schema(&self) -> SchemaRef {
+        self.input.schema()
+    }
+
+    fn output_partitioning(&self) -> Partitioning {
+        self.input.output_partitioning()
+    }
+
+    fn output_ordering(&self) -> Option<&[PhysicalSortExpr]> {
+        self.input.output_ordering()
+    }
+
+    fn maintains_input_order(&self) -> bool {
+        false
+    }
+
+    fn children(&self) -> Vec<Arc<dyn ExecutionPlan>> {
+        vec![self.input.clone()]
+    }
+
+    fn with_new_children(
+        self: Arc<Self>,
+        children: Vec<Arc<dyn ExecutionPlan>>,
+    ) -> DataFusionResult<Arc<dyn ExecutionPlan>> {
+        assert!(!children.is_empty());
+        Ok(Arc::new(Self {
+            offset: self.offset,
+            time_index_column_name: self.time_index_column_name.clone(),
+            input: children[0].clone(),
+            metric: self.metric.clone(),
+        }))
+    }
+
+    fn execute(
+        &self,
+        partition: usize,
+        context: Arc<TaskContext>,
+    ) -> DataFusionResult<SendableRecordBatchStream> {
+        let baseline_metric = BaselineMetrics::new(&self.metric, partition);
+
+        let input = self.input.execute(partition, context)?;
+        let schema = input.schema();
+        let time_index = schema
+            .column_with_name(&self.time_index_column_name)
+            .expect("time index column not found")
+            .0;
+        Ok(Box::pin(SeriesNormalizeStream {
+            offset: self.offset,
+            time_index,
+            schema,
+            input,
+            metric: baseline_metric,
+        }))
+    }
+
+    fn fmt_as(&self, t: DisplayFormatType, f: &mut std::fmt::Formatter) -> std::fmt::Result {
+        match t {
+            DisplayFormatType::Default => {
+                write!(f, "PromSeriesNormalizeExec: offset=[{}]", self.offset)
+            }
+        }
+    }
+
+    fn metrics(&self) -> Option<MetricsSet> {
+        Some(self.metric.clone_inner())
+    }
+
+    fn statistics(&self) -> Statistics {
+        self.input.statistics()
+    }
+}
+
+pub struct SeriesNormalizeStream {
+    offset: Millisecond,
+    time_index: usize,
+
+    schema: SchemaRef,
+    input: SendableRecordBatchStream,
+    metric: BaselineMetrics,
+}
+
+impl SeriesNormalizeStream {
+    pub fn normalize(&self, input: RecordBatch) -> ArrowResult<RecordBatch> {
+        // TODO(ruihang): maybe the input is not timestamp millisecond array
+        let ts_column = input
+            .column(self.time_index)
+            .as_any()
+            .downcast_ref::<TimestampMillisecondArray>()
+            .unwrap();
+
+        // bias the timestamp column by offset
+        let ts_column_biased = if self.offset == 0 {
+            ts_column.clone()
+        } else {
+            TimestampMillisecondArray::from_iter(
+                ts_column.iter().map(|ts| ts.map(|ts| ts - self.offset)),
+            )
+        };
+        let mut columns = input.columns().to_vec();
+        columns[self.time_index] = Arc::new(ts_column_biased);
+
+        // sort the record batch
+        let ordered_indices = compute::sort_to_indices(&columns[self.time_index], None, None)?;
+        let ordered_columns = columns
+            .iter()
+            .map(|array| compute::take(array, &ordered_indices, None))
+            .collect::<ArrowResult<Vec<_>>>()?;
+        RecordBatch::try_new(input.schema(), ordered_columns)
+    }
+}
+
+impl RecordBatchStream for SeriesNormalizeStream {
+    fn schema(&self) -> SchemaRef {
+        self.schema.clone()
+    }
+}
+
+impl Stream for SeriesNormalizeStream {
+    type Item = ArrowResult<RecordBatch>;
+
+    fn poll_next(mut self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<Option<Self::Item>> {
+        let poll = match self.input.poll_next_unpin(cx) {
+            Poll::Ready(batch) => {
+                let _timer = self.metric.elapsed_compute().timer();
+                Poll::Ready(batch.map(|batch| batch.and_then(|batch| self.normalize(batch))))
+            }
+            Poll::Pending => Poll::Pending,
+        };
+        self.metric.record_poll(poll)
+    }
+}
+
+#[cfg(test)]
+mod test {
+    use datafusion::arrow::array::Float64Array;
+    use datafusion::arrow::datatypes::{DataType, Field, Schema};
+    use datafusion::from_slice::FromSlice;
+    use datafusion::physical_plan::memory::MemoryExec;
+    use datafusion::prelude::SessionContext;
+    use datatypes::arrow::array::TimestampMillisecondArray;
+    use datatypes::arrow_array::StringArray;
+
+    use super::*;
+
+    const TIME_INDEX_COLUMN: &str = "timestamp";
+
+    fn prepare_test_data() -> MemoryExec {
+        let schema = Arc::new(Schema::new(vec![
+            Field::new(TIME_INDEX_COLUMN, TimestampMillisecondType::DATA_TYPE, true),
+            Field::new("value", DataType::Float64, true),
+            Field::new("path", DataType::Utf8, true),
+        ]));
+        let timestamp_column = Arc::new(TimestampMillisecondArray::from_slice([
+            60_000, 120_000, 0, 30_000, 90_000,
+        ])) as _;
+        let value_column = Arc::new(Float64Array::from_slice([0.0, 1.0, 10.0, 100.0, 1000.0])) as _;
+        let path_column =
+            Arc::new(StringArray::from_slice(["foo", "foo", "foo", "foo", "foo"])) as _;
+        let data = RecordBatch::try_new(
+            schema.clone(),
+            vec![timestamp_column, value_column, path_column],
+        )
+        .unwrap();
+
+        MemoryExec::try_new(&[vec![data]], schema, None).unwrap()
+    }
+
+    #[tokio::test]
+    async fn test_sort_record_batch() {
+        let memory_exec = Arc::new(prepare_test_data());
+        let normalize_exec = Arc::new(SeriesNormalizeExec {
+            offset: 0,
+            time_index_column_name: TIME_INDEX_COLUMN.to_string(),
+            input: memory_exec,
+            metric: ExecutionPlanMetricsSet::new(),
+        });
+        let session_context = SessionContext::default();
+        let result = datafusion::physical_plan::collect(normalize_exec, session_context.task_ctx())
+            .await
+            .unwrap();
+        let result_literal = datatypes::arrow::util::pretty::pretty_format_batches(&result)
+            .unwrap()
+            .to_string();
+
+        let expected = String::from(
+            "+---------------------+-------+------+\
+            \n| timestamp           | value | path |\
+            \n+---------------------+-------+------+\
+            \n| 1970-01-01T00:00:00 | 10    | foo  |\
+            \n| 1970-01-01T00:00:30 | 100   | foo  |\
+            \n| 1970-01-01T00:01:00 | 0     | foo  |\
+            \n| 1970-01-01T00:01:30 | 1000  | foo  |\
+            \n| 1970-01-01T00:02:00 | 1     | foo  |\
+            \n+---------------------+-------+------+",
+        );
+
+        assert_eq!(result_literal, expected);
+    }
+
+    #[tokio::test]
+    async fn test_offset_record_batch() {
+        let memory_exec = Arc::new(prepare_test_data());
+        let normalize_exec = Arc::new(SeriesNormalizeExec {
+            offset: 1_000, // offset 1s
+            time_index_column_name: TIME_INDEX_COLUMN.to_string(),
+            input: memory_exec,
+            metric: ExecutionPlanMetricsSet::new(),
+        });
+        let session_context = SessionContext::default();
+        let result = datafusion::physical_plan::collect(normalize_exec, session_context.task_ctx())
+            .await
+            .unwrap();
+        let result_literal = datatypes::arrow::util::pretty::pretty_format_batches(&result)
+            .unwrap()
+            .to_string();
+
+        let expected = String::from(
+            "+---------------------+-------+------+\
+            \n| timestamp           | value | path |\
+            \n+---------------------+-------+------+\
+            \n| 1969-12-31T23:59:59 | 10    | foo  |\
+            \n| 1970-01-01T00:00:29 | 100   | foo  |\
+            \n| 1970-01-01T00:00:59 | 0     | foo  |\
+            \n| 1970-01-01T00:01:29 | 1000  | foo  |\
+            \n| 1970-01-01T00:01:59 | 1     | foo  |\
+            \n+---------------------+-------+------+",
+        );
+
+        assert_eq!(result_literal, expected);
+    }
+}
diff --git a/src/promql/src/lib.rs b/src/promql/src/lib.rs
index 11415d1838a2..552b4fd0ff60 100644
--- a/src/promql/src/lib.rs
+++ b/src/promql/src/lib.rs
@@ -14,3 +14,4 @@
 
 pub mod engine;
 pub mod error;
+pub mod extension_plan;