diff --git a/apps/recorder/.gitignore b/apps/recorder/.gitignore
index 1b5bbe1..3d2eadd 100644
--- a/apps/recorder/.gitignore
+++ b/apps/recorder/.gitignore
@@ -27,3 +27,5 @@ node_modules
 dist/
 temp/*
 !temp/.gitkeep
+tests/resources/mikan/classic_episodes/*/*
+!tests/resources/mikan/classic_episodes/parquet/tiny.parquet
\ No newline at end of file
diff --git a/apps/recorder/Cargo.toml b/apps/recorder/Cargo.toml
index 4cb8713..a8a9b8e 100644
--- a/apps/recorder/Cargo.toml
+++ b/apps/recorder/Cargo.toml
@@ -154,7 +154,11 @@ icu = "2.0.0"
 tracing-tree = "0.4.0"
 num_cpus = "1.17.0"
 headers-accept = "0.1.4"
-polars = { version = "0.49.1", features = ["parquet"], optional = true }
+polars = { version = "0.49.1", features = [
+    "parquet",
+    "lazy",
+    "diagonal_concat",
+], optional = true }
 
 [dev-dependencies]
 inquire = { workspace = true }
diff --git a/apps/recorder/examples/mikan_collect_classic_eps.rs b/apps/recorder/examples/mikan_collect_classic_eps.rs
index 0a5fd54..257e797 100644
--- a/apps/recorder/examples/mikan_collect_classic_eps.rs
+++ b/apps/recorder/examples/mikan_collect_classic_eps.rs
@@ -2,6 +2,7 @@ use std::collections::HashSet;
 
 use chrono::{DateTime, Duration, FixedOffset, NaiveDate, NaiveTime, TimeZone, Utc};
 use fetch::{HttpClientConfig, fetch_html};
+use itertools::Itertools;
 use lazy_static::lazy_static;
 use nom::{
     IResult, Parser,
@@ -398,6 +399,136 @@ async fn scrape_mikan_classic_episode_table_page_from_rev_id(
     scrape_mikan_classic_episode_table_page(mikan_client, page, Some((rev_idx, total))).await
 }
 
+async fn merge_mikan_classic_episodes_and_strip_columns() -> RecorderResult<()> {
+    use polars::prelude::*;
+
+    let dir = TEST_FOLDER.join("parquet");
+    let files = std::fs::read_dir(dir)?;
+
+    let parquet_paths = files
+        .filter_map(|f| f.ok())
+        .filter_map(|f| {
+            let path = f.path();
+            if let Some(ext) = path.extension()
+                && ext == "parquet"
+                && path
+                    .file_stem()
+                    .is_some_and(|f| f.to_string_lossy().starts_with("rev_"))
+            {
+                Some(path)
+            } else {
+                None
+            }
+        })
+        .collect::<Vec<_>>();
+
+    if parquet_paths.is_empty() {
+        return Err(RecorderError::without_source(
+            "No parquet files found to merge".into(),
+        ));
+    }
+
+    println!("Found {} parquet files to merge", parquet_paths.len());
+
+    // 读取并合并所有 parquet 文件
+    let mut all_dfs = Vec::new();
+    for path in &parquet_paths {
+        println!("Reading {path:?}");
+        let file = std::fs::File::open(path)?;
+        let df = ParquetReader::new(file).finish().map_err(|e| {
+            let message = format!("Failed to read parquet file {path:?}: {e}");
+            RecorderError::with_source(Box::new(e), message)
+        })?;
+        all_dfs.push(df);
+    }
+
+    let lazy_frames: Vec<LazyFrame> = all_dfs.into_iter().map(|df| df.lazy()).collect();
+
+    let merged_df = concat_lf_diagonal(&lazy_frames, UnionArgs::default())
+        .map_err(|e| {
+            let message = format!("Failed to concat DataFrames: {e}");
+            RecorderError::with_source(Box::new(e), message)
+        })?
+        .sort(
+            ["publish_at_timestamp"],
+            SortMultipleOptions::default().with_order_descending(true),
+        )
+        .unique(
+            Some(vec![
+                "mikan_fansub_id".to_string(),
+                "mikan_episode_id".to_string(),
+            ]),
+            UniqueKeepStrategy::First,
+        )
+        .collect()
+        .map_err(|e| {
+            let message = format!("Failed to collect lazy DataFrame: {e}");
+            RecorderError::with_source(Box::new(e), message)
+        })?;
+
+    fn select_columns_and_write(
+        merged_df: DataFrame,
+        name: &str,
+        columns: &[&str],
+    ) -> RecorderResult<()> {
+        let result_df = merged_df
+            .lazy()
+            .sort(["publish_at_timestamp"], SortMultipleOptions::default())
+            .select(columns.iter().map(|c| col(*c)).collect_vec())
+            .collect()
+            .map_err(|e| {
+                let message = format!("Failed to sort and select columns: {e}");
+                RecorderError::with_source(Box::new(e), message)
+            })?;
+
+        let output_path = TEST_FOLDER.join(format!("parquet/{name}.parquet"));
+        let mut output_file = std::fs::File::create(&output_path)?;
+
+        ParquetWriter::new(&mut output_file)
+            .set_parallel(true)
+            .with_compression(ParquetCompression::Zstd(Some(
+                ZstdLevel::try_new(22).unwrap(),
+            )))
+            .finish(&mut result_df.clone())
+            .map_err(|e| {
+                let message = format!("Failed to write merged parquet file: {e}");
+                RecorderError::with_source(Box::new(e), message)
+            })?;
+
+        println!("Merged {} rows into {output_path:?}", result_df.height());
+        Ok(())
+    }
+
+    select_columns_and_write(merged_df.clone(), "tiny", &["fansub_name", "original_name"])?;
+    select_columns_and_write(
+        merged_df.clone(),
+        "lite",
+        &[
+            "mikan_fansub_id",
+            "fansub_name",
+            "mikan_episode_id",
+            "original_name",
+        ],
+    )?;
+    select_columns_and_write(
+        merged_df,
+        "full",
+        &[
+            "id",
+            "publish_at_timestamp",
+            "mikan_fansub_id",
+            "fansub_name",
+            "mikan_episode_id",
+            "original_name",
+            "magnet_link",
+            "file_size",
+            "torrent_link",
+        ],
+    )?;
+
+    Ok(())
+}
+
 #[tokio::main]
 async fn main() -> RecorderResult<()> {
     std::fs::create_dir_all(TEST_FOLDER.join("html"))?;
@@ -442,5 +573,12 @@ async fn main() -> RecorderResult<()> {
         page.save_to_files()?;
     }
 
+    // 合并所有 parquet 文件
+    println!("\nMerging all parquet files...");
+
+    merge_mikan_classic_episodes_and_strip_columns().await?;
+
+    println!("Merge completed!");
+
     Ok(())
 }
diff --git a/apps/recorder/tests/resources/mikan/classic_episodes/parquet/tiny.parquet b/apps/recorder/tests/resources/mikan/classic_episodes/parquet/tiny.parquet
new file mode 100644
index 0000000..c7a7a69
Binary files /dev/null and b/apps/recorder/tests/resources/mikan/classic_episodes/parquet/tiny.parquet differ