Added concat support for s3 and hybrid-s3.

Author: s3rius

Cargo.lock

@@ -25,6 +25,7 @@ url = "^2.5.4"
 prometheus = "^0.13.4"
 actix-web-prom = "^0.9.0"
 dyn-clone = "^1.0.17"
+tempdir = "0.3.7"
 actix-cors = "0.7.0"
 wildmatch = "2.4.0"
 md-5 = "^0.10.6"
@@ -69,7 +70,6 @@ test_rmq = []
 integration_tests = ["test_redis", "test_rmq"]
 
 [dev-dependencies]
-tempdir = "0.3.7"
 actix-rt = "2.10.0"
 httptest = "0.16.1"
 actix-http = "3.9.0"

Cargo.toml

@@ -182,6 +182,7 @@ Parameters:
 * `--s3-profile` - Name of the section from `~/.aws/credentials` file;
 * `--s3-headers` - JSON object with additional header to every S3 request (Useful for setting ACLs);
 * `--s3-force-path-style` - use path style URL. It appends bucket name at the end of the URL;
+* `--s3-concat-concurrent-downloads` - Number of concurrent downloads of partial files from S3. When performing concatenation, Rustus downloads all partial files from S3 and concatenates them into a single file. This parameter controls the number of concurrent downloads. Default value is 10.
 
 Required parameter are only `--s3-url` and `--s3-bucket`.
 
@@ -203,7 +204,8 @@ Required parameter are only `--s3-url` and `--s3-bucket`.
         --s3-headers '{"x-amz-acl": "public-read"}' \
         --force-fsync \
         --data-dir "./data/" \
-        --dir-structure "{year}/{month}/{day}"
+        --dir-structure "{year}/{month}/{day}" \
+        --s3-concat-concurrent-downloads 10
     ```
 
 === "ENV"
@@ -225,6 +227,7 @@ Required parameter are only `--s3-url` and `--s3-bucket`.
     export RUSTUS_DIR_STRUCTURE="{year}/{month}/{day}"
     export RUSTUS_FORCE_FSYNC="true"
     export RUSTUS_S3_FORCE_PATH_STYLE="true"
+    export RUSTUS_S3_CONCAT_CONCURRENT_DOWNLOADS="true"
 
     rustus
     ```
@@ -259,6 +262,8 @@ Parameters:
 * `--s3-headers` - JSON object with additional header to every S3 request (Useful for setting ACLs);
 * `--s3-force-path-style` - use path style URL. It appends bucket name at the end of the URL;
 * `--dir-structure` - pattern of a directory structure on s3;
+* `--s3-concat-concurrent-downloads` - Number of concurrent downloads of partial files from S3. When performing concatenation, Rustus downloads all partial files from S3 and concatenates them into a single file. This parameter controls the number of concurrent downloads. Default value is 10.
+
 
 Required parameter are only `--s3-url` and `--s3-bucket`.
 
@@ -278,7 +283,9 @@ Required parameter are only `--s3-url` and `--s3-bucket`.
         --s3-session-token "token" \
         --s3-force-path-style \
         --s3-headers '{"x-amz-acl": "public-read"}' \
-        --dir-structure "{year}/{month}/{day}"
+        --dir-structure "{year}/{month}/{day}" \
+        --s3-concat-concurrent-downloads 10
+
     ```
 
 === "ENV"
@@ -298,6 +305,7 @@ Required parameter are only `--s3-url` and `--s3-bucket`.
     export RUSTUS_S3_HEADERS='{"x-amz-acl": "public-read"}'
     export RUSTUS_S3_FORCE_PATH_STYLE="true"
     export RUSTUS_DIR_STRUCTURE="{year}/{month}/{day}"
+    export RUSTUS_S3_CONCAT_CONCURRENT_DOWNLOADS="10"
 
     rustus
     ```

docs/configuration.md

@@ -118,6 +118,21 @@ pub struct DataStorageOptions {
     /// This parameter is required fo s3-based storages.
     #[arg(long, env = "RUSTUS_S3_HEADERS")]
     pub s3_headers: Option<String>,
+
+    /// Number of concurrent downloads of partial files
+    /// from S3.
+    /// When performing concatenation, Rustus downloads
+    /// all partial files from S3 and concatenates them
+    /// into a single file.
+    ///
+    /// This parameter controls the number of concurrent
+    /// downloads.
+    #[arg(
+        long,
+        env = "RUSTUS_S3_CONCAT_CONCURRENT_DOWNLOADS",
+        default_value = "10"
+    )]
+    pub s3_concat_concurrent_downloads: usize,
 }
 
 #[derive(Parser, Debug, Clone)]

src/config.rs

@@ -12,11 +12,13 @@ use crate::utils::dir_struct::substr_time;
 use actix_web::{HttpRequest, HttpResponse, HttpResponseBuilder};
 use bytes::Bytes;
 use chrono::{DateTime, Utc};
+use futures::{StreamExt, TryStreamExt};
 use s3::{
     command::Command,
     request::{tokio_backend::HyperRequest, Request as S3Request},
     Bucket,
 };
+use tokio::io::AsyncWriteExt;
 
 use super::file_storage::FileDataStorage;
 
@@ -32,6 +34,7 @@ pub struct S3HybridDataStorage {
     bucket: Bucket,
     local_storage: FileDataStorage,
     dir_struct: String,
+    concurrent_concat_downloads: usize,
 }
 
 impl S3HybridDataStorage {
@@ -50,6 +53,7 @@ impl S3HybridDataStorage {
         data_dir: PathBuf,
         dir_struct: String,
         force_fsync: bool,
+        concurrent_concat_downloads: usize,
     ) -> Self {
         let local_storage = FileDataStorage::new(data_dir, dir_struct.clone(), force_fsync);
         let creds = s3::creds::Credentials::new(
@@ -91,6 +95,7 @@ impl S3HybridDataStorage {
             bucket: *bucket,
             local_storage,
             dir_struct,
+            concurrent_concat_downloads,
         }
     }
 
@@ -170,12 +175,67 @@ impl DataStorage for S3HybridDataStorage {
 
     async fn concat_files(
         &self,
-        _file_info: &FileInfo,
-        _parts_info: Vec<FileInfo>,
+        file_info: &FileInfo,
+        parts_info: Vec<FileInfo>,
     ) -> RustusResult<()> {
-        Err(RustusError::Unimplemented(
-            "Hybrid s3 cannot concat files.".into(),
-        ))
+        let dir = tempdir::TempDir::new(&file_info.id)?;
+        let mut download_futures = vec![];
+
+        // At first we need to download all parts.
+        for part_info in &parts_info {
+            let part_key = self.get_s3_key(&part_info.id, part_info.created_at);
+            let part_out = dir.path().join(&part_info.id);
+            // Here we create a future which downloads the part
+            // into a temporary file.
+            download_futures.push(async move {
+                let part_file = tokio::fs::File::create(&part_out).await?;
+                let mut writer = tokio::io::BufWriter::new(part_file);
+                let mut reader = self.bucket.get_object_stream(&part_key).await?;
+                while let Some(chunk) = reader.bytes().next().await {
+                    let mut chunk = chunk?;
+                    writer.write_all_buf(&mut chunk).await?;
+                }
+                Ok::<_, RustusError>(())
+            });
+        }
+        // Here we await all download futures.
+        // We use buffer_unordered to limit the number of concurrent downloads.
+        futures::stream::iter(download_futures)
+            // Number of concurrent downloads.
+            .buffer_unordered(self.concurrent_concat_downloads)
+            // We use try_collect to collect all results
+            // and return an error if any of the futures returned an error.
+            .try_collect::<Vec<_>>()
+            .await?;
+
+        let key = self.get_s3_key(&file_info.id, file_info.created_at);
+        let output_path = dir.path().join(&file_info.id);
+        let output_file = tokio::fs::OpenOptions::new()
+            .create_new(true)
+            .write(true)
+            .open(&output_path)
+            .await?;
+        let mut writer = tokio::io::BufWriter::new(output_file);
+        for part_info in &parts_info {
+            let part_out = dir.path().join(&part_info.id);
+            let part_file = tokio::fs::File::open(&part_out).await?;
+            let mut reader = tokio::io::BufReader::new(part_file);
+            tokio::io::copy(&mut reader, &mut writer).await?;
+            writer.flush().await?;
+            tokio::fs::remove_file(part_out).await?;
+        }
+        writer.flush().await?;
+        let output_file = writer.into_inner();
+        output_file.sync_all().await?;
+
+        // We reopen the file to upload it to S3.
+        // This is needed because we need to open the file in read mode.
+        let output_file = tokio::fs::File::open(&output_path).await?;
+        let mut reader = tokio::io::BufReader::new(output_file);
+
+        self.bucket.put_object_stream(&mut reader, key).await?;
+        tokio::fs::remove_file(output_path).await?;
+        Ok(())
     }
 
     async fn remove_file(&self, file_info: &FileInfo) -> RustusResult<()> {