Content dedup on both sides, stored checksums, spool-free zips, upload stats
ci / docker (push) Successful in 12s

- sha256+crc32 hashed during upload streaming; unique index per album
- duplicate content returns the existing photo (race-safe via 23505)
- client hashes locally (WebCrypto) and skips the transfer entirely for
  content the album already has
- zip downloads stream S3->response directly using the stored crc32;
  pre-hash photos spool once and self-heal (crc via zip, sha via reprocess)
- upload UI: overall progress bar, bytes, live speed, ETA
This commit is contained in:
2026-07-17 14:41:07 +02:00
parent a6819809a7
commit 6259ca84d8
11 changed files with 296 additions and 60 deletions
+82 -26
View File
@@ -5,6 +5,7 @@ use axum::http::HeaderMap;
use axum::Json;
use futures::StreamExt;
use serde::Deserialize;
use sha2::Digest;
use tokio::io::AsyncWriteExt;
use uuid::Uuid;
@@ -60,17 +61,23 @@ pub async fn upload(
.unwrap_or("application/octet-stream")
.to_string();
// Stream the request body to a temp file so large raws never sit in memory.
// Stream the request body to a temp file so large raws never sit in
// memory, hashing as it flows: sha256 for duplicate detection, crc32 for
// spool-free zip downloads.
let dir = tempfile::tempdir().map_err(anyhow::Error::from)?;
let path = dir.path().join("upload.bin");
let mut file = tokio::fs::File::create(&path)
.await
.map_err(anyhow::Error::from)?;
let mut stream = body.into_data_stream();
let mut sha = sha2::Sha256::new();
let mut crc = crc32fast::Hasher::new();
let mut size: i64 = 0;
while let Some(chunk) = stream.next().await {
let chunk = chunk.map_err(|e| ApiError::bad_request(format!("upload aborted: {e}")))?;
size += chunk.len() as i64;
sha.update(&chunk);
crc.update(&chunk);
file.write_all(&chunk).await.map_err(anyhow::Error::from)?;
}
file.flush().await.map_err(anyhow::Error::from)?;
@@ -78,6 +85,20 @@ pub async fn upload(
if size == 0 {
return Err(ApiError::bad_request("empty upload"));
}
let sha256 = hex::encode(sha.finalize());
let crc32 = i64::from(crc.finalize());
// Same content already in this album? Return it — re-dragging a folder
// after a partial upload just fills the gaps instead of duplicating.
let existing: Option<Photo> =
sqlx::query_as("select * from photos where album_id = $1 and sha256 = $2")
.bind(album_id)
.bind(&sha256)
.fetch_optional(&state.db)
.await?;
if let Some(existing) = existing {
return Ok(Json(existing));
}
// Upload to S3 first, then create the row and enqueue processing in one
// transaction — a photo row can never exist without its job, and a failed
@@ -86,30 +107,33 @@ pub async fn upload(
let key = s3::original_key(photo_id, &filename);
s3::put_file(&state, &key, &path, &content_type).await?;
let result: ApiResult<(sqlx::Transaction<'static, sqlx::Postgres>, Photo)> = async {
let mut tx = state.db.begin().await?;
let photo: Photo = sqlx::query_as(
"insert into photos (id, album_id, filename, content_type, size_bytes, status)
values ($1, $2, $3, $4, $5, $6)
returning *",
)
.bind(photo_id)
.bind(album_id)
.bind(&filename)
.bind(&content_type)
.bind(size)
.bind(PhotoStatus::Uploaded.as_str())
.fetch_one(&mut *tx)
.await?;
jobs::enqueue(
&mut *tx,
JobKind::ProcessPhoto,
serde_json::json!({ "photo_id": photo_id }),
)
.await?;
Ok((tx, photo))
}
.await;
let result: Result<(sqlx::Transaction<'static, sqlx::Postgres>, Photo), sqlx::Error> =
async {
let mut tx = state.db.begin().await?;
let photo: Photo = sqlx::query_as(
"insert into photos (id, album_id, filename, content_type, size_bytes, status, sha256, crc32)
values ($1, $2, $3, $4, $5, $6, $7, $8)
returning *",
)
.bind(photo_id)
.bind(album_id)
.bind(&filename)
.bind(&content_type)
.bind(size)
.bind(PhotoStatus::Uploaded.as_str())
.bind(&sha256)
.bind(crc32)
.fetch_one(&mut *tx)
.await?;
jobs::enqueue(
&mut *tx,
JobKind::ProcessPhoto,
serde_json::json!({ "photo_id": photo_id }),
)
.await?;
Ok((tx, photo))
}
.await;
let (tx, photo) = match result {
Ok(pair) => pair,
@@ -118,7 +142,24 @@ pub async fn upload(
if let Err(cleanup) = s3::delete_prefix(&state, &s3::photo_prefix(photo_id)).await {
tracing::error!("failed to clean up s3 after aborted upload: {cleanup:#}");
}
return Err(e);
// Concurrent identical upload beat us to the unique index — hand
// back the winner instead of an error.
let unique_violation = matches!(
&e,
sqlx::Error::Database(db) if db.code().as_deref() == Some("23505")
);
if unique_violation {
let winner: Option<Photo> =
sqlx::query_as("select * from photos where album_id = $1 and sha256 = $2")
.bind(album_id)
.bind(&sha256)
.fetch_optional(&state.db)
.await?;
if let Some(winner) = winner {
return Ok(Json(winner));
}
}
return Err(e.into());
}
};
if let Err(e) = tx.commit().await {
@@ -133,6 +174,21 @@ pub async fn upload(
Ok(Json(photo))
}
/// Client-side dedup support: lets the uploader skip transferring files whose
/// content already exists in the album.
pub async fn by_hash(
State(state): State<AppState>,
Path((album_id, sha256)): Path<(Uuid, String)>,
) -> ApiResult<Json<Photo>> {
let photo: Option<Photo> =
sqlx::query_as("select * from photos where album_id = $1 and sha256 = $2")
.bind(album_id)
.bind(&sha256)
.fetch_optional(&state.db)
.await?;
photo.map(Json).ok_or_else(ApiError::not_found)
}
pub async fn delete(
State(state): State<AppState>,
Path(photo_id): Path<Uuid>,