feat(mongodb): bson dumps and database backup/restore

crates/dbx-web/Cargo.toml

@@ -34,10 +34,14 @@ tracing = "0.1"
 async-stream = "0.3"
 futures = "0.3"
 rustls = { version = "0.23", features = ["aws-lc-rs"] }
-tokio-util = "0.7"
+tokio-util = { version = "0.7", features = ["io"] }
 reqwest = { version = "0.12", default-features = false, features = ["json", "stream", "rustls-tls"] }
 chrono = "0.4"
 aes-gcm = "0.10"
 base64 = "0.22"
 pbkdf2 = "0.12"
 sha2 = "0.10"
+tempfile = "3"
+
+[dev-dependencies]
+reqwest = { version = "0.12", default-features = false, features = ["multipart"] }

crates/dbx-web/src/main.rs

@@ -988,6 +988,25 @@ async fn main() {
         .route("/mongo/export/progress/{exportId}", get(routes::mongodb_import_export::export_progress))
         .route("/mongo/export/download/{exportId}", get(routes::mongodb_import_export::export_download))
         .route("/mongo/export/cancel", post(routes::mongodb_import_export::cancel_export))
+        .route("/mongo/dump/catalog", post(routes::mongodb_dump::catalog))
+        .route(
+            "/mongo/dump/source",
+            post(routes::mongodb_dump::prepare_source).layer(DefaultBodyLimit::max(
+                routes::table_import::import_request_body_limit_for_upload(web_body_limit_bytes()),
+            )),
+        )
+        .route("/mongo/dump/source/release", post(routes::mongodb_dump::release_source))
+        .route("/mongo/dump/upload-limit", get(routes::mongodb_dump::upload_limit))
+        .route(
+            "/mongo/dump/source/upload",
+            post(routes::mongodb_dump::upload_restore_source).layer(DefaultBodyLimit::max(
+                routes::table_import::import_request_body_limit_for_upload(web_body_limit_bytes()),
+            )),
+        )
+        .route("/mongo/dump/export", post(routes::mongodb_dump::start_dump))
+        .route("/mongo/dump/restore", post(routes::mongodb_dump::start_restore))
+        .route("/mongo/dump/progress/{taskId}", get(routes::mongodb_dump::progress))
+        .route("/mongo/dump/cancel", post(routes::mongodb_dump::cancel))
         // History
         .route("/history", get(routes::history::load_history).delete(routes::history::clear_history))
         .route("/history/save", post(routes::history::save_history))

crates/dbx-web/src/routes/mod.rs

@@ -17,6 +17,7 @@ pub mod jdbc;
 pub mod layout;
 pub mod mcp_policy;
 pub mod mongo;
+pub mod mongodb_dump;
 pub mod mongodb_import_export;
 #[cfg(feature = "mq-admin")]
 pub mod mq;

crates/dbx-web/src/routes/mongodb_dump.rs

@@ -0,0 +1,402 @@
+use std::path::{Component, Path as FilePath, PathBuf};
+use std::sync::Arc;
+use std::time::Duration;
+
+use axum::{
+    extract::{Multipart, Path, State},
+    response::Sse,
+    Json,
+};
+use dbx_core::{
+    mongodb_dump::{
+        self, MongoDatabaseDumpProgress, MongoDatabaseDumpRequest, MongoDatabaseRestoreRequest, MongoDumpFormat,
+        MongoRestoreSourceRequest,
+    },
+    transfer,
+};
+use serde::Deserialize;
+use tokio::io::AsyncWriteExt;
+
+use super::export_download::export_download_filename;
+use crate::{
+    error::AppError,
+    sse::{TransferProgressChannel, TransferReplayEventKind},
+    state::{WebExportFile, WebState},
+};
+
+#[derive(Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct CatalogRequest {
+    connection_id: String,
+    database: String,
+}
+
+pub async fn catalog(
+    State(state): State<Arc<WebState>>,
+    Json(request): Json<CatalogRequest>,
+) -> Result<Json<serde_json::Value>, AppError> {
+    let catalog = mongodb_dump::inspect_mongodb_database_dump(&state.app, &request.connection_id, &request.database)
+        .await
+        .map_err(AppError::from)?;
+    Ok(Json(serde_json::to_value(catalog).map_err(|e| AppError::from(e.to_string()))?))
+}
+
+fn upload_path(root: &FilePath, filename: &str) -> Result<PathBuf, AppError> {
+    let path = FilePath::new(filename);
+    if filename.is_empty()
+        || filename.contains(['\\', ':', '\0'])
+        || path.components().any(|c| !matches!(c, Component::Normal(_)))
+    {
+        return Err(AppError::from("Invalid dump upload path".to_string()));
+    }
+    if path.components().count() > 4 {
+        return Err(AppError::from("Dump directory nesting is too deep".to_string()));
+    }
+    Ok(root.join(path))
+}
+
+pub async fn prepare_source(
+    State(state): State<Arc<WebState>>,
+    multipart: Multipart,
+) -> Result<Json<serde_json::Value>, AppError> {
+    receive_source(state, multipart, false).await
+}
+
+pub async fn upload_restore_source(
+    State(state): State<Arc<WebState>>,
+    multipart: Multipart,
+) -> Result<Json<serde_json::Value>, AppError> {
+    receive_source(state, multipart, true).await
+}
+
+pub async fn upload_limit() -> Json<usize> {
+    Json(crate::web_body_limit_bytes())
+}
+
+async fn field_text(mut field: axum::extract::multipart::Field<'_>, limit: usize) -> Result<String, AppError> {
+    let mut bytes = Vec::new();
+    while let Some(chunk) = field.chunk().await.map_err(|e| AppError::bad_request(e.to_string()))? {
+        if bytes.len().saturating_add(chunk.len()) > limit {
+            return Err(AppError::bad_request("Dump metadata field is too large"));
+        }
+        bytes.extend_from_slice(&chunk);
+    }
+    String::from_utf8(bytes).map_err(|e| AppError::bad_request(e.to_string()))
+}
+
+async fn receive_source(
+    state: Arc<WebState>,
+    mut multipart: Multipart,
+    acquiring: bool,
+) -> Result<Json<serde_json::Value>, AppError> {
+    let tmp = state.data_dir.join("tmp");
+    tokio::fs::create_dir_all(&tmp).await.map_err(|e| AppError::from(e.to_string()))?;
+    let upload =
+        tempfile::Builder::new().prefix("mongo-upload-").tempdir_in(&tmp).map_err(|e| AppError::from(e.to_string()))?;
+    let mut format = None;
+    let mut gzip = false;
+    let mut paths = Vec::new();
+    let mut bytes = 0usize;
+    let mut manifest: Option<Vec<mongodb_dump::MongoDumpFile>> = None;
+    let mut request: Option<MongoDatabaseRestoreRequest> = None;
+    let mut expected_upload = std::collections::HashMap::<String, u64>::new();
+    while let Some(mut field) = multipart.next_field().await.map_err(|e| AppError::from(e.to_string()))? {
+        match field.name().unwrap_or("") {
+            "file" => {
+                if paths.len() >= 100_000 {
+                    return Err(AppError::from("Too many dump files".to_string()));
+                }
+                let filename =
+                    field.file_name().ok_or_else(|| AppError::from("Missing dump filename".to_string()))?.to_string();
+                if format.is_none() {
+                    return Err(AppError::bad_request("Dump format must precede files"));
+                }
+                if !acquiring
+                    && format == Some(MongoDumpFormat::Directory)
+                    && !filename.ends_with(".metadata.json")
+                    && !filename.ends_with(".metadata.json.gz")
+                {
+                    return Err(AppError::bad_request(
+                        "Directory preview accepts metadata only; upload BSON after confirmation",
+                    ));
+                }
+                if acquiring && request.is_none() {
+                    return Err(AppError::bad_request("Restore plan must precede files"));
+                }
+                let expected_size = if acquiring {
+                    Some(
+                        expected_upload
+                            .remove(&filename)
+                            .ok_or_else(|| AppError::bad_request("File is not in the selected restore plan"))?,
+                    )
+                } else {
+                    None
+                };
+                let mut file_bytes = 0u64;
+                let path = upload_path(upload.path(), &filename)?;
+                tokio::fs::create_dir_all(path.parent().unwrap()).await.map_err(|e| AppError::from(e.to_string()))?;
+                let mut output = tokio::fs::OpenOptions::new()
+                    .write(true)
+                    .create_new(true)
+                    .open(&path)
+                    .await
+                    .map_err(|e| AppError::from(e.to_string()))?;
+                while let Some(chunk) = field.chunk().await.map_err(|e| AppError::from(e.to_string()))? {
+                    if let Some(request) = &request {
+                        if transfer::is_cancelled(&request.task_id).await {
+                            return Err(AppError::bad_request("MongoDB dump/restore cancelled"));
+                        }
+                    }
+                    bytes = bytes.saturating_add(chunk.len());
+                    file_bytes = file_bytes.saturating_add(chunk.len() as u64);
+                    if expected_size.is_some_and(|size| file_bytes > size) {
+                        return Err(AppError::bad_request("File size changed after preview"));
+                    }
+                    let limit = if !acquiring && format == Some(MongoDumpFormat::Directory) {
+                        32 * 1024 * 1024
+                    } else {
+                        crate::web_body_limit_bytes()
+                    };
+                    if bytes > limit {
+                        let mut error =
+                            AppError::from(format!("Dump upload exceeds the configured {} byte limit", limit));
+                        error.status = axum::http::StatusCode::PAYLOAD_TOO_LARGE;
+                        return Err(error);
+                    }
+                    output.write_all(&chunk).await.map_err(|e| AppError::from(e.to_string()))?;
+                }
+                output.flush().await.map_err(|e| AppError::from(e.to_string()))?;
+                if expected_size.is_some_and(|size| file_bytes != size) {
+                    return Err(AppError::bad_request("File size changed after preview"));
+                }
+                paths.push((filename, path));
+            }
+            "format" => {
+                let value = field_text(field, 32).await?;
+                format = Some(
+                    serde_json::from_value::<MongoDumpFormat>(serde_json::Value::String(value))
+                        .map_err(|e| AppError::from(e.to_string()))?,
+                );
+            }
+            "gzip" => {
+                let value = field_text(field, 16).await?;
+                gzip = value.parse::<bool>().map_err(|_| AppError::from("Invalid gzip option".to_string()))?;
+            }
+            "manifest" if !acquiring => {
+                let value = field_text(field, 16 * 1024 * 1024).await?;
+                manifest = Some(serde_json::from_str(&value).map_err(|e| AppError::bad_request(e.to_string()))?);
+            }
+            "request" if acquiring => {
+                let value = field_text(field, 1024 * 1024).await?;
+                if request.is_some() {
+                    return Err(AppError::bad_request("Duplicate restore plan"));
+                }
+                request = Some(serde_json::from_str(&value).map_err(|e| AppError::bad_request(e.to_string()))?);
+                expected_upload = mongodb_dump::mongodb_restore_upload_files(request.as_ref().unwrap())
+                    .map_err(AppError::from)?
+                    .into_iter()
+                    .map(|f| (f.path, f.size_bytes))
+                    .collect();
+            }
+            _ => return Err(AppError::from("Unexpected dump upload field".to_string())),
+        }
+    }
+    let format = format.ok_or_else(|| AppError::from("Dump format is required".to_string()))?;
+    if paths.is_empty() && manifest.is_none() {
+        return Err(AppError::from("No dump files uploaded".to_string()));
+    }
+    if acquiring {
+        if !expected_upload.is_empty() {
+            return Err(AppError::bad_request("Selected restore files are missing"));
+        }
+        let request = request.ok_or_else(|| AppError::bad_request("Missing restore plan"))?;
+        if transfer::is_cancelled(&request.task_id).await {
+            return Err(AppError::bad_request("MongoDB dump/restore cancelled"));
+        }
+        let preview = mongodb_dump::attach_mongodb_restore_upload(&request, upload).await.map_err(AppError::from)?;
+        return Ok(Json(serde_json::to_value(preview).map_err(|e| AppError::from(e.to_string()))?));
+    }
+    if format == MongoDumpFormat::Directory {
+        let manifest = manifest.ok_or_else(|| AppError::bad_request("Directory preview requires a file manifest"))?;
+        let expected = manifest
+            .iter()
+            .filter(|f| f.path.ends_with(".metadata.json") || f.path.ends_with(".metadata.json.gz"))
+            .collect::<Vec<_>>();
+        if expected.len() != paths.len()
+            || paths.iter().any(|(name, path)| {
+                !expected
+                    .iter()
+                    .any(|f| f.path == *name && std::fs::metadata(path).is_ok_and(|m| m.len() == f.size_bytes))
+            })
+        {
+            return Err(AppError::bad_request("Uploaded metadata does not match the file manifest"));
+        }
+        let preview =
+            mongodb_dump::prepare_mongodb_directory_catalog(upload, manifest, gzip).await.map_err(AppError::from)?;
+        return Ok(Json(serde_json::to_value(preview).map_err(|e| AppError::from(e.to_string()))?));
+    }
+    let path = match format {
+        MongoDumpFormat::Archive if paths.len() == 1 => paths[0].1.clone(),
+        MongoDumpFormat::Archive => return Err(AppError::from("Archive restore accepts one file".to_string())),
+        MongoDumpFormat::Directory => {
+            if paths.iter().any(|(name, _)| FilePath::new(name).components().count() < 2) {
+                return Err(AppError::from("Directory uploads require database-relative paths".to_string()));
+            }
+            upload.path().to_owned()
+        }
+    };
+    let preview = mongodb_dump::prepare_owned_mongodb_restore_source(
+        MongoRestoreSourceRequest { path: path.to_string_lossy().into_owned(), format, gzip },
+        upload,
+    )
+    .await
+    .map_err(AppError::from)?;
+    Ok(Json(serde_json::to_value(preview).map_err(|e| AppError::from(e.to_string()))?))
+}
+
+#[derive(Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct SourceRef {
+    source_ref: String,
+}
+pub async fn release_source(Json(request): Json<SourceRef>) -> Json<serde_json::Value> {
+    Json(serde_json::json!({ "released": mongodb_dump::release_mongodb_restore_source(&request.source_ref) }))
+}
+
+#[derive(Deserialize)]
+pub struct DumpWrapper {
+    request: MongoDatabaseDumpRequest,
+}
+#[derive(Deserialize)]
+pub struct RestoreWrapper {
+    request: MongoDatabaseRestoreRequest,
+}
+
+enum Task {
+    Dump(MongoDatabaseDumpRequest, WebExportFile),
+    Restore(MongoDatabaseRestoreRequest),
+}
+
+async fn start_task(state: Arc<WebState>, task: Task) -> Result<Json<serde_json::Value>, AppError> {
+    let id = match &task {
+        Task::Dump(request, _) => &request.task_id,
+        Task::Restore(request) => &request.task_id,
+    }
+    .clone();
+    let key = format!("mongo-database:{id}");
+    let channel = Arc::new(TransferProgressChannel::new());
+    {
+        let mut channels = state.transfer_progress_channels.write().await;
+        if channels.contains_key(&key) {
+            return Err(AppError::from("Database dump/restore task already exists".to_string()));
+        }
+        channels.insert(key.clone(), channel.clone());
+    }
+    let task_id = id.clone();
+    dbx_core::export_runtime::spawn_export_task(async move {
+        let mut latest = MongoDatabaseDumpProgress::initial(&task_id);
+        let mut callback = |progress: MongoDatabaseDumpProgress| {
+            latest = progress;
+            if latest.status == "running" {
+                if let Ok(json) = serde_json::to_string(&latest) {
+                    channel.send(json, TransferReplayEventKind::Progress);
+                }
+            }
+        };
+        let result = match task {
+            Task::Dump(request, file) => {
+                let result = mongodb_dump::dump_mongodb_database(
+                    &state.app,
+                    &request,
+                    |id| {
+                        let id = id.to_string();
+                        Box::pin(async move { transfer::is_cancelled(&id).await })
+                    },
+                    &mut callback,
+                )
+                .await;
+                if result.is_ok() {
+                    state.export_files.write().await.insert(task_id.clone(), file);
+                }
+                result
+            }
+            Task::Restore(request) => {
+                mongodb_dump::restore_mongodb_database(
+                    &state.app,
+                    &request,
+                    |id| {
+                        let id = id.to_string();
+                        Box::pin(async move { transfer::is_cancelled(&id).await })
+                    },
+                    &mut callback,
+                )
+                .await
+            }
+        };
+        if let Err(error) = result {
+            latest.error_message = Some(error);
+            if latest.status == "running" {
+                latest.status = "error".into();
+                latest.phase = "done".into();
+            }
+        }
+        if let Ok(json) = serde_json::to_string(&latest) {
+            channel.send(json, TransferReplayEventKind::Terminal);
+        }
+        transfer::clear_cancelled(&task_id).await;
+        tokio::spawn(async move {
+            tokio::time::sleep(Duration::from_secs(60)).await;
+            state.transfer_progress_channels.write().await.remove(&key);
+        });
+    });
+    Ok(Json(serde_json::json!({ "taskId": id })))
+}
+
+pub async fn start_dump(
+    State(state): State<Arc<WebState>>,
+    Json(body): Json<DumpWrapper>,
+) -> Result<Json<serde_json::Value>, AppError> {
+    let mut request = body.request;
+    if request.format != MongoDumpFormat::Archive {
+        return Err(AppError::from("Web database exports use MongoDB archive format".to_string()));
+    }
+    let tmp = state.data_dir.join("tmp");
+    tokio::fs::create_dir_all(&tmp).await.map_err(|e| AppError::from(e.to_string()))?;
+    let ext = if request.gzip { "archive.gz" } else { "archive" };
+    let filename = export_download_filename(&request.file_path, &request.database, ext);
+    request.file_path =
+        tmp.join(format!("mongo_database_{}.{ext}", uuid::Uuid::new_v4())).to_string_lossy().into_owned();
+    let file = WebExportFile { file_path: request.file_path.clone(), download_filename: filename, format: ext.into() };
+    start_task(state, Task::Dump(request, file)).await
+}
+
+pub async fn start_restore(
+    State(state): State<Arc<WebState>>,
+    Json(body): Json<RestoreWrapper>,
+) -> Result<Json<serde_json::Value>, AppError> {
+    start_task(state, Task::Restore(body.request)).await
+}
+
+pub async fn progress(
+    State(state): State<Arc<WebState>>,
+    Path(id): Path<String>,
+) -> Result<Sse<impl futures::Stream<Item = Result<axum::response::sse::Event, std::convert::Infallible>>>, AppError> {
+    let channel = state
+        .transfer_progress_channels
+        .read()
+        .await
+        .get(&format!("mongo-database:{id}"))
+        .cloned()
+        .ok_or_else(|| AppError::from("Database task not found".to_string()))?;
+    Ok(crate::sse::sse_from_transfer_channel(channel))
+}
+
+#[derive(Deserialize)]
+#[serde(rename_all = "camelCase")]
+pub struct CancelRequest {
+    task_id: String,
+}
+pub async fn cancel(Json(request): Json<CancelRequest>) -> Json<serde_json::Value> {
+    transfer::set_cancelled(&request.task_id).await;
+    Json(serde_json::json!({ "cancelled": true }))
+}

crates/dbx-web/src/routes/mongodb_import_export.rs

@@ -17,13 +17,40 @@ use futures::stream::Stream;
 use futures::StreamExt;
 use serde::Deserialize;
 use tokio::io::AsyncWriteExt;
+use tokio_util::io::ReaderStream;
 
 use crate::error::AppError;
 use crate::routes::export_download::{attachment_content_disposition, export_download_filename};
+use crate::sse::{TransferProgressChannel, TransferReplayEventKind};
 use crate::state::{WebExportFile, WebState};
 
 const MONGO_IMPORT_PROGRESS_TTL: Duration = Duration::from_secs(30);
 
+struct RemoveFileOnDrop(PathBuf);
+
+impl Drop for RemoveFileOnDrop {
+    fn drop(&mut self) {
+        let _ = std::fs::remove_file(&self.0);
+    }
+}
+
+struct ExportDownloadStream {
+    // Fields are dropped in declaration order, so Windows closes the file before deleting it.
+    chunks: ReaderStream<tokio::fs::File>,
+    _cleanup: RemoveFileOnDrop,
+}
+
+impl Stream for ExportDownloadStream {
+    type Item = std::io::Result<Bytes>;
+
+    fn poll_next(
+        self: std::pin::Pin<&mut Self>,
+        cx: &mut std::task::Context<'_>,
+    ) -> std::task::Poll<Option<Self::Item>> {
+        std::pin::Pin::new(&mut self.get_mut().chunks).poll_next(cx)
+    }
+}
+
 fn initial_import_progress(import_id: &str, started_at: Instant) -> MongoImportProgress {
     MongoImportProgress {
         import_id: import_id.to_string(),
@@ -153,11 +180,11 @@ pub async fn preview_import(
 
     if let Some((source_ref, file_path)) = uploaded_file {
         let file_path_str = file_path.to_string_lossy().to_string();
-        let format = match format.or_else(|| mongodb_import_export::format_from_path(&file_path_str).ok()) {
+        let format = match format {
             Some(format) => format,
             None => {
                 cleanup_uploaded_import_path(&file_path).await;
-                return Err(AppError::from("Unsupported MongoDB import file type".to_string()));
+                return Err(AppError::from("MongoDB import format is required".to_string()));
             }
         };
         let preview = mongodb_import_export::preview_mongodb_import_file(&MongoImportPreviewRequest {
@@ -269,7 +296,7 @@ pub async fn execute_import(
     let state_clone = state.clone();
     tokio::spawn(async move {
         let tx_clone = tx.clone();
-        let _result = mongodb_import_export::import_mongodb_file_core(
+        let result = mongodb_import_export::import_mongodb_file_core(
             &app,
             &req,
             |id| {
@@ -279,6 +306,17 @@ pub async fn execute_import(
             |progress| send_import_progress(&tx_clone, &progress),
         )
         .await;
+        if let Err(error) = result {
+            let current = tx.borrow().clone();
+            if let Ok(mut progress) = serde_json::from_str::<MongoImportProgress>(&current) {
+                if progress.status == MongoImportStatus::Running {
+                    progress.phase = MongoImportPhase::Done;
+                    progress.status = MongoImportStatus::Error;
+                    progress.error_message = Some(error.display_message());
+                    send_import_progress(&tx, &progress);
+                }
+            }
+        }
         cleanup_uploaded_import_source(&req.file_path).await;
         schedule_import_progress_cleanup(state_clone, req.import_id.clone());
     });
@@ -402,26 +440,35 @@ pub async fn start_export(
     let ext = match req.format {
         MongoExportFormat::Csv => "csv",
         MongoExportFormat::Ndjson => "ndjson",
+        MongoExportFormat::Bson if req.gzip => "bson.gz",
+        MongoExportFormat::Bson => "bson",
     };
-    let tmp_file = tmp_dir.join(format!("mongo_export_{export_id}.{ext}"));
+    let tmp_file = tmp_dir.join(format!("mongo_export_{}.{ext}", uuid::Uuid::new_v4()));
     let file_path = tmp_file.to_string_lossy().to_string();
     let download_filename = export_download_filename(&req.file_path, &req.collection, ext);
     req.file_path = file_path.clone();
 
-    state
-        .export_files
-        .write()
-        .await
-        .insert(export_id.clone(), WebExportFile { file_path, download_filename, format: ext.to_string() });
-
+    let export_file = WebExportFile { file_path, download_filename, format: ext.to_string() };
+    let channel_key = format!("mongo-export:{export_id}");
     let tx = {
-        let mut channels = state.sse_channels.write().await;
-        channels.entry(export_id.clone()).or_insert_with(|| tokio::sync::broadcast::channel::<String>(256).0).clone()
+        let mut channels = state.transfer_progress_channels.write().await;
+        let channel = Arc::new(TransferProgressChannel::new());
+        channels.insert(channel_key.clone(), channel.clone());
+        channel
     };
 
     let app = state.app.clone();
     let state_clone = state.clone();
     dbx_core::export_runtime::spawn_export_task(async move {
+        let mut latest = MongoExportProgress {
+            export_id: req.export_id.clone(),
+            status: MongoExportStatus::Running,
+            documents_read: 0,
+            bytes_written: 0,
+            total_documents: None,
+            error_message: None,
+            elapsed_ms: 0,
+        };
         let result = mongodb_import_export::export_mongodb_query_core(
             &app,
             &req,
@@ -430,32 +477,38 @@ pub async fn start_export(
                 Box::pin(async move { transfer::is_cancelled(&id).await })
             },
             |progress| {
-                if let Ok(json) = serde_json::to_string(&progress) {
-                    let _ = tx.send(json);
+                latest = progress;
+                if latest.status == MongoExportStatus::Running {
+                    if let Ok(json) = serde_json::to_string(&latest) {
+                        tx.send(json, TransferReplayEventKind::Progress);
+                    }
                 }
             },
         )
         .await;
 
-        if let Err(error) = result {
-            let _ = tokio::fs::remove_file(&req.file_path).await;
-            state_clone.export_files.write().await.remove(&req.export_id);
-            let progress = MongoExportProgress {
-                export_id: req.export_id.clone(),
-                status: MongoExportStatus::Error,
-                documents_read: 0,
-                bytes_written: 0,
-                total_documents: None,
-                error_message: Some(error),
-                elapsed_ms: 0,
-            };
-            if let Ok(json) = serde_json::to_string(&progress) {
-                let _ = tx.send(json);
+        match result {
+            Ok(_) => {
+                // The download must exist before a terminal event can reach the browser.
+                state_clone.export_files.write().await.insert(req.export_id.clone(), export_file);
+            }
+            Err(error) => {
+                let _ = tokio::fs::remove_file(&req.file_path).await;
+                if latest.status != MongoExportStatus::Cancelled {
+                    latest.status = MongoExportStatus::Error;
+                }
+                latest.error_message = Some(error);
             }
         }
+        if let Ok(json) = serde_json::to_string(&latest) {
+            tx.send(json, TransferReplayEventKind::Terminal);
+        }
+        transfer::clear_cancelled(&req.export_id).await;
 
-        tokio::time::sleep(Duration::from_secs(5)).await;
-        state_clone.remove_sse_channel(&req.export_id).await;
+        tokio::spawn(async move {
+            tokio::time::sleep(Duration::from_secs(30)).await;
+            state_clone.transfer_progress_channels.write().await.remove(&channel_key);
+        });
     });
 
     Ok(Json(serde_json::json!({ "exportId": export_id })))
@@ -465,12 +518,14 @@ pub async fn export_progress(
     State(state): State<Arc<WebState>>,
     Path(export_id): Path<String>,
 ) -> Result<Sse<impl Stream<Item = Result<axum::response::sse::Event, std::convert::Infallible>>>, AppError> {
-    let tx = {
-        let mut channels = state.sse_channels.write().await;
-        channels.entry(export_id).or_insert_with(|| tokio::sync::broadcast::channel::<String>(256).0).clone()
-    };
-    let rx = tx.subscribe();
-    Ok(crate::sse::sse_from_channel(rx))
+    let channel = state
+        .transfer_progress_channels
+        .read()
+        .await
+        .get(&format!("mongo-export:{export_id}"))
+        .cloned()
+        .ok_or_else(|| AppError::from("Export not found".to_string()))?;
+    Ok(crate::sse::sse_from_transfer_channel(channel))
 }
 
 pub async fn cancel_export(
@@ -491,17 +546,27 @@ pub async fn export_download(
         .await
         .remove(&export_id)
         .ok_or_else(|| AppError::from("Export file not found".to_string()))?;
-    let data = tokio::fs::read(&export_file.file_path).await.map_err(|e| AppError::from(e.to_string()))?;
-    let _ = tokio::fs::remove_file(&export_file.file_path).await;
+    export_file_response(export_file).await
+}
+
+pub(crate) async fn export_file_response(export_file: WebExportFile) -> Result<Response, AppError> {
+    let file = tokio::fs::File::open(&export_file.file_path).await.map_err(|e| AppError::from(e.to_string()))?;
     let content_type = match export_file.format.as_str() {
         "csv" => "text/csv; charset=utf-8",
-        _ => "application/x-ndjson; charset=utf-8",
+        "ndjson" => "application/x-ndjson; charset=utf-8",
+        "bson.gz" | "archive.gz" => "application/gzip",
+        "bson" => "application/bson",
+        _ => "application/octet-stream",
+    };
+    let download = ExportDownloadStream {
+        chunks: ReaderStream::new(file),
+        _cleanup: RemoveFileOnDrop(PathBuf::from(&export_file.file_path)),
     };
     Response::builder()
         .status(StatusCode::OK)
         .header(header::CONTENT_TYPE, content_type)
         .header(header::CONTENT_DISPOSITION, attachment_content_disposition(&export_file.download_filename))
-        .body(Body::from(data))
+        .body(Body::from_stream(download))
         .map_err(|error| AppError::from(error.to_string()))
 }
 
@@ -509,6 +574,152 @@ pub async fn export_download(
 mod tests {
     use super::*;
 
+    #[tokio::test]
+    async fn download_keeps_file_until_body_is_consumed_or_abandoned() {
+        for extension in ["bson", "bson.gz"] {
+            for consume in [true, false] {
+                let directory = tempfile::tempdir().unwrap();
+                let path = directory.path().join(format!("records.{extension}"));
+                let bytes = vec![42u8; 100_000];
+                std::fs::write(&path, &bytes).unwrap();
+                let response = export_file_response(WebExportFile {
+                    file_path: path.to_str().unwrap().into(),
+                    download_filename: format!("records.{extension}"),
+                    format: extension.into(),
+                })
+                .await
+                .unwrap_or_else(|error| panic!("{}", error.message));
+                assert!(path.exists(), "download file was removed before streaming");
+                assert_eq!(
+                    response.headers()[header::CONTENT_TYPE],
+                    if extension == "bson" { "application/bson" } else { "application/gzip" }
+                );
+                assert!(response.headers()[header::CONTENT_DISPOSITION]
+                    .to_str()
+                    .unwrap()
+                    .contains(&format!("records.{extension}")));
+                if consume {
+                    assert_eq!(axum::body::to_bytes(response.into_body(), bytes.len()).await.unwrap(), bytes);
+                } else {
+                    let mut stream = response.into_body().into_data_stream();
+                    assert!(!stream.next().await.unwrap().unwrap().is_empty());
+                    drop(stream);
+                }
+                assert!(!path.exists(), "download file leaked after the response ended");
+            }
+        }
+    }
+
+    #[tokio::test]
+    async fn export_failure_and_cancellation_are_replayed_to_late_subscribers() {
+        use axum::response::IntoResponse;
+        use dbx_core::{connection::AppState, storage::Storage};
+
+        for cancelled in [false, true] {
+            let directory = tempfile::tempdir().unwrap();
+            let storage = Storage::open(&directory.path().join("storage.db")).await.unwrap();
+            let state = Arc::new(WebState::for_tests(Arc::new(AppState::new(storage)), directory.path().into()));
+            let id = uuid::Uuid::new_v4().to_string();
+            let expected = if cancelled { MongoExportStatus::Cancelled } else { MongoExportStatus::Error };
+            if cancelled {
+                transfer::set_cancelled(&id).await;
+            }
+            let Json(started) = start_export(
+                State(state.clone()),
+                Json(StartExportRequest {
+                    request: MongoExportRequest {
+                        export_id: id.clone(),
+                        connection_id: "missing".into(),
+                        database: "test".into(),
+                        collection: "empty".into(),
+                        filter: None,
+                        projection: None,
+                        sort: None,
+                        collation: None,
+                        format: MongoExportFormat::Bson,
+                        gzip: false,
+                        include_header: false,
+                        file_path: "empty.bson".into(),
+                        execution_id: None,
+                    },
+                }),
+            )
+            .await
+            .unwrap_or_else(|error| panic!("{}", error.message));
+            assert_eq!(started["exportId"], id);
+            let key = format!("mongo-export:{id}");
+            tokio::time::timeout(Duration::from_secs(5), async {
+                loop {
+                    let latest = state.transfer_progress_channels.read().await.get(&key).unwrap().latest();
+                    if latest.is_some_and(|json| {
+                        serde_json::from_str::<MongoExportProgress>(&json).unwrap().status == expected
+                    }) {
+                        break;
+                    }
+                    tokio::time::sleep(Duration::from_millis(10)).await;
+                }
+            })
+            .await
+            .unwrap();
+            assert!(!state.export_files.read().await.contains_key(&id));
+            let response = export_progress(State(state.clone()), Path(id))
+                .await
+                .unwrap_or_else(|error| panic!("{}", error.message));
+            let mut stream = response.into_response().into_body().into_data_stream();
+            let chunk = tokio::time::timeout(Duration::from_secs(1), stream.next()).await.unwrap().unwrap().unwrap();
+            assert!(String::from_utf8_lossy(&chunk)
+                .contains(&format!("\"status\":{}", serde_json::to_string(&expected).unwrap())));
+        }
+    }
+
+    #[tokio::test]
+    async fn early_import_failure_finishes_progress_and_cleans_source() {
+        use dbx_core::{connection::AppState, storage::Storage};
+
+        let directory = tempfile::tempdir().unwrap();
+        let storage = Storage::open(&directory.path().join("storage.db")).await.unwrap();
+        let state = Arc::new(WebState::for_tests(Arc::new(AppState::new(storage)), directory.path().into()));
+        let upload_dir = import_upload_dir(directory.path());
+        std::fs::create_dir_all(&upload_dir).unwrap();
+        let path = upload_dir.join("empty.bson");
+        std::fs::write(&path, []).unwrap();
+        let id = uuid::Uuid::new_v4().to_string();
+        let Json(started) = execute_import(
+            State(state.clone()),
+            Json(ExecuteImportWrapper {
+                request: MongoImportRequest {
+                    import_id: id.clone(),
+                    connection_id: "missing".into(),
+                    database: "test".into(),
+                    collection: "empty".into(),
+                    file_path: path.to_str().unwrap().into(),
+                    source_ref: None,
+                    format: mongodb_import_export::MongoImportFormat::Bson,
+                    parse_options: Default::default(),
+                    batch_size: 1,
+                    execution_id: None,
+                },
+            }),
+        )
+        .await
+        .unwrap_or_else(|error| panic!("{}", error.message));
+        assert_eq!(started["importId"], id);
+        tokio::time::timeout(Duration::from_secs(5), async {
+            loop {
+                let progress = state.table_import_channels.read().await.get(&id).unwrap().borrow().clone();
+                let progress: MongoImportProgress = serde_json::from_str(&progress).unwrap();
+                if progress.status == MongoImportStatus::Error && !path.exists() {
+                    assert_eq!(progress.phase, MongoImportPhase::Done);
+                    assert!(progress.error_message.unwrap().contains("Batch size"));
+                    break;
+                }
+                tokio::time::sleep(Duration::from_millis(10)).await;
+            }
+        })
+        .await
+        .unwrap();
+    }
+
     #[test]
     fn source_ref_rejects_path_traversal() {
         let data_dir = std::env::temp_dir().join(format!("dbx-mongo-import-data-{}", uuid::Uuid::new_v4()));

crates/dbx-web/tests/mongodb_dump_http.rs

@@ -0,0 +1,118 @@
+use futures::StreamExt;
+use reqwest::multipart::{Form, Part};
+use std::process::Stdio;
+use std::time::Duration;
+
+#[tokio::test]
+async fn database_dump_upload_and_progress_http_contract() {
+    let directory = tempfile::tempdir().unwrap();
+    let port = std::net::TcpListener::bind("127.0.0.1:0").unwrap().local_addr().unwrap().port();
+    let log_path = directory.path().join("server.log");
+    let mut server = tokio::process::Command::new(env!("CARGO_BIN_EXE_dbx-web"))
+        .env("DBX_DATA_DIR", directory.path())
+        .env("DBX_PORT", port.to_string())
+        .env("DBX_DISABLE_PASSWORD", "1")
+        .kill_on_drop(true)
+        .stdout(Stdio::null())
+        .stderr(std::fs::File::create(&log_path).unwrap())
+        .spawn()
+        .unwrap();
+    let base = format!("http://127.0.0.1:{port}/api/mongo/dump");
+    let client = reqwest::Client::new();
+    tokio::time::timeout(Duration::from_secs(20), async {
+        loop {
+            if client.get(format!("http://127.0.0.1:{port}/api/auth/check")).send().await.is_ok() {
+                break;
+            }
+            if let Some(status) = server.try_wait().unwrap() {
+                panic!("Web server exited {status}: {}", std::fs::read_to_string(&log_path).unwrap());
+            }
+            tokio::time::sleep(Duration::from_millis(50)).await;
+        }
+    })
+    .await
+    .unwrap();
+
+    let bad = Form::new()
+        .text("format", "archive")
+        .text("gzip", "false")
+        .part("file", Part::bytes(vec![0]).file_name("../escape.archive"));
+    let response = client.post(format!("{base}/source")).multipart(bad).send().await.unwrap();
+    assert!(!response.status().is_success());
+    assert!(response.text().await.unwrap().contains("Invalid dump upload path"));
+
+    let form = Form::new()
+        .text("format", "directory")
+        .text("gzip", "false")
+        .text("manifest", serde_json::json!([
+            {"path":"source/records.bson","sizeBytes":0},
+            {"path":"source/records.metadata.json","sizeBytes":r#"{"collectionName":"records","options":{},"indexes":[],"type":"collection"}"#.len()}
+        ]).to_string())
+        .part(
+            "file",
+            Part::text(r#"{"collectionName":"records","options":{},"indexes":[],"type":"collection"}"#)
+                .file_name("source/records.metadata.json"),
+        );
+    let response = client.post(format!("{base}/source")).multipart(form).send().await.unwrap();
+    assert!(response.status().is_success(), "{}", response.text().await.unwrap());
+    let preview: serde_json::Value = response.json().await.unwrap();
+    assert_eq!(preview["databases"], serde_json::json!(["source"]));
+    assert_eq!(preview["collections"][0]["name"], "records");
+    assert!(preview["collections"][0]["documents"].is_null());
+    assert_eq!(preview["collections"][0]["sourceFiles"].as_array().unwrap().len(), 2);
+    let source_ref = preview["sourceRef"].as_str().unwrap();
+    let request = serde_json::json!({ "taskId":"acquire", "connectionId":"missing", "database":"target", "sourceDatabase":"source", "sourceRef":source_ref, "collections":["records"] });
+    let acquisition = Form::new()
+        .text("format", "directory")
+        .text("gzip", "false")
+        .text("request", request.to_string())
+        .part("file", Part::bytes(Vec::new()).file_name("source/records.bson"))
+        .part(
+            "file",
+            Part::text(r#"{"collectionName":"records","options":{},"indexes":[],"type":"collection"}"#)
+                .file_name("source/records.metadata.json"),
+        );
+    let response = client.post(format!("{base}/source/upload")).multipart(acquisition).send().await.unwrap();
+    assert!(response.status().is_success(), "{}", response.text().await.unwrap());
+    let acquired: serde_json::Value = response.json().await.unwrap();
+    let acquired_ref = acquired["sourceRef"].as_str().unwrap();
+    let id = uuid::Uuid::new_v4().to_string();
+    let response = client.post(format!("{base}/restore")).json(&serde_json::json!({ "request": {
+        "taskId": id, "connectionId": "missing", "database": "target", "sourceDatabase": "source", "sourceRef": acquired_ref
+    }})).send().await.unwrap();
+    assert!(response.status().is_success());
+    tokio::time::sleep(Duration::from_millis(100)).await;
+    let events = client.get(format!("{base}/progress/{id}")).send().await.unwrap();
+    assert!(events.status().is_success());
+    tokio::time::timeout(Duration::from_secs(5), async {
+        let mut stream = events.bytes_stream();
+        let mut text = String::new();
+        while let Some(chunk) = stream.next().await {
+            text.push_str(&String::from_utf8_lossy(&chunk.unwrap()));
+            if text.contains("\"status\":\"error\"") {
+                return;
+            }
+        }
+        panic!("Missing terminal restore event: {text}");
+    })
+    .await
+    .unwrap();
+    let response: serde_json::Value = client
+        .post(format!("{base}/source/release"))
+        .json(&serde_json::json!({ "sourceRef": source_ref }))
+        .send()
+        .await
+        .unwrap()
+        .json()
+        .await
+        .unwrap();
+    assert_eq!(response["released"], true);
+    client
+        .post(format!("{base}/source/release"))
+        .json(&serde_json::json!({"sourceRef":acquired_ref}))
+        .send()
+        .await
+        .unwrap();
+    server.kill().await.unwrap();
+    server.wait().await.unwrap();
+}