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

docs/content/docs/mongodb.cn.mdx

@@ -101,26 +101,39 @@ GridFS 工作台用于管理 Bucket 和文件：
 
 ## 导入与导出
 
-Collection 文档提供独立的 CSV/JSON 导入和 CSV/NDJSON 导出，和关系型 Table Import 不是同一条工作流。CSV 每一行生成一个文档，嵌套值保持嵌套，`$oid`、`$date` 等 Extended JSON 类型可以保留。
+Collection 文档提供独立的 CSV/JSON/NDJSON 与 BSON 导入导出，和关系型 Table Import 不是同一条工作流。CSV 每一行生成一个文档，嵌套值保持嵌套，BSON 备份保留原生 BSON 类型。
 
 ### 导入数据
 
 1. 在集合上右键选择「导入数据」。
-2. 选择 `.csv`、`.json` 或 `.ndjson` 文件。桌面端读取本地路径，Web 端上传到临时文件。
-3. 对照预览调整编码、分隔符、表头、空值和类型策略。
+2. 先明确选择格式，再选择 `.csv`、`.json`、`.ndjson`/`.jsonl`、`.bson` 或 `.bson.gz` 文件。桌面端读取本地路径，Web 端上传到临时文件；更换文件不会覆盖已选择的格式。
+3. 对照预览检查数据。文本格式支持编码和类型设置，CSV 可逐列覆盖推断类型；BSON 导入不应用 CSV 文本转换。
 4. 确认以追加方式写入。重复 `_id` 会停止导入，已提交的批次不会回滚。
 
 CSV 按 MongoDB Compass：首行表头，嵌套字段用点号路径（`address.city`），数组索引（`tags[0]`、`matrix[0][1]`），ObjectId 写成 24 位十六进制，日期为 ISO-8601，空单元格为 null。导入采样前 1000 行推断列类型；`_id` 的十六进制会还原为 ObjectId，`recognizeObjectIdHex` 可将此规则扩展到其他列。JSON/NDJSON 使用 Extended JSON（`$oid`、`$date`）。全部按字符串会把每个单元格当文本。
 
-类型保真：`Double` 导出再导入会变为 `Decimal128`（十进制文本无损）。非 `_id` 的 ObjectId 字段会往返为 String，除非打开 `recognizeObjectIdHex`。小的 `Int64` 可能收窄为 `Int32`；ISO 日期格式的字符串会变为 `DateTime`。Extended JSON 类型（`$binary`、`$regularExpression`、`$timestamp`、`$minKey`、`$maxKey`、`$code`）通过 JSON 文本单元格往返。空数组 `[]` 和空对象 `{}` 往返正确。
+CSV 类型保真：`Double` 导出再导入会变为 `Decimal128`（十进制文本无损）。非 `_id` 的 ObjectId 字段会往返为 String，除非打开 `recognizeObjectIdHex`。小的 `Int64` 可能收窄为 `Int32`；ISO 日期格式的字符串会变为 `DateTime`。Extended JSON 类型（`$binary`、`$regularExpression`、`$timestamp`、`$minKey`、`$maxKey`、`$code`）通过 JSON 文本单元格往返。空数组 `[]` 和空对象 `{}` 往返正确。
 
-首期限制：只支持追加（无清空或 upsert），无按字段类型覆盖，CSV 导出从前 10000 个文档发现字段（之后的字段会被静默丢弃；变化 schema 请改用 NDJSON），无 JSON 数组导出格式，不支持 Excel/GridFS/BSON 二进制导入，失败批次不会自动重试。
+集合导入限制：只支持追加（无清空或 upsert），CSV 导出从前 10000 个文档发现字段（之后的字段会被静默丢弃；变化 schema 请改用 NDJSON 或 BSON），无 JSON 数组导出格式，不支持 Excel/GridFS 导入，失败批次不会自动重试。集合级 BSON 导入只恢复文档，不恢复配套元数据和索引。
 
 ### 导出数据
 
-- 在集合上右键选择「导出数据」，再选 CSV 或 NDJSON，会像表导出一样直接打开保存文件对话框。
+- 在集合上右键选择「导出数据」，再选 CSV、NDJSON、BSON 或 Gzip BSON，会像表导出一样直接打开保存文件对话框。
 - CSV 嵌套对象写为点号列（`address.city`），数组写为索引列（`tags[0]`、`tags[1]`）。NDJSON 每行一个规范 Extended JSON 文档。
-- 字段数超过 256 的 Collection 请改用 NDJSON。
+- 字段数超过 256 的 Collection 请改用 NDJSON 或 BSON。
+
+### 数据库备份与恢复
+
+使用 MongoDB 原生驱动时，在数据库节点右键选择「备份数据库」或「恢复数据库」。数据库备份包含选中集合、集合选项、索引和视图。
+
+- 桌面端支持官方目录布局与 MongoDB archive 格式，均可使用 gzip。Web 端导出 archive 文件，恢复时支持 archive 或上传目录。
+- 目录备份包含 `<数据库>/<集合>.bson` 和 `.metadata.json`，或对应的 `.gz` 文件。MongoDB archive 是独立的流式格式，不是 ZIP 或 tar。
+- 格式与 gzip 选项需明确选择。目录恢复的「读取备份」只读取清单和元数据；确认源库、目标库和集合后，才上传选中的数据文件。
+- 恢复按单遍流式读取和写入进行。可选的 `objcheck` 默认关闭，开启后在同一遍读取中逐文档校验；BSON 边界、gzip 完整性和 archive 校验和检查始终保留。
+- 默认追加写入。「恢复前删除选中的目标同名集合」只删除并重建这些集合。已有 `_id` 可能导致 `E11000`；失败或取消不会回滚此前的写入。
+- Web 上传复用服务端 `DBX_MAX_UPLOAD_MB` 配置。桌面端直接读取本地文件，无需 HTTP 上传。
+
+上述格式可与官方 `mongodump`/`mongorestore` 对支持的集合双向互通。本流程不提供 oplog 回放、时间点一致性保证、UUID 保留，也不处理内部数据库、时间序列集合和加密集合；这些场景请使用官方工具。
 
 如果只需要当前可见选择，仍可从表格复制当前文档、选中文档或字段值为普通 JSON 或 Extended JSON。
 

docs/content/docs/mongodb.mdx

@@ -101,26 +101,39 @@ Dropping a bucket removes its files and chunks. Verify the bucket name and backu
 
 ## Import and Export
 
-Collection documents have a dedicated CSV/JSON import and CSV/NDJSON export workflow. This is separate from relational Table Import: each CSV row becomes a document, nested values stay nested, and Extended JSON types such as `$oid` and `$date` can be preserved.
+Collection documents have dedicated CSV/JSON/NDJSON and BSON import/export workflows. This is separate from relational Table Import: each CSV row becomes a document, nested values stay nested, and BSON dumps preserve native BSON types.
 
 ### Import data
 
 1. Right-click a collection and choose **Import data**.
-2. Select a `.csv`, `.json`, or `.ndjson` file. Desktop reads the local path; Web uploads it to a temporary server file.
-3. Review encoding, delimiter, header, empty-value, and type settings against the preview.
+2. Choose the format explicitly, then select a `.csv`, `.json`, `.ndjson`/`.jsonl`, `.bson`, or `.bson.gz` file. Desktop reads the local path; Web uploads it to a temporary server file. Changing the file does not override the selected format.
+3. Review the preview. Text formats expose encoding and type settings; CSV supports per-column type overrides. BSON imports do not apply CSV text conversions.
 4. Confirm append-only insert. Duplicate `_id` values stop the import. Batches that already committed are not rolled back.
 
 CSV follows MongoDB Compass: a header row, nested fields as dotted paths (`address.city`), array indices (`tags[0]`, `matrix[0][1]`), ObjectId as 24-character hex, dates as ISO-8601, and empty cells as null. Import samples the first 1000 rows to infer column types; `_id` hex is restored as ObjectId, and `recognizeObjectIdHex` extends this to other columns. JSON/NDJSON use Extended JSON (`$oid`, `$date`). All strings keeps every cell as text.
 
-Type fidelity: `Double` exports and reimports as `Decimal128` (lossless for decimal text). Non-`_id` ObjectId fields round-trip as String unless `recognizeObjectIdHex` is on. Small `Int64` may narrow to `Int32`; ISO-date-looking strings become `DateTime`. Extended JSON types (`$binary`, `$regularExpression`, `$timestamp`, `$minKey`, `$maxKey`, `$code`) round-trip through JSON-text cells. Empty array `[]` and empty object `{}` round-trip correctly.
+CSV type fidelity: `Double` exports and reimports as `Decimal128` (lossless for decimal text). Non-`_id` ObjectId fields round-trip as String unless `recognizeObjectIdHex` is on. Small `Int64` may narrow to `Int32`; ISO-date-looking strings become `DateTime`. Extended JSON types (`$binary`, `$regularExpression`, `$timestamp`, `$minKey`, `$maxKey`, `$code`) round-trip through JSON-text cells. Empty array `[]` and empty object `{}` round-trip correctly.
 
-First-release limits: append only (no truncate or upsert), no per-field type override, CSV export discovers fields from the first 10000 documents (later fields are silently dropped; use NDJSON for variable schemas), no JSON-array export format, no Excel/GridFS/BSON binary import, and no automatic retry of failed batches.
+Collection-import limits: append only (no truncate or upsert), CSV export discovers fields from the first 10000 documents (later fields are silently dropped; use NDJSON or BSON for variable schemas), no JSON-array export format, no Excel/GridFS import, and no automatic retry of failed batches. Collection-level BSON import restores documents only, not the companion metadata or indexes.
 
 ### Export data
 
-- Right-click a collection and choose **Export data**, then CSV or NDJSON. This opens the save-file dialog immediately, the same as table export.
+- Right-click a collection and choose **Export data**, then CSV, NDJSON, BSON, or gzip BSON. This opens the save-file dialog immediately, the same as table export.
 - CSV writes nested objects as dotted columns (`address.city`) and arrays as indexed columns (`tags[0]`, `tags[1]`). NDJSON writes one canonical Extended JSON document per line.
-- Collections with more than 256 distinct fields should use NDJSON.
+- Collections with more than 256 distinct fields should use NDJSON or BSON.
+
+### Database backups
+
+With the native MongoDB driver, right-click a database and choose **Back up database** or **Restore database**. Database backups include selected collections, collection options, indexes and views.
+
+- Desktop supports the official dump directory layout and MongoDB archive format, optionally gzip-compressed. Web exports archive files and restores either archives or uploaded directories.
+- A dump directory contains `<database>/<collection>.bson` and `.metadata.json`, or their `.gz` variants. MongoDB archive is a separate stream format, not ZIP or tar.
+- Choose the format and gzip option explicitly. For directory restore, **Read backup** reads the catalog and metadata; selected data files are uploaded only after confirming the source database, target database and collections.
+- Restore reads and writes data in one pass. Optional `objcheck` validates each BSON document in that same pass and defaults off. Framing, gzip and archive checksum checks remain enabled.
+- Restore appends by default. **Drop selected target collections before restoring** deletes and recreates only those collections. Existing `_id` values can cause `E11000`; failures or cancellation do not roll back earlier writes.
+- Web uploads use the shared `DBX_MAX_UPLOAD_MB` server setting. Desktop reads local files without an HTTP upload.
+
+These formats interoperate with `mongodump`/`mongorestore` for supported collections. Oplog replay, point-in-time guarantees, UUID preservation, internal databases, time-series collections and encrypted collections are outside this workflow; use the official tools for those cases.
 
 Copy the current document, selected documents, or field values as plain JSON or Extended JSON from the grid when you only need the visible selection.