# SFTP Parquet Import Integration

The Data Connector for `SFTP Parquet` enables you to import Parquet files stored on your SFTP server to Treasure AI.

## Prerequisites

- Basic knowledge of Treasure AI.
- Before using this integration, determine valid protocols for your environment.
- Check your accepted IP range and port if you are using a firewall. Server administrators sometimes change the default port number from TCP 22 for security reasons.
- "PuTTY" and other formats are not supported.


## Limitations and Supported

- **Parquet files only** - This integration supports Parquet files exclusively. Other file formats (CSV, JSON, TSV, etc.) are not supported.
- Files must be in valid Parquet format.
- The row group size of the parquet file should be around 128 MB to avoid Out of Memory


## Static IP Address of Treasure Data Integration

If your security policy requires IP whitelisting, you must add Treasure Data's IP addresses to your allowlist to ensure a successful connection.

Please find the complete list of static IP addresses, organized by region, at the following [document](/apis/endpoints/ip-addresses-integrations-result-workers)

## Use the Treasure Console to Create Your Connection

### Create a New Connection

In Treasure AI, you must create and configure the data connection prior to running your query. As part of the data connection, you provide authentication to access the integration.

1. Open **Treasure Console**.
2. Navigate to **Integrations Hub** >  **Catalog**.
3. Search for and select SFTP Parquet.


![New Authentication](/assets/sftp_parquet_create_authentication.c0f74bbdbd0859b4db4b777d64451d640d0237cc62be5dd85d8c55ac1a40858f.a126b8b7.webp)
4. Select **Create Authentication**.

![Authentication Detail](/assets/sftp_parquet_create_authentication_detail.edece0f1d93926a18f8a85860c58b5ebbea9f3a236abb8ed6031206451c3edf3.a126b8b7.webp)
5. The following dialog opens. Edit the parameters. Select **Continue**.

| Parameters | Description |
|  --- | --- |
| **Host** | The host information of the remote SFTP instance, for example, an IP address. |
| **Port** | The connection port on the remote SFTP instance, the default is 22. |
| **User** | The user name used to connect to the remote SFTP instance. |
| **Authentication mode** | The way you choose to authenticate with your SFTP server. |
| **Secret key file** | Required if 'public/private key pair' is selected from `Authentication Mode`. (RSA, DSS, ECDSA, and ED25519 are supported.) |
| **Passphrase for secret key file** | (Optional) If required, provide a passphrase for the provided secret file. |
| **Retry limit** | The number of times to retry a failed connection (default 10). |
| **Timeout** | Connection timeout in seconds (default 600). |
| **Maximum SFTP connections** | The maximum number of concurrent SFTP connections (integer, default 1, minimum 1, maximum 10). Controls connection pool size for parallel file transfers. |


1. Enter a name for your connection.
2. Choose to share the authentication with others or not.
3. Select **Continue.**


## Transfer Your Data to Treasure AI

After creating the authenticated connection, you are automatically taken to Authentications.

1. Search for the connection you created.
2. Select **New Source**.
3. Type a name for your **Source** in the Data Transfer Name field.
4. Select **Next**.
5. Edit the following parameters:


| Parameters | Description |
|  --- | --- |
| User directory root | Check if the path prefix is under the user directory Ex: /home/test_user |
| Path prefix
 | Prefix of target files, and it must point to a folder (string, required). The path prefix has to be a folder path. If a partial file name is included in the file path, you will receive an `invalid path_prefix:xxx` error message.
The path_prefix must be a folder path containing Parquet files.
 |
| Path match pattern | Type a regular expression to query file paths. If a file path doesn't match the specified pattern, the file is skipped. For example, if you specify the pattern `.parquet$`, then a file is skipped if its path doesn't match the pattern. |
| Incremental | Enables incremental loading (boolean, optional. default: true). If incremental loading is enabled, the config diff for the next execution will include last_path parameter so that the next execution skips files before the path. Otherwise, last_path is not included. |
| Start after path | Only paths lexicographically greater than this will be imported. |


1. Select **Next**.


The Data Settings page can be modified for your needs or you can skip the page.

![New data source](/assets/sftp_parquet_create_source.e339833b61b0a85a88c3b49aa72f05f511b5245b140695ad9effa97302187a0b.a126b8b7.webp)

![Data source detail](/assets/sftp_parquet_create_source_detail.c3a0d77cc09da3090b4ee900298e1796ec79b5039afca21bfb2ac8770e34bd88.a126b8b7.webp)

### Data Preview

You can see a [preview](/products/customer-data-platform/integration-hub/batch/import/previewing-your-source-data) of your data before running the import by selecting Generate Preview. Data preview is optional and you can safely skip to the next page of the dialog if you choose to.

1. Select **Next**. The Data Preview page opens.
2. If you want to preview your data, select **Generate Preview**.
3. Verify the data.


### Data Placement

For data placement, select the target database and table where you want your data placed and indicate how often the import should run.

1. Select **Next.** Under Storage, you will create a new or select an existing database and create a new or select an existing table for where you want to place the imported data.
2. Select a **Database** > **Select an existing** or **Create New Database**.
3. Optionally, type a database name.
4. Select a **Table**> **Select an existing** or **Create New Table**.
5. Optionally, type a table name.
6. Choose the method for importing the data.
  - **Append** (default)-Data import results are appended to the table.
If the table does not exist, it will be created.
  - **Always Replace**-Replaces the entire content of an existing table with the result output of the query. If the table does not exist, a new table is created.
  - **Replace on New Data**-Only replace the entire content of an existing table with the result output when there is new data.
7. Select the **Timestamp-based Partition Key** column.
If you want to set a different partition key seed than the default key, you can specify the long or timestamp column as the partitioning time. As a default time column, it uses upload_time with the add_time filter.
8. Select the **Timezone** for your data storage.
9. Under **Schedule**, you can choose when and how often you want to run this query.


#### Run once

1. Select **Off**.
2. Select **Scheduling Timezone**.
3. Select **Create & Run Now**.


#### Repeat Regularly

1. Select **On**.
2. Select the **Schedule**. The UI provides these four options: *@hourly*, *@daily* and *@monthly* or custom *cron*.
3. You can also select **Delay Transfer** and add a delay of execution time.
4. Select **Scheduling Timezone**.
5. Select **Create & Run Now**.


After your transfer has run, you can see the results of your transfer in **Data Workbench** > **Databases.**

# Import with SFTP Parquet via Treasure Workflow

Create and run a workflow

```yaml
_export:
  td:
    database: workflow_sftp_parquet
    table: workflow_sftp_parquet
+import_from_sftp_parquet:
  td_load>: imports/seed.yml
  database: ${td.database}
  table: ${td.table}
```

Modify the *seed.yml* file with your SFTP connection details for the import.

```yaml
in:
  type: sftp_parquet
  host: HOST
  port: <PORT, default is 22>
  auth_method: key_pair
  user: USER
  secret_key_file:
    content: |
      -----BEGIN RSA PRIVATE KEY-----
      Proc-Type: 4,ENCRYPTED
      DEK-Info: AES-128-CBC...
      ...
      -----END RSA PRIVATE KEY-----
  secret_key_passphrase: PASSPHRASE
  user_directory_is_root: true
  timeout: 600
  path_prefix: /path/to/parquet/files
out:
  mode: append
```

| Configuration Parameters | Value |
|  --- | --- |
| host: | (string, required) |
| port: | (string, default: 22) |
| auth_method: | (string ['password', 'key_pair'], required) |
| user: | (string, required) |
| password: | (string, default: null) |
| secret_key_file: | (string, default: null). OpenSSH format is required. |
| secret_key_passphrase: | (string, default: "") |
| user_directory_is_root: | (boolean, default: true) |
| timeout: sftp connection timeout seconds | (integer, default: 600) |
| maximum_sftp_connections: maximum concurrent SFTP connections | (integer, default: 1, minimum: 1, maximum: 10) |
| path_prefix: Prefix of output paths | (string, required) |
| incremental: enables incremental loading | (boolean, optional. default: true). If incremental loading is enabled, config diff for the next execution will include last_path parameter so that next execution skips files before the path. Otherwise, last_path will not be included. |
| path_match_pattern: | regexp to match file paths. If a file path doesn't match with this pattern, the file will be skipped (regexp string, optional) |
| total_file_count_limit: | maximum number of files to read (integer, optional) |
| min_task_size (experimental): | minimum size of a task. If this is larger than 0, one task includes multiple input files. This is useful if too many number of tasks impacts performance of output or executor plugins badly. (integer, optional) |


# Import with SFTP via the CLI (TD Toolbelt)

## Install TD Toolbelt

Install the most current [Treasure Data Toolbelt](https://toolbelt.treasuredata.com/).

```
$ td --version
```

## Create Seed Config File (seed.yml)

Prepare *seed.yml,* as shown in the following example, with your SFTP Parquet details. We support two authentication methods: Public / Private Key Pair and Password.

### Public and Private Key Pair Authentication

Create *seed.yml* with the following content.

```yaml
in:
  type: sftp_parquet
  host: HOST
  port: <PORT, default is 22>
  auth_method: key_pair
  user: USER
  secret_key_file:
    content: |
      -----BEGIN RSA PRIVATE KEY-----
      Proc-Type: 4,ENCRYPTED
      DEK-Info: AES-128-CBC...
      ...
      -----END RSA PRIVATE KEY-----
  secret_key_passphrase: PASSPHRASE
  user_directory_is_root: true
  timeout: 600
  path_prefix: /path/to/parquet/files
out:
  mode: append
  exec: {}
```

`secret\_key\_file` requires OpenSSH format.

### Password Authentication

Create *seed.yml* with the following content.

```yaml
in:
  type: sftp_parquet
  host: HOST
  port: <PORT, default is 22>
  auth_method: password
  user: USER
  password: PASSWORD
  user_directory_is_root: true
  timeout: 600
  path_prefix: /path/to/parquet/files
out:
  mode: append
  exec: {}
```

You can use the following special characters in the password: "#$!*@"

The SFTP Parquet integration imports all Parquet files that match the specified prefix. path_prefix must point to a folder containing Parquet files (e.g. path_prefix: `path/to/parquet/files`–> `path/to/parquet/files/data_201501.parquet`, `path/to/parquet/files/data_201502.parquet`, …, `path/to/parquet/files/data_201505.parquet`).

## Guess Fields (Generate load.yml)

Use *connector:guess*. This command automatically reads the source file and assesses (uses logic to guess) the file format.

```
$ td connector:guess seed.yml -o load.yml
```

If you open *load.yml*, you see the guessed file format definitions. Since Parquet files are self-describing, the column schema is automatically detected from the Parquet metadata.

```yaml
in:
  type: sftp_parquet
  host: HOST
  port: <PORT, default is 22>
  auth_method: key_pair
  user: USER
  secret_key_file:
    content: |
      -----BEGIN RSA PRIVATE KEY-----
      Proc-Type: 4,ENCRYPTED
      DEK-Info: AES-128-CBC...
      ...
      -----END RSA PRIVATE KEY-----
  secret_key_passphrase: PASSPHRASE
  user_directory_is_root: true
  timeout: 600
  path_prefix: /path/to/parquet/files
out:
  mode: append
  exec: {}
```

Then, you can preview how the system will parse the file by using the *preview* command.

```
td connector:preview load.yml
```

The guess command reads the Parquet file metadata to determine column names and types automatically.

If the system detects your column name or column type unexpectedly, you can modify the schema, though Parquet files typically have well-defined schemas.

You also must create a database and table prior to executing the data load job. Follow these steps:

```bash
td database:create td_sample_db
td table:create td_sample_db td_sample_table
```

## Execute Load Job

Submit the load job. It may take a couple of hours, depending on the size of the data. Specify the Treasure AI database and table where the data should be stored.

It's also recommended to specify *--time-column* option, because Treasure AI's storage is partitioned by time (see [data partitioning](https://docs.treasure.ai/smart/project-product-documentation/data-partitioning-in-treasure-data)) If the option is not provided, the integration chooses the first *long* or *timestamp* column as the partitioning time. The type of the column specified by *--time-column* must be either of *long* and *timestamp* type.

If your data doesn't have a time column, you can add a time column by using *add_time* filter option. For more details, see [add_time filter plugin](https://docs.treasure.ai/smart/project-product-documentation/add_time-filter-function).

```bash
td connector:issue load.yml --database td_sample_db --table td_sample_table --time-column created_at
```

The connector:issue command assumes that you have already created a *database(td_sample_db)*and a *table(td_sample_table)*. If the database or the table do not exist in TD, the connector:issue command fails. If this happens, [create the database](https://docs.treasure.ai/smart/project-product-documentation/creating-or-viewing-a-database) and [create a table](https://docs.treasure.ai/smart/project-product-documentation/creating-or-viewing-tables) manually, or use *--auto-create-table* option with *td connector:issue* command to auto create the database and table:

```bash
td connector:issue load.yml --database td_sample_db --table td_sample_table --time-column created_at --auto-create-table
```

The integration does not sort records on server-side. To use time-based partitioning effectively, sort records beforehand.

If you have a field called `time`, you don't have to specify the `--time-column` option.

```bash
td connector:issue load.yml --database td_sample_db --table td_sample_table
```

## Scheduled Execution

You can schedule periodic integration execution for incremental SFTP Parquet file import. We configure our scheduler carefully to ensure high availability. Using this feature means you no longer need a *cron* daemon on your local data center.

For the scheduled import, the integration for SFTP Parquet imports all files that match with the specified prefix (e.g. path_prefix: `path/to/parquet/files` –> `path/to/parquet/files/data_201501.parquet`, `path/to/parquet/files/data_201502.parquet`, …, `path/to/parquet/files/data_201505.parquet`) at first and remembers the last path (`path/to/parquet/files/data_201505.parquet`) for the next execution.

On the second and on subsequent runs, it imports only files that come after the last path in alphabetical (lexicographic) order. (`path/to/parquet/files/data_201506.parquet`, …)

### Create the Schedule

A new schedule can be created using the *td connector:create* command. The following are required: the name of the schedule, the cron-style schedule, the database and table where the data will be stored, and the integration configuration file.

```bash
td connector:create \
daily_parquet_import \
"10 0 * * *" \
td_sample_db \
td_sample_table \
load.yml
```

It's also recommended to specify the *--time-column* option, because Treasure AI's storage is partitioned by time.

```bash
td connector:create \
daily_parquet_import \
"10 0 * * *" \
td_sample_db \
td_sample_table \
load.yml \
--time-column created_at
```

The `cron` parameter also accepts three special options: `@hourly`, `@daily` and `@monthly`.

By default, the schedule is set up in UTC timezone. You can set the schedule in a timezone using -t or --timezone option. The `--timezone` option supports only extended timezone formats like 'Asia/Tokyo', 'America/Los_Angeles', etc. Timezone abbreviations like PST, and CST are *not* supported and may lead to unexpected schedules.

### List the Schedules

You can see the list of currently scheduled entries by running the command *td connector:list*.

```
td connector:list
```

### Show the Setting and Schedule History

*td connector:show* shows the execution setting of a schedule entry.

```
td connector:show daily_parquet_import
```

*td connector:history* shows the execution history of a schedule entry. To investigate the results of each individual run, use *td job jobid*.

```bash
td connector:history daily_parquet_import
```

### Delete the Schedule

*td connector:delete* will remove the schedule.

```bash
td connector:delete daily_parquet_import
```

## Modes for out plugin

You can specify the file import mode in the *out* section of seed.yml.

### append (default)

This is the default mode, and records are appended to the target table.

```yaml
in:
  ...
out:
  mode: append
```

### replace

This mode replaces data in the target table. Note that any manual schema changes made to the target table will remain intact with this mode.

```
in:
  ...
out:
  mode: replace
```

# Import from SFTP Parquet Server via Workflow

For sample workflows of importing Parquet files from your STFP server, view [Treasure Boxes](https://github.com/treasure-data/treasure-boxes/tree/master/td_load/sftp).