Files Overview

Naming

You can assign custom schema and table names while building your connector. Column names are generated based on the uploaded files. For a CSV, there will be one column for each entry in the first line. For a JSON file, there will be one column for each top level field in the JSON object.

If you change the name of a file in your bucket, all data in that file will be re-imported, which may result in duplicate values within that table.

Primary keys

Fivetran adds the below three columns as composite primary keys for a table. In all modes, _file and _line are composite primary keys. The column _modified is an additional primary key that is added if Modified File Merge is set to the append_file option.

Excluding source data

Inclusion and exclusion of source data in file storage integrations is done via specification of a folder path. The integration will sync all files under a certain folder path, even if they are themselves in nested subfolders of the specified folder. Excluding fields from source files is not currently supported.

Updating data

After the initial load of historical data, Fivetran only pulls incremental updates of any new or modified files from the source, adding an extra column to your tables called fivetran_synced (TIMESTAMP) that indicates the start time of the job that synced this row.

We do not recommend manually changing the contents of your files. Doing so will result in the following behaviours:

The integration has file-level granularity. If you modify a file, we update the table in the destination to match the contents of the file. Adding rows to a CSV that has already been synced will change the last_modified date, which will then trigger a re-sync of the contents of that file. The file name and line numbers will be used as primary keys. Whatever you change in your file will be represented in the file in the destination, but you may lose the previous version of the file that corresponded to the old record.

Frequency of updates

By default, all storage services sync new and modified files every 15 minutes. Depending on the size of each update or the number of tables, it may take longer. In that case, the integration will update at the next 15 minute interval, i.e.

1. Update X Start: 9:00am

2. Update X Finish: 9:18am

3. Update Y Start: 9:30am

You can change the update interval in the Dashboard. A storage service sync that encounters an error will re-try periodically after the shorter of the sync frequency or 1 hour.

Deleted data

We handle deleted data based on last modified dates, and we treat deleted files (and column) and deleted rows differently.

If you delete a file in your source bucket, it no longer exists, so it will not have a last modified date. Therefore, the deletion will not be reflected in your destination - so data from the deleted file will still remain in your destination.

Deleted columns will change the last modified date, but the deletions will not appear in your destination.

Deleted rows, on the other hand, will have a last modified date, so the change will be reflected in your destination.

Sync strategy

Magic Folder connectors don’t support incremental syncs. We use the last modified date of the files to detect changes in the files of your cloud folder.

We do the following:

• If you add a new file to your cloud folder, we sync the file as a new table in your destination.
• If you modify a file in your cloud folder, we sync the file’s content again to the destination table. We delete the previous data of the table and don’t capture deletes.
• If you delete a file from your cloud folder, we don’t delete the table from your destination.

If we find files of the same name but with different extensions in the folder, we check the last modified dates of the files and sync only the least recently modified file. For example, if you add a file called abc.csv in your cloud folder, and then add another file called abc.json, we check the last modified date of both files and sync abc.csv to your destination.

Sync limitations

We do not sync files inside nested folders in your cloud folder. We do not sync empty folders or unsupported file types.

We don’t sync empty worksheets. We don’t sync a worksheet if the entire first row of the worksheet is empty. We don’t sync a column if its first row is empty.

You can find more information on the individual connector pages.

Folder path (optional)

This folder path is used to specify a portion of the bucket in which you’d like Fivetran to look for files. Any files under the specified folder and all of its nested subfolders will be examined for files we can upload. If no prefix is supplied, we’ll look through the entire bucket for files to sync.

File pattern (optional)

The file pattern is a regular expression that we use to decide whether or not to sync certain files. It applies to everything under the prefix. For instance, suppose under the prefix logs you had three folders: 2017, 2016, and errors. Using the pattern \d\d\d\d/.*, you could exclude all the files in the errors folder, because \d\d\d\d only applies to the folders whose name consists of four consecutive digits, and because .* after / applies to any files in these folders. If you’re not sure what regular expression to use, you can leave this field blank, and we’ll sync everything under the prefix. If you’re feeling particularly bold, you can learn to write your own regex here and test it out using this tool.

File type

The file type is used to let Fivetran know that even files without a file extension ought to be parsed as this file type. For example, if you have an automated CSV output system that saves files without a .csv extension, you can specify the CSV type and we will sync them correctly as CSVs. Selecting “infer” will let Fivetran infer from a file’s extension (.csv, .tsv, .json, .avro, or .log) what to sync. If you do choose a file type, every file we examine will be interpreted as the file type you select, so make sure everything Fivetran syncs has the same file type!

Compression

The compression format is used to let Fivetran know that even files without a compression extension should be decompressed using the selected compression format. For example, if you have an automated CSV output system that GZIPs files to save space, but saves them without a .gzip extension, you can set this field to GZIP. The integration will then decompress every file that it examines using GZIP. If all of your compressed files are correctly marked with a matching compression extension (.bz2, .gz, .gzip, .tar, or .zip), you can select “infer”.

Error handling

Selecting skip ignores any improperly formatted data within a file, allowing you to sync only valid data. Choosing fail enables you to prevent a file from syncing if any improperly formatted data is detected. With either option you will receive a notification on your dashboard if errors are encountered.

Modified file merge

When a previously synced file is modified, should the rows in the destination be replaced or should the new rows be appended to the table. upsert_file option will replace records in destination, using the filename and line number as the primary key. append_file option will append records.

NOTE: Select the upsert_file option to replicate the file contents to the destination. Select the append_file option to track the file modification history.

Archive folder pattern (optional)

If there are multiple files within archive (TAR or ZIP) folders, you can use the archive folder pattern to filter those as well. For example, the archive folder pattern .*json will sync from an archive folder only those files that end in a .json file extension.

Null sequence (optional)

CSVs have no native notion of a null character. However, some CSV generators have created one, using characters such as \N to represent null. Note: text is un-escaped before the null sequence is matched, so don’t use the escape character in your null sequence. Only use this field if you are sure your CSVs have a null sequence.

Delimiter (optional)

The delimiter is a character used in CSV files to separate one field from the next. If this is left blank, Fivetran will infer the delimiter for each file, and files of many different types of delimiters can be stored in the same folder with no problems. If this is not left blank, then all CSV files in your search path will be parsed with this delimiter.

Escape character (optional)

CSVs have a special rule for escaping quotation marks as opposed to other characters - they require two consecutive double quotes to represent an escaped double quote. However, some CSV generators do not follow this rule and use other characters like backslash for escaping. Only use this field if you are sure your CSVs have a different escape character.

Skip header lines (optional)

Some CSV generating programs include additional header lines at the top of the file. They consist of few lines that do not match the format of the rest of the rows in the file. These header rows can cause undesired behavior in Fivetran because we attempt to parse them as if they were records in your CSV. By setting this value, you can skip fixed-length headers at the beginning of your CSV files.

Note: Only use this field if you are sure your CSVs have more than one header line. We will skip these lines before we map the header to row data for the CSVs.

Skip footer lines (optional)

Some CSV generating programs include a footer at the bottom of the file. It consists of a few lines that do not match the format of the rest of the rows in the file. These footer rows can cause undesired behavior in Fivetran because we attempt to parse them as if they were records in your CSV. By setting this value, you can skip fixed-length footers at the end of your CSV files.

Headerless files (optional)

Some CSV generating programs do not include column name headers for the files — they only contain data rows. By setting this value, you request Fivetran to generate generic column names following the convention of column_0, column_1, … column_n to map the rows.

Note: Only use this option if you are sure your CSVs do not have the header line, and the column orderings remain the same. Otherwise, it can cause undesired behavior in the schema. For example, if the column order changes in the source, we won’t be able to identify column changes. If you have no control of the CSV generator and still want to turn on the empty file option, we recommend always keeping the error handling option “on” to catch potential errors.