Skip to main content
Version: Next


Loads bytes from a file. Saves bytes to a file.



file [-f|--follow] [-m|--mmap] [-t|--timeout=<duration>] <path>


file [-a|--append] [-r|--real-time] [--uds] <path>


The file loader acquires raw bytes from a file. The file saver writes bytes to a file or a Unix domain socket.

When used as from file <path> | ... or ... | to file <path>, Tenzir uses the following heuristics to infer the format based on the filename:

  • If the filename is eve.json, use suricata
  • If the file extension is ndjson, use json.
  • Otherwise, if the filename has an extension, try to use this as the format (this will produce an error if there is no such format)
  • If the filename has no extension (that includes the reserved path -), the default format is JSON.

<path> (Loader, Saver)

The path to the file to load/save. If intermediate directories do not exist, the saver will create them.

When ~ is the first character, it will be substituted with the value of the $HOME environment variable.

The path - is a reserved value and means stdin for the loader and stdout for the saver.

-f|--follow (Loader)

Do not stop when the end of file is reached, but rather to wait for additional data to be appended to the input.

This flag has the semantics of the "tail -f" idiom in Unix.

-m|--mmap (Loader)

Use the mmap(2) system call to map the file and produce only one single chunk of bytes, instead of producing data piecemeal via read(2). This option effectively gives the downstream parser full control over reads.

For the feather and parquet parsers, this significantly reduces memory usage and improves performance.

-t|--timeout=<duration> (Loader)

Wait at most for the provided duration when performing a blocking call to the system call read(2).

This flags comes in handy in combination with --follow to produce a steady pulse of input in the pipeline execution, as input (even if empty) drives the processing forward.

-a|--append (Saver)

Append to the file instead of overwriting it with a new file.

-r|--real-time (Saver)

Immediately synchronize the file with every chunk of bytes instead of buffering bytes to batch filesystem write operations.

--uds (Saver)

Interpret path as a Unix domain socket and connect to it.


Read JSON from stdin via from and convert it to CSV:

from - read json | to stdout | write csv

Read 1 MiB from a file /tmp/data and write the bytes another file /tmp/1mb, blocking if /tmp/data is less than 1 MiB until the file reaches this size:

load file -f /tmp/data | head 1 Mi | save file /tmp/1mb