Skip to main content
Version: v4.13


Executes a system command and hooks its stdin and stdout into the pipeline.


shell <command>


The shell operator executes the provided command by spawning a new process. The input of the operator is forwarded to the child's standard input. Similarly, the child's standard output is forwarded to the output of the operator.


The command to execute and hook into the pipeline processing.

The value of command is a single string. If you would like to pass a command line as you would on the shell, use single or double quotes for escaping, e.g., shell 'jq -C' or shell "jq -C". The command is interpreted by /bin/sh -c.


Show a live log from the tenzir-node service:

shell "journalctl -u tenzir-node -f | read json"

Consider the use case of converting CSV to JSON:

tenzir 'read csv | write json' | jq -C

The write json operator produces NDJSON. Piping this output to jq generates a colored, tree-structured variation that is (arguably) easier to read. Using the shell operator, you can integrate Unix tools that rely on stdin/stdout for input/output as "native" operators that process raw bytes. For example, in this pipeline:

write json | save stdout

The write operator produces raw bytes and save accepts raw bytes. The shell operator therefore fits right in the middle:

write json | shell "jq -C" | save stdout

Using user-defined operators, we can expose this (potentially verbose) post-processing more succinctly in the pipeline language:

write json | shell "jq -C" | save stdout

Now you can use jsonize as a custom operator in a pipeline:

tenzir 'read csv | where field > 42 | jsonize' < file.csv

This mechanism allows for wrapping also more complex invocation of tools. Zeek, for example, converts packets into structured network logs. Tenzir already has support for consuming Zeek output with the formats zeek-json and zeek-tsv. But that requires attaching yourself downstream of a Zeek instance. Sometimes you want instant Zeek analytics given a PCAP trace.

With the shell operator, you can script a Zeek invocation and readily post-process the output with a rich set of operators, to filter, reshape, enrich, or route the logs as structured data. Let's define a zeek operator for that:

shell "zeek -r - LogAscii::output_to_stdout=T
| read zeek-json

Processing a PCAP trace now is a matter of calling the zeek operator:

gunzip -c example.pcap.gz |
tenzir 'zeek | select id.orig_h, id.orig_p, id.resp_h | head 3'
{"id": {"orig_h": null, "resp_h": null, "resp_p": null}}
{"id": {"orig_h": "", "resp_h": "", "resp_p": 0}}
{"id": {"orig_h": "", "resp_h": "", "resp_p": 22}}

NB: because zeek (= shell) reads bytes, we can drop the implicit load stdin source operator in this pipeline.