pipexec - create a directed graph of processes and pipes



pipexec − create a directed graph of processes and pipes




pipexec creates an arbitrary network (directed graph) of processes and pipes in between - even cycles are possible. It overcomes the shortcomings of shells that are typically only able to create non cyclic trees.

pipexec also monitors all it’s child processes and is able to restart the whole network of processes and pipes if one crashes. Therefore pipexec can be used in SYSV-init or systemd configuration to run a network of processes.



print help and version information

−l logfd

use the given file descriptor for text logging. If a ’s’ is specified, syslog is used. Example: Specifying ’2’ means log to stderr.

−j logfd

use the given file descriptor for json logging. If a ’s’ is specified, syslog is used. Example: Specifying ’2’ means log to stderr. As this is meant to be parsed by other programs, this is an official and supported interface which is described in the JSON LOGGING chapter.

−p pidfile

with pipexec it is possible to handle pipes within SYSV-init scripts. In some environments (e.g. RHEL6, Debian7) the start and stop routines need a pid file. If this option is given, pipexec writes its own pid into the file shortly after start of pipexec.


if one sub-process (child) gets killed and this options is given, all other sub-processes are also killed. Afterwards all processes are restarted.

−s sleep_time

the time interval in seconds before a restart. This option makes only sense when also the ’−k’ option is specified.


Inside a shell it is possible to start processes and redirect the output to other processes.

cat Chap1.txt Chap2.txt | grep bird | wc −l

Three processes are created: the standard output (file descriptor (fd) 1) of the ’cat’ process is connected to the standard input (fd 0) of the ’grep’ command, and the standard output of the ’grep’ command is connected to the standard input (fd 0) of the ’wc’ process.

Please note that the assignment between names and file descriptor number is pure historical and has no technical background.

find / 1> >(grep .txt) 2> >(wc >/tmp/w.log)

In this more complex example, the fd 1 of the ’find’ process is connected to fd 0 of ’grep’ and fd 2 is connected to fd 0 of ’wc’.

The limitation using this way of specifying processes and pipes is, that it is not possible to have any cycles. It is impossible to e.g. pass a fd of ’wc’ either to ’grep’ or to ’find’.

pipexec overcomes these limitations. It makes it possible to link any two arbitrary file descriptors in a set of processes.


When building up a network of processes and pipes, there is the need to specify each element separately.

The processes will be the nodes in the network (directed graph), the connections of the file descriptors between to processes are the edges. Each node (process) has a unique name assigned to it. This makes it possible to differentiate between using the same command more than once.

The format of specifying a process is
[ NAME /path/to/command arg1 arg2 ... argN ]

The first parameter ’NAME’ must be a unique name. The second parameter must be the full path of the command to execute. Please note that always the full path must be specified, there is no PATH environment variable handling (execv(2) is used internally to span new processes). The following parameters are the parameters passed to the command.

The whole definition must be enclosed in square brackets. The square brackets must be given separately - before and after them must be a space.

The format of specifying a pipe between processes is


The names are the names of the processes, the numbers are the number of the file descriptor that should be used to build the pipe in between. When using pipexec from a shell (like bash) there is the need to escape the brackets or use quotation marks.


pipexec can log in JSON format. This is an official supported interface which is defined in this chapter. This can be seen as stable as no changes will be made during small version upgrades. Of course additions will be made also within minor upgrades, but this should be fine because of the underlaying JSON format.

Each JSON log object (e.g. line) will be in a separate line.

The following keys are always present:

The timestamp is the time(0) (seconds since epoch) when the log in generated.


The pid of the pipexec process.


An id that specifies defined log objects.


An indicator in which internal module this log was generated.


One of debug, info, warning, error.


A short message describing the event.

Depending on the id there might be additional fields which are described in the next section.

id = 0

id of 0 is a special case: this is used for internal logs only. The logs are not documented and may change at any time. The information can be used to get an idea what currently happens in pipexec.

id = 1

This gives information about the process ids (pids) of the commands. The field command contains the command (i.e. the part in the command before the colon. The field command_pid contains the pid of the command.

id = 2

This log message is emitted when a child (command) exits. It contains some information about the exit status and termination reason of the command. The field command_pid contains the pid of the command which just terminated. The field status is the value which is set by waitpid(2). normal_exit, child_status and child_signaled are WIFEXITED, WEXITSTATUS and WIFSIGNALED of the status respectively.


pipexec returns 1 if any of the child processes fails else 0 is returned.


The shell command
cat Chap1.txt Chap2.txt | grep bird | wc −l

is equivalent to
pipexec [ CAT /bin/cat Chap1.txt Chap2.txt ] \
[ GREP /usr/bin/grep bird ] [ WC /usr/bin/wc −l ] \
"{CAT:1>GREP:0}" "{GREP:1>WC:0}"

The pipexec equivalent is longer and more complex in this example. But pipexec can build cycles that are impossible within a shell:
pipexec [ A /bin/cmd1 ] [ B /bin/cmd2 ] "{A:1>B:0}" "{B:1>A:0}"

When using json log, you get output like:
{"timestamp":1655706460,"pipexec_pid":42850,"id":1,"type":"exec","serverity":"info","message":"New child forked","command":"A","command_pid":"42851"}
{"timestamp":1655706886,"pipexec_pid":42869,"id":2,"type":"tracing","serverity":"info","message":"child exit","command_pid":"42870","status":"1","normal_exit":"1","child_status":"1","child_signaled":"0"}

For more examples see the ptee(1) and peet(1) man pages.


bash(1), ptee(1), peet(1), execv(2)


Written by Andreas Florath ([email protected])


Copyright © 2015,2022 by Andreas Florath ([email protected]). License GPLv2+: GNU GPL version 2 or later <http://gnu.org/licenses/gpl.html>.

Updated 2024-01-29 - jenkler.se | uex.se