Text processing

`shuf`

Shuffle lines into random order, or pick random lines, numbers or words.

shuf [OPTION...] [FILE] | shuf -e [ARG...] | shuf -i LO-HI

9 of 9 examples run in Debian 13.7 on 2026-10-31 · coreutils 9.7-3 · manual: shuf(1)

shuf prints its input lines in a random order. With -n N it stops after N lines, which makes it a random sampler: one random line from a file, five random test cases, a random winner. It can also shuffle its arguments (-e) or a range of integers (-i 1-100) without any input file, and with -r it samples with replacement, so values can repeat, like dice rolls.

Every example on this page uses real randomness, so its output differs each time the page is built; that is the point of the tool. When you need a repeatable shuffle (a fixed train/test split, a reproducible test order), give shuf a fixed stream of bytes with --random-source: the same bytes always produce the same order.

shuf is a GNU coreutils program. sort -R also shuffles, but it keeps identical lines next to each other, so it is not a true shuffle.

Examples

Shuffle all lines

$ shuf todo.txt

explain

output

buy milk
fix login bug
deploy v2.4
write tests
fix logout bug

All five lines, in a random order that changes every run.

Pick random lines

$ shuf -n 2 todo.txt

explain

output

write tests
fix login bug

-n 2 outputs only two lines, picked at random without repeats.

A random line from a file

$ shuf -n 1 data/fruits.txt

explain

output

apple

One line chosen uniformly at random. apple appears three times in the file, so it is three times as likely as fig to be picked.

A random number in a range

$ shuf -i 1-6 -n 1

explain

output

-i 1-6 shuffles the integers 1 to 6 instead of reading input; -n 1 keeps one, like rolling a die.

Shuffle words given on the command line

$ shuf -e red green blue yellow

explain

output

blue
green
yellow
red

-e treats each argument as an input line, so no file or pipe is needed.

Sample with replacement

$ shuf -r -n 10 -e H T | paste -s -d ' '

explain

output

H H H T T T T H H H

-r lets values repeat, so this is ten independent coin tosses. Without -r, -n 10 could not return more than the two values that exist.

A repeatable shuffle

$ shuf --random-source=<(yes 42) -i 1-10 | paste -s -d ' '
> shuf --random-source=<(yes 42) -i 1-10 | paste -s -d ' '

output

3 7 5 10 1 6 9 8 2 4
3 7 5 10 1 6 9 8 2 4

--random-source reads "random" bytes from a file instead of the system's generator. Here the endless output of yes 42 acts as a seed: both runs print the same order, and a different seed string gives a different one. This is for reproducibility, not security.

Split data into random halves

$ tail -n +2 data/users.csv | shuf --random-source=<(yes split) > shuffled.csv
> head -n 4 shuffled.csv > train.csv
> tail -n +5 shuffled.csv > test.csv
> wc -l train.csv test.csv

output

  4 train.csv
  4 test.csv
  8 total

Shuffle once with a fixed seed, then cut the result: the 8 users are divided into two groups of 4, and rerunning gives the same groups.

Write the result to a file

$ shuf -n 3 -o picks.txt todo.txt
> wc -l picks.txt

output

3 picks.txt

-o writes to a file, which may even be the input file, since shuf reads everything first.

Common pitfalls

sort -R is not a shuffle: it sorts by a random hash of each line, so duplicate lines always end up adjacent. Use shuf when duplicates must be spread out.
Without -n, shuf holds the whole input in memory before printing anything. With -n, GNU shuf can sample a large input without keeping all of it.
--random-source with a short file fails with an end-of-file error once the bytes run out. Use an endless stream such as <(yes seed).
shuf -i and -e cannot be combined with an input file.

GNU vs BSD and macOS

The examples above ran the GNU/Linux version. These are the differences that matter on macOS and the BSDs (written from their documentation, not run here).

macOS and the BSDs have no shuf in the base system. Install GNU coreutils (where it may be named gshuf), or use sort -R on macOS, with the duplicate caveat above.

Options worth knowing

A selection, described in our own words. For the full list, see shuf(1) or shuf --help.

Option	What it does
`-n, --head-count COUNT`	Output at most COUNT lines.
`-e, --echo`	Treat each command-line argument as an input line.
`-i, --input-range LO-HI`	Shuffle the integers from LO to HI.
`-r, --repeat`	Allow output lines to repeat (sampling with replacement).
`-o, --output FILE`	Write the result to FILE.
`--random-source FILE`	Get random bytes from FILE, for repeatable results.
`-z, --zero-terminated`	Lines end with NUL instead of newline.

Examples

Common pitfalls

GNU vs BSD and macOS

Options worth knowing

Related