205

I have a couple of big files that I would like to compress. I can do this with for example

tar cvfj big-files.tar.bz2 folder-with-big-files

The problem is that I can't see any progress, so I don't have a clue how long it will take or anything like that. Using v I can at least see when each file is completed, but when the files are few and large this isn't the most helpful.

Is there a way I can get tar to show more detailed progress? Like a percentage done or a progress bar or estimated time left or something. Either for each single file or all of them or both.

0

17 Answers 17

218

I prefer oneliners like this:

tar cf - /folder-with-big-files -P | pv -s $(du -sb /folder-with-big-files | awk '{print $1}') | gzip > big-files.tar.gz

It will have output like this:

4.69GB 0:04:50 [16.3MB/s] [==========================>        ] 78% ETA 0:01:21

For OSX (from Kenji's answer)

tar cf - /folder-with-big-files -P | pv -s $(($(du -sk /folder-with-big-files | awk '{print $1}') * 1024)) | gzip > big-files.tar.gz

Explanation:

  • tar tarball tool
  • cf create file
  • - use stdout instead of a file (to be able to pipe the output to the next command)
  • /folder-with-big-files The input folder to zip
  • -P use absolute paths (not necessary, see comments)

pipe to

  • pv progress monitor tool
  • -s use the following size as the total data size to transfer (for % calculation)
    • $(...) evaluate the expression
    • du -sb /folder-with-big-files disk usage summarize in one line with bytes. Returns eg 8367213097 folder-with-big-files
    • pipe (|) to awk '{print $1}' which returns only the first part of the du output (the bytes, removing the foldername)

pipe to

  • gzip gzip compression tool
  • big-files.tar.gz output file name
20
  • 3
    On OSX, du does not take -b argument, needed to fallback to : $((du -sk /folder-with | awk '{print $1}') * 1024))
    – ıɾuǝʞ
    Nov 29, 2013 at 10:14
  • 10
    Nice, a one liner. Can you explain it? Or does it just magically work somehow?
    – Kissaki
    Oct 4, 2014 at 18:02
  • 4
    Ok, I have it pv $FILE.tgz | tar xzf - -C $DEST_DIR Nov 28, 2014 at 7:57
  • 4
    Note that the progress doesn't show until the du command finishes which could take a while depending on the size, complexity, and fragmentation of the directory.
    – Rooster242
    Jan 10, 2019 at 20:11
  • 13
    I'm a bit late to the party, but I was wondering why this answer suggests the use of the -P option on tar. That seems like bad advice, given that the OP didn't mention a need for absolute paths in the tarball (and having them can cause real headaches when it comes time to extract the archive). Feb 15, 2019 at 16:25
83

Use pv. To report the progress correctly, pv needs to know how many bytes you are throwing at it. So, the first step is to calculate the size (in kbyte). You can also completely drop the progress bar and just let pv tell you how many bytes it has seen; it would report a 'done that much and that fast'.

% SIZE=`du -sk folder-with-big-files | cut -f 1`

And then:

% tar cvf - folder-with-big-files | pv -p -s ${SIZE}k | \ 
     bzip2 -c > big-files.tar.bz2
6
  • Cool. pv doesn't seem to come with Mac OS X, but will try this out once I have a computer with MacPorts on it. Could you explain what you are doing there though? Not quite sure what the first line does exactly.
    – Svish
    Jul 28, 2010 at 12:10
  • 4
    first line: fetch info about how many bytes will be handled. second line: use the size from the first line to allow pv to render 'progress'. since you are piping data, pv does not know how many more bytes will come.
    – akira
    Jul 22, 2011 at 10:25
  • One addition: SIZE=$(($SIZE * 1000 / 1024)) - I don't know whether or not this is a quirk on my particular platform, so I'm not adding it to the answer: du returns size where 1 kb = 1024 bytes, while pv seems to be expecting 1 kb = 1000 bytes. (I'm on Ubuntu 10.04)
    – Izkata
    Dec 11, 2011 at 2:27
  • 2
    @lzkata you could always ask du to use your preferred blocksize, e.g. du -s --block-size=1000, or just work with plain bytes, e.g. drop the k's from the du and pv calls. Nevertheless, I would expect both to use 1024 unless told otherwise, e.g. the --si switch on du, for example.
    – Legolas
    Feb 23, 2012 at 11:05
  • 2
    or just drop the k-stuff and just use plain bytes (du -sb and pv -s without any modifier). that should end all the confusion.
    – akira
    Feb 23, 2012 at 11:10
35

better progress bar..

apt-get install pv dialog

(pv -n file.tgz | tar xzf - -C target_directory ) \
2>&1 | dialog --gauge "Extracting file..." 6 50

enter image description here

1
  • 2
    This is works for extraction, but you still need to do one of the more complicated commands for creation (which was the original question). It could still be combined with those; it's just more complicated.
    – Daniel H
    Aug 9, 2014 at 5:05
31

Check out the --checkpoint and --checkpoint-action options in the tar info page (as for my distribution, the description for these options is not contained in the man page → RTFI).

See https://www.gnu.org/software/tar/manual/html_section/checkpoints.html

With these (and maybe the functionality to write your own checkpoint command), you can calculate a percentage…

2
  • 3
    This should be the correct answer. Others just explain extra tools (not installed by default, besides) to achieve something similar. Nov 15, 2016 at 14:47
  • 1
    @Sardathrion Maybe because it's GNU-tar specific.
    – phk
    Feb 24, 2017 at 11:52
28

Inspired by helper’s answer

Another way is use the native tar options

FROMSIZE=`du -sk --apparent-size ${FROMPATH} | cut -f 1`;
CHECKPOINT=`echo ${FROMSIZE}/50 | bc`;
echo "Estimated: [==================================================]";
echo -n "Progess:   [";
tar -c --record-size=1K --checkpoint="${CHECKPOINT}" --checkpoint-action="ttyout=>" -f - "${FROMPATH}" | bzip2 > "${TOFILE}";
echo "]"

the result is like

Estimated: [==================================================]
Progess:   [>>>>>>>>>>>>>>>>>>>>>>>

a complete example here

4
  • This is an amazing approach - thanks for sharing it! The progress bar might not completely match the estimate bar more often than not but it's usually only a "off-by-one" error and given that it clearly states, that this is an estimate I'd stay that's still alright :)
    – Raven
    Aug 28, 2020 at 13:53
  • 1
    It's correct, is just an estimation. However, use the du --apparent-size option because it match better with the tar command result. I will update the answer with this option (and the referred link, take a look)
    – campisano
    Aug 28, 2020 at 17:35
  • I'm actually using this for the extraction of a compressed archive, so I am using xz --robot --list to get the uncompressed size and feed that into your solution. But thanks for the update nonetheless!
    – Raven
    Aug 29, 2020 at 6:51
  • This is a really great tar hack! I love it. A great suggestion is to modify to use [--->] style by: (a) using 49 in the CHECKPOINT, and (b) using --checkpoint-action="ttyout=\b->". That is a backspace. And a space after the [ .
    – not2qubit
    Oct 13, 2023 at 21:54
28

Using only tar

tar has the option (since v1.12) to print status information on signals using --totals=$SIGNO, e.g.:

tar --totals=USR1 -czf output.tar input.file
Total bytes written: 6005319680 (5.6GiB, 23MiB/s)

The Total bytes written: [...] information gets printed on every USR1 signal, e.g.:

pkill -SIGUSR1 tar

Source:

2
  • And to know the total original size, we can use du -hs /path. But how can we estimate the total bytes to be written when using the -z flag? I assume it would be less than the original size
    – lucidbrot
    Nov 11, 2019 at 17:07
  • 1
    I have asked a Q related to my comment on unix.se
    – lucidbrot
    Nov 12, 2019 at 19:27
6

Method based upon tqdm:

tar -v -xf tarfile.tar -C TARGET_DIR | tqdm --total $(tar -tvf tarfile.tar | wc -l) > /dev/null
4

You won't see how much is left, but you will see that it is progressing with a very simple addition to your command --checkpoint=1000

# tar  xf  file.tar --checkpoint=1000
tar: Read checkpoint 1000
tar: Read checkpoint 2000
tar: Read checkpoint 3000
tar: Read checkpoint 4000
tar: Read checkpoint 5000
tar: Read checkpoint 6000
...

More details on https://www.gnu.org/software/tar/manual/html_section/tar_25.html

3

Just noticed the comment about MacOS, and while I think the solution from @akira (and pv) is much neater I thought I'd chase a hunch and a quick playaround in my MacOS box with tar and sending it a SIGINFO signal. Funnily enough, it worked :) if you're on a BSD-like system, this should work, but on a Linux box, you might need to send a SIGUSR1, and/or tar might not work the same way.

The down side is that it will only provide you with an output (on stdout) showing you how far through the current file it is since I'm guessing it has no idea about how big the data stream it's getting is.

So yes, an alternative approach would be to fire up tar and periodically send it SIGINFOs anytime you want to know how far it's gotten. How to do this?

The ad-hoc, manual approach

If you want to be able to check status on an ad-hoc basis, you can hit control-T (as Brian Swift mentioned) in the relevant window which will send the SIGINFO signal across. One issue with that is it will send it to your entire chain I believe, so if you are doing:

% tar cvf - folder-with-big-files | bzip2 -c > big-files.tar.bz2

You will also see bzip2 report it's status along with tar:

a folder-with-big-files/big-file.imgload 0.79  cmd: bzip2 13325 running 
      14 0.27u 1.02s 

      adding folder-with-big-files/big-file.imgload (17760256 / 32311520)

This works nicely if you just want to check if that tar you're running is stuck, or just slow. You probably don't need to worry too much about formatting issues in this case, since it's only a quick check..

The sort of automated approach

If you know it's going to take a while, but want something like a progress indicator, an alternative would be to fire off your tar process and in another terminal work out it's PID and then throw it into a script that just repeatedly sends a signal over. For example, if you have the following scriptlet (and invoke it as say script.sh PID-to-signal interval-to-signal-at):

#!/bin/sh

PID=$1
INTERVAL=$2
SIGNAL=29      # excuse the voodoo, bash gets the translation of SIGINFO, 
               # sh won't..

kill -0 $PID   # invoke a quick check to see if the PID is present AND that
               # you can access it..

echo "this process is $$, sending signal $SIGNAL to $PID every $INTERVAL s"
while [ $? -eq 0 ]; do
     sleep $INTERVAL;
     kill -$SIGNAL $PID;    # The kill signalling must be the last statement
                            # or else the $? conditional test won't work
done
echo "PID $PID no longer accessible, tar finished?"

If you invoke it this way, since you're targeting only tar you'll get an output more like this

a folder-with-big-files/tinyfile.1
a folder-with-big-files/tinyfile.2
a folder-with-big-files/tinyfile.3
a folder-with-big-files/bigfile.1
adding folder-with-big-files/bigfile.1 (124612 / 94377241)
adding folder-with-big-files/bigfile.1 (723612 / 94377241)
...

which I admit, is kinda pretty.

Last but not least - my scripting is kinda rusty, so if anyone wants to go in and clean up/fix/improve the code, go for your life :)

5
  • 2
    If running tar on the command line, typing control-T will send it a SIGINFO. If this was in a script it would be done with kill -INFO pid Apr 23, 2012 at 4:58
  • Completely forgot about control-T, I clearly have gotten used to spamming too many console windows for my own good..
    – tanantish
    Apr 23, 2012 at 20:21
  • 1
    why can't I see -SIGINFO when doing kill -l Jun 12, 2013 at 2:32
  • @FelipeAlvarez Because you have probably not been on a BSD/MAC OS. See also: SIGINFO on GNU Linux (Arch Linux) missing
    – Murmel
    Nov 12, 2019 at 19:17
  • @tanantish You could get rid of the "voodoo" (your wording not mine ;P ) by using SIGNAL=$(kill -l SIGINFO), which would have the advantage of failing on systems without SIGINFO
    – Murmel
    Nov 12, 2019 at 19:21
2

Inspired by Noah Spurrier’s answer

function tar {
  local bf so
  so=${*: -1}
  case $(file "$so" | awk '{print$2}') in
  XZ) bf=$(xz -lv "$so" |
    perl -MPOSIX -ane '$.==11 && print ceil $F[5]/50688') ;;
  gzip) bf=$(gzip -l "$so" |
    perl -MPOSIX -ane '$.==2 && print ceil $F[1]/50688') ;;
  directory) bf=$(find "$so" -type f | xargs du -B512 --apparent-size |
    perl -MPOSIX -ane '$bk += $F[0]+1; END {print ceil $bk/100}') ;;
  esac
  command tar "$@" --blocking-factor=$bf \
    --checkpoint-action='ttyout=%u%\r' --checkpoint=1
}

Source

1
  • 23
    A little context and explanation maybe?
    – Kissaki
    Oct 4, 2014 at 18:03
2

On macOS, first make sure that you have all the commands available, and install the missing ones (e.g. pv) using brew.

If you only want to tar without compression, go with:

tar -c folder-with-big-files | pv -s $[$(du -sk folder-with-big-files | awk '{print $1}') * 1024] > folder-with-big-files.tar

If you want to compress, go with:

tar cf - folder-with-big-files -P | pv -s $[$(du -sk folder-with-big-files | awk '{print $1}') * 1024] | gzip > folder-with-big-files.tar.gz

Note: It may take a while before the progress bar appears. Try on a smaller folder first to make sure it works, then move to folder-with-big-files.

2

For simple extracting with pv.

pv mysql.tar.gz | tar -x

You'll get an output like this:

 249MiB 0:00:19 [14.0MiB/s] [==>                               ] 10% ETA 0:02:44

Here it is in action:

enter image description here

To install pv on macOS just use Homebrew with:

brew install pv

On other systems, you can look at the source repo here.

1

If you known the file number instead of total size of all of them:

an alternative (less accurate but suitable) is to use the -l option and send in the unix pipe the filenames instead of data content.

Let's have 12345 files into mydir, command is:

[myhost@myuser mydir]$ tar cfvz ~/mytarfile.tgz .|pv -s 12345 -l > /dev/null 

you can know such value in advance (because of your use case) or use some command like find+wc to discover it:

[myhost@myuser mydir]$ find | wc -l
12345
2
  • So, why not put this command into sub-command? =)
    – Kirby
    Jan 9, 2018 at 14:12
  • tar cfvz ~/mytarfile.tgz . | pv -s $(find . | wc -l) -l > /dev/null. Does it work for you?
    – Kirby
    Jan 9, 2018 at 14:18
1

better looking progress bar

Install the dialog and pv commands with

sudo apt-get install dialog pv

and then execute tar like this

(tar cf - /folder-with-big-files | pv -n -s $(du -sb /folder-with-big-files | awk '{print $1}') | gzip -9 > big-files.tar) 2>&1 | dialog --gauge 'Your backup is in progress...' 7 70
1

If you are ok with using 7z:

7z a example.tar example/

This will show the scanning drive stage

Scanning the drive:
2608M 139834 Scan  example/file.txt

as well as some other useful information to watch.

Scanning the drive:
157318 folders, 601997 files, 13683142277 bytes (13 GiB)           

Creating archive: example.tar

Items to compress: 759315

    3% 29587 + example/file.txt
0

Here are some numbers of a prometheus (metrics data) backup on Debian/buster AMD64:

root# cd /path/to/prometheus/
root# tar -cf - ./metrics | ( pv -p --timer --rate --bytes > prometheus-metrics.tar )

Canceled this job as there was not enough disk-space available.

Experimenting with zstd as compressor for tar with monitoring the progress using pv:

root# apt-get update
root# apt-get install zstd pv

root# tar -c --zstd -f - ./metrics | ( pv -p --timer --rate --bytes > prometheus-metrics.tar.zst )
10.2GiB 0:11:50 [14.7MiB/s]

root# du -s -h prometheus
62G    prometheus

root# du -s -h prometheus-metrics.tar.zst
11G    prometheus-metrics.tar.zst
0

In my daily use I don't need to know the exact percent-level progress of the operation, only if it is working and (sometimes) how much it is near completion.

I solve this need minimally by showing the number of files processed in its own line; in Bash:

let n=0; tar zcvf files.tgz directory | while read LINE; do printf "\r%d" $((n++)) ; done ; echo

As I use this a lot, I defined a function alias in .bashrc:

function pvl { declare -i n=0; while read L ; do printf "\r%d" $((++n)) ; done ; echo ; }

Then simply:

tar zcvf files.tgz directory | pvl

I can compute the number of files in advance if needed with find directory | wc -l (Or better using the same function shown [find directory | pvl] to squash my impatience!).

Another example, setting rights for a virtual website (after that, a chown -R is fast because the filenames are in the filesystem cache):

find /site -print -type d -exec chmod 2750 "{}" \; -o -type f -exec chmod 640 "{}" | pvl

It's true this lateral processing could slow the main operation, but I think printing a return character and a few digits cannot be too expensive (besides that, waiting for the next equal sign to appear or percent digit to change feels slow compared with the subjective blazing speed of changing digits!).

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .