183

So pulling open a file with cat and then using grep to get matching lines only gets me so far when I am working with the particular log set that I am dealing with. It need a way to match lines to a pattern, but only to return the portion of the line after the match. The portion before and after the match will consistently vary. I have played with using sed or awk, but have not been able to figure out how to filter the line to either delete the part before the match, or just return the part after the match, either will work. This is an example of a line that I need to filter:

2011-11-07T05:37:43-08:00 <0.4> isi-udb5-ash4-1(id1) /boot/kernel.amd64/kernel: [gmp_info.c:1758](pid 40370="kt: gmp-drive-updat")(tid=100872) new group: <15,1773>: { 1:0-25,27-34,37-38, 2:0-33,35-36, 3:0-35, 4:0-9,11-14,16-32,34-38, 5:0-35, 6:0-15,17-36, 7:0-16,18-36, 8:0-14,16-32,34-36, 9:0-10,12-36, 10-11:0-35, 12:0-5,7-30,32-35, 13-19:0-35, 20:0,2-35, down: 8:15, soft_failed: 1:27, 8:15, stalled: 12:6,31, 20:1 }

The portion I need is everything after "stalled".

The background behind this is that I can find out how often something stalls:

cat messages | grep stalled | wc -l

What I need to do is find out how many times a certain node has stalled (indicated by the portion before each colon after "stalled". If I just grep for that (ie 20:) it may return lines that have soft fails, but no stalls, which doesn't help me. I need to filter only the stalled portion so I can then grep for a specific node out of those that have stalled.

For all intents and purposes, this is a freebsd system with standard GNU core utils, but I cannot install anything extra to assist.

6
  • @Gilles, Odd how that didn't pop up when I searched, though I didn't use the title I eventually went with...but it didn't show up in the screen below my title. Anyway, that aside, that might get me where I want, though I need the entire line after the match, not the first word - but might not take much of a change.
    – MaQleod
    Nov 7, 2011 at 23:52
  • Its title sucked. I stole yours which is very nice. Take the sed solution and don't treat whitespace specially. Nov 7, 2011 at 23:55
  • @Gilles, that is something I'm not entirely sure how to do. I am still learning sed.
    – MaQleod
    Nov 8, 2011 at 0:06
  • similar to unix.stackexchange.com/questions/24089/… as well. Nov 8, 2011 at 0:43
  • 1
    @shaa0601 I don't understand your question, it's especially difficult to follow in a comment with no formatting. Ask a new, self-contained question. Aug 28, 2014 at 15:37

6 Answers 6

227

The canonical tool for that would be sed.

sed -n -e 's/^.*stalled: //p'

Detailed explanation:

  • -n means not to print anything by default.
  • -e is followed by a sed command.
  • s is the pattern replacement command.
  • The regular expression ^.*stalled: matches the pattern you're looking for, plus any preceding text (.* meaning any text, with an initial ^ to say that the match begins at the beginning of the line). Note that if stalled: occurs several times on the line, this will match the last occurrence.
  • The match, i.e. everything on the line up to stalled: , is replaced by the empty string (i.e. deleted).
  • The final p means to print the transformed line.

If you want to retain the matching portion, use a backreference: \1 in the replacement part designates what is inside a group \(…\) in the pattern. Here, you could write stalled: again in the replacement part; this feature is useful when the pattern you're looking for is more general than a simple string.

sed -n -e 's/^.*\(stalled: \)/\1/p'

Sometimes you'll want to remove the portion of the line after the match. You can include it in the match by including .*$ at the end of the pattern (any text .* followed by the end of the line $). Unless you put that part in a group that you reference in the replacement text, the end of the line will not be in the output.

As a further illustration of groups and backreferences, this command swaps the part before the match and the part after the match.

sed -n -e 's/^\(.*\)\(stalled: \)\(.*\)$/\3\2\1/p'

To get the part after the first occurrence of the string instead of last (for those lines where the string can occur several times), a common trick is to replace that string once with a newline character (which is the one character that won't occur inside a line), and then remove everything up to that newline:

sed -n '
  /stalled: / {
    s//\
/
    s/.*\n//p
  }'

With some sed implementations, the first s command can be written s//\n/ though that's not standard/portable.

11
  • 3
    @MaQleod Oh, it's waiting for input on standard input, which here is the terminal because you haven't redirected it. Here you'd do an input redirection sed … <messages, since you want to process data from a file. To act on data produced by another command, you'd use a pipe: somecommand | sed …. Nov 8, 2011 at 1:02
  • 2
    right, end of day blackout there. command works perfectly, thanks.
    – MaQleod
    Nov 8, 2011 at 16:37
  • 1
    Best sed explanation I've seen so far -- thanks! Sep 16, 2016 at 17:47
  • 1
    @ungalcrys Shorter version of what? This isn't equivalent to any of the commands in my answer. I'd recommend writing it as sed 's/^.*stalled//' since -r is specific to Linux and doesn't work on other systems such as macOS and here you aren't getting any benefit from it. Aug 9, 2017 at 10:19
  • 1
    @ungalcrys The difference is their behavior on non-matching lines. sed 's/^.*stalled//' prints them unchanged, my first command skips them. This may not matter for this particular question. Aug 9, 2017 at 10:37
126

The other canonical tool you already use: grep:

For example:

grep -o 'stalled.*'

Has the same result as the second option of Gilles:

sed -n -e 's/^.*\(stalled: \)/\1/p'

The -o flag returns the --only-matching part of the expression, so not the entire line which is - of course - normally done by grep.

To remove the "stalled :" from the output, we can use a third canonical tool, cut:

grep -o 'stalled.*' | cut -f2- -d:

The cut command uses delimiter : and prints field 2 till the end. It's a matter of preference of course, but the cut syntax I find very easy to remember.

7
  • 3
    Thanks for mentioning the -o option! I wanted to point out that grep doesn't recognize the \n as a newline, so your first example only matches to the first n character. For example, echo "Hello Anne" | grep -o 'A[^\n]*' returns the string A. However, echo "Hello Anne" | grep -o 'A.*' returns the expected Anne, since . matches any character except the newline.
    – adamlamar
    Mar 16, 2015 at 21:52
  • 2
    Note that the quotes around the cut delimiter -d':' are removed by @poige. I find it easier to remember with quotes, e.g. with -d' ' or -d';'. Jul 10, 2017 at 20:44
  • According to your finding it should be easier to remember to use quotes with -f 2 too. Seriously, why not?
    – poige
    Aug 26, 2017 at 10:26
  • Because a delimiter like a semi-colon ; rather than a colon : will be interpreted differently if not quoted. Of course that's logical behavior, but still I like to rely on muscle memory. I don't like to quote the delimiter one time but not the other time. Just personal preference, like I said before: easier to remember. Oct 7, 2017 at 18:09
  • the period that is part of the .* is needed, worked well for me: cat filename | grep 'Return only this line xyz text' | grep -o 'xyz.*' returns xyz text
    – ron
    Dec 12, 2017 at 19:01
11

Yet another canonical tool you considered awk could be used with the following line:

awk -F"stalled" '/stalled/{print $2}' messages

Detailed explanation:

  • -F defines a separator for the line, i.e., "stalled". Everything before the separator is addressed with $1 and everything after with $2.
  • /reg-ex/ Searches for the matching regular expression, in this case "stalled".
  • {print $<n>} - prints n column. Since your separator is defined as stalled, everything after stalled is considered to be the second column.
1
  • think for the awk solution - this solution works the best for me
    – B.Kocis
    Nov 19, 2021 at 9:42
4

I used ifconfig | grep eth0 | cut -f3- -d: to take this

    [root@MyPC ~]# ifconfig
    eth0  Link encap:Ethernet  HWaddr AC:B4:CA:DD:E6:F8
          inet addr:192.168.0.2  Bcast:192.168.0.255  Mask:255.255.255.0
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:78998810244 errors:1 dropped:0 overruns:0 frame:1
          TX packets:20113430261 errors:0 dropped:0 overruns:0 carrier:0
          collisions:0 txqueuelen:1000
          RX bytes:110947036025418 (100.9 TiB)  TX bytes:15010653222322 (13.6 TiB)

and make it look like this

    [root@MyPC ~]# ifconfig | grep eth0 | cut -f3- -d:
    C4:7A:4D:F6:B8
3
  • 3
    Does this answer the question? Mar 31, 2017 at 4:56
  • 2
    You can use cat /sys/class/net/*/address, no parsing required. Dec 13, 2017 at 16:58
  • 1
    if only C4:7A:4D:F6:B8 appeared in your initial code block
    – Zodzie
    Sep 2, 2020 at 21:24
1

Using Perl (i.e. Perl5) and Raku (previously known as Perl6):

Perl:

perl -pe 's/^.*stalled: //; #leaves non-matching and/or blank lines intact

Or:

perl -nE '/^.*stalled: (.*)/ and say $1;'  #removes non-matching lines

Raku:

raku -pe 's/^.*stalled\:\s//;' #leaves non-matching and/or blank lines intact

Or:

raku -ne '/^.*stalled\:\s (.*)/ and say ~$0;' #removes non-matching lines

OUTPUT (for 2nd Perl and 2nd Raku examples above):

12:6,31, 20:1 }

The code above is virtually identical between the two languages. The most significant difference is that in Raku all non-alnum/non-underscore characters must be escaped to be 'understood literally' by the Raku regex engine.

Other minor differences include the fact that:

  1. Raku changes capture numbering to start from $0 (Perl starts from $1),
  2. in Raku a leading ~ tilde is used to stringify the match object, and
  3. in Perl a -E commandline flag must be used to enable the say function.

http://www.wall.org/~larry/natural.html
https://www.perl.org/
https://www.raku.org/

6
  • 1
    Or perl -ne 'print if s/^.*stalled: //' or perl -ne 'print $& if /^.*stalled: \K.*/s' to preserve the original line delimiter if any. Sep 15, 2021 at 5:05
  • 1
    One advantage of using perl is that you can replace * with *? to get the part after the first occurrence of "stalled: " on the line. Sep 15, 2021 at 5:08
  • @StéphaneChazelas Very nice! AFAIK the Raku equivalents are raku -ne '.put if s/^.*stalled\:\s//' and raku -ne 'put ~$/ if /^.*stalled\:\s<(.*/;' . The second one can be written as: raku -ne 'put $/.Str if /^.* stalled\:\s <(.*)> /;', which may be a tad more readable. Sep 15, 2021 at 5:32
  • @StéphaneChazelas agreed on the non-greedy *? notation, but if there are multiple instances of "stalled: " it might be fun to try a non-greedy match earlier in the regex, à la: perl -ne 'print ${^POSTMATCH} if /^.*?stalled: /ps' . Sep 15, 2021 at 5:36
  • 1
    ITYM perl -ne 'print ${^POSTMATCH} if /stalled: /p' Sep 15, 2021 at 5:42
0

there seems to a simpler way. just do:

sed "s/installed.*//g"

which removes all the words after "installed".

for i in *
do
    se=$(echo $i|sed "s/---.*//g")
    echo $se
    mv "$i" $se
done
1
  • Your solution removes the portion after the pattern. The OP needed to return the portion after the pattern, not remove it.
    – Stephane
    Oct 5, 2021 at 12:31

Not the answer you're looking for? Browse other questions tagged or ask your own question.