Zizzencs
Zizzencs

Reputation: 5148

Best way to simulate "group by" from bash?

Suppose you have a file that contains IP addresses, one address in each line:

10.0.10.1
10.0.10.1
10.0.10.3
10.0.10.2
10.0.10.1

You need a shell script that counts for each IP address how many times it appears in the file. For the previous input you need the following output:

10.0.10.1 3
10.0.10.2 1
10.0.10.3 1

One way to do this is:

cat ip_addresses |uniq |while read ip
do
    echo -n $ip" "
    grep -c $ip ip_addresses
done

However it is really far from being efficient.

How would you solve this problem more efficiently using bash?

(One thing to add: I know it can be solved from perl or awk, I'm interested in a better solution in bash, not in those languages.)

ADDITIONAL INFO:

Suppose that the source file is 5GB and the machine running the algorithm has 4GB. So sort is not an efficient solution, neither is reading the file more than once.

I liked the hashtable-like solution - anybody can provide improvements to that solution?

ADDITIONAL INFO #2:

Some people asked why would I bother doing it in bash when it is way easier in e.g. perl. The reason is that on the machine I had to do this perl wasn't available for me. It was a custom built linux machine without most of the tools I'm used to. And I think it was an interesting problem.

So please, don't blame the question, just ignore it if you don't like it. :-)

Upvotes: 317

Views: 222328

Answers (17)

F. Hauri  - Give Up GitHub
F. Hauri - Give Up GitHub

Reputation: 71007

GROUP BY under

Regarding this SO thread, there are some different answer regarding different needs.

1. Counting IP as SO request (GROUP BY IP address).

As IP are easy to convert to single integer, for small bunch of address, if you need to repeat this kind of operation many time, using a pure function could be a lot more efficient!

Pure (no fork!)

There is a way, using a function. This way is very quick as there is no fork!...

countIp () { 
    local -a _ips=(); local _a
    while IFS=. read -a _a ;do
        ((_ips[_a<<24|${_a[1]}<<16|${_a[2]}<<8|${_a[3]}]++))
    done
    for _a in ${!_ips[@]} ;do
        printf "%.16s %4d\n" \
          $(($_a>>24)).$(($_a>>16&255)).$(($_a>>8&255)).$(($_a&255)) ${_ips[_a]}
    done
}

Note: IP addresses are converted to 32bits unsigned integer value, used as index for array. This use simple bash arrays!

time countIp < ip_addresses 
10.0.10.1    3
10.0.10.2    1
10.0.10.3    1
real    0m0.001s
user    0m0.004s
sys     0m0.000s

time sort ip_addresses | uniq -c
      3 10.0.10.1
      1 10.0.10.2
      1 10.0.10.3
real    0m0.010s
user    0m0.000s
sys     0m0.000s

On my host, doing so is a lot quicker than using forks, upto approx 1'000 addresses, but take approx 1 entire second when I'll try to sort'n count 10'000 addresses.

2. GROUP BY duplicates (files content)

By using checksum you could indentfy duplicate files somewhere:

find . -type f -exec sha1sum {} + |
    sort |
        sed '
          :a;
          $s/^[^ ]\+ \+//;
          N;
          s/^\([^ ]\+\) \+\([^ ].*\)\n\1 \+\([^ ].*\)$/\1 \2\o11\3/;
          ta;
          s/^[^ ]\+ \+//;
          P;
          D;
          ba
    '

This will print all duplicates, by line, separated by Tabulation ($'\t' or octal 011 ou could change /\1 \2\o11\3/; by /\1 \2|\3/; for using | as separator).

./b.txt   ./e.txt
./a.txt   ./c.txt    ./d.txt

Could be written as (with | as separator):

find . -type f -exec sha1sum {} + | sort | sed ':a;$s/^[^ ]\+ \+//;N;
  s/^\([^ ]\+\) \+\([^ ].*\)\n\1 \+\([^ ].*\)$/\1 \2|\3/;ta;s/^[^ ]\+ \+//;P;D;ba'

Pure way

By using nameref, you could build bash arrays holding all duplicates:

declare -iA sums='()'
while IFS=' ' read -r sum file ;do
    declare -n list=_LST_$sum
    list+=("$file")
    sums[$sum]+=1
done < <(
    find . -type f -exec sha1sum {} +
)

From there, you have a bunch of arrays holding all duplicates file name as separated element:

for i in ${!sums[@]};do
     declare -n list=_LST_$i
     printf "%d %d %s\n" ${sums[$i]} ${#list[@]} "${list[*]}"
done

This may output something like:

2 2 ./e.txt ./b.txt
3 3 ./c.txt ./a.txt ./d.txt

Where count of files by md5sum (${sums[$shasum]}) match count of element in arrays ${_LST_ShAsUm[@]}.

for i in ${!sums[@]};do
    declare -n list=_LST_$i
    echo ${list[@]@A}
done
declare -a _LST_22596363b3de40b06f981fb85d82312e8c0ed511=([0]="./e.txt" [1]="./b.txt")
declare -a _LST_f572d396fae9206628714fb2ce00f72e94f2258f=([0]="./c.txt" [1]="./a.txt" [2]="./d.txt")

Note that this method could handle spaces and special characters in filenames!

3. GROUP BY columns in a table

As efficient sample using awk was provided by Anonymous, here is a pure solution.

So you want to sumarize columns 3 to last column and group by columns 1 and 2, having table.txt looking like

US|A|1000|2000
US|B|1000|2000
US|C|1000|2000
UK|1|1000|2000
UK|1|1000|2000|3000
UK|1|1000|2000|3000|4000

For not too big tables, you could:

myfunc() {
    local -iA restabl='()';
    local IFS=+
    while IFS=\| read -ra ar; do
        restabl["${ar[0]}|${ar[1]}"]+="${ar[*]:2}"
    done
    for i in ${!restabl[@]} ;do
        printf '%s|%s\n' "$i" "${restabl[$i]}"
    done
}

Could ouput something like:

myfunc <table.txt 
UK|1|19000
US|A|3000
US|C|3000
US|B|3000

And to have table sorted:

myfunc() {
    local -iA restabl='()';
    local IFS=+ sorted=()
    while IFS=\| read -ra ar; do
        sorted[64#${ar[0]}${ar[1]}]="${ar[0]}|${ar[1]}"
        restabl["${ar[0]}|${ar[1]}"]+="${ar[*]:2}"
    done
    for i in ${sorted[@]} ;do
        printf '%s|%s\n' "$i" "${restabl[$i]}"
    done
}

Must return:

myfunc <table 
UK|1|19000
US|A|3000
US|B|3000
US|C|3000

Upvotes: 2

RARE Kpop Manifesto
RARE Kpop Manifesto

Reputation: 2915

A combination of awk + sort (with version sort flag) is probably fastest (if ur environment has awk at all):

echo "${input...}" |

{m,g}awk '{ __[$+_]++ } END { for(_ in __) { print "",+__[_],_ } }' FS='^$' OFS='\t' | 

gsort -t$'\t' -k 3,3 -V

Only the post GROUP-BY summary rows are being sent to the sorting utility - which is far less system intensive sort compared to pre-sorting the input rows for no reason.

For small inputs, e.g. fewer than 1000 rows or so, just directly sort|uniq -c it.

    3   10.0.10.1
    1   10.0.10.2
    1   10.0.10.3

Upvotes: 1

ufopilot
ufopilot

Reputation: 3985

Importing data to sqlite db and using sql syntax (just an other idea). I know it's too much for this example but would be useful for complex queries with multiple files (tables)

#!/bin/bash
trap clear_db EXIT
clear_db(){ rm -f "mydb$$"; }

# add header to input_file (IP)
INPUT_FILE=ips.txt

# import file into db
sqlite3 -csv mydb$$ ".import ${INPUT_FILE} mytable"

# using sql statements on table 'mytable' 
sqlite3 mydb$$ -separator " "  "SELECT IP, COUNT(*) FROM mytable GROUP BY IP;"

10.0.10.1 3
10.0.10.2 1
10.0.10.3 1

Upvotes: 2

TKF
TKF

Reputation: 638

This does not answer the count element of the original question, but this question is the first search engine result when searching for what I wanted to achieve, so I thought this may help someone as it relates to 'group by' functionality.

I wanted to order files based on groupings of them, where the presence of some string in the filename determined the group.

It uses a temporary grouping/ordering prefix which is removed after ordering; sed substitute expressions (s#pattern#replacement#g) match the target string and prepend an integer to the line corresponding to the desired sort order of that target string. Then, grouping prefix is removed with cut.

Note that the sed expressions could be joined (e.g. sed -e '<expr>; <expr>; <expr>;') but here they're split for readability.

It's not pretty and probably not fast (I'm dealing with <50 items) but it at-least conceptually simple and doesn't require learning awk.

#!/usr/bin/env bash

for line in $(find /etc \
    | sed -E -e "s#^(.*${target_string_A}.*)#${target_string_A_sort_index}:\1#;" \
    | sed -E -e "s#^(.*${target_string_B}.*)#${target_string_B_sort_index}:\1#;" \
    | sed -E -e "s#^/(.*)#00:/\1#;" \
    | sort \
    | cut -c4-
)
do
    echo "${line}"
done

e.g. Input

/this/is/a/test/a
/this/is/a/test/b
/this/is/a/test/c
/this/is/a/special/test/d
/this/is/a/another/test/e
#!/usr/bin/env bash

for line in $(find /etc \
    | sed -E -e "s#^(.*special.*)#10:\1#;" \
    | sed -E -e "s#^(.*another.*)#05:\1#;" \
    | sed -E -e "s#^/(.*)#00:/\1#;" \
    | sort \
    | cut -c4-
)
do
    echo "${line}"
done
/this/is/a/test/a
/this/is/a/test/b
/this/is/a/test/c
/this/is/a/another/test/e
/this/is/a/special/test/d

Upvotes: 1

Aron Curzon
Aron Curzon

Reputation: 2644

Most of the other solutions count duplicates. If you really need to group key value pairs, try this:

Here is my example data:

find . | xargs md5sum
fe4ab8e15432161f452e345ff30c68b0 a.txt
30c68b02161e15435ff52e34f4fe4ab8 b.txt
30c68b02161e15435ff52e34f4fe4ab8 c.txt
fe4ab8e15432161f452e345ff30c68b0 d.txt
fe4ab8e15432161f452e345ff30c68b0 e.txt

This will print the key value pairs grouped by the md5 checksum.

cat table.txt | awk '{print $1}' | sort | uniq  | xargs -i grep {} table.txt
30c68b02161e15435ff52e34f4fe4ab8 b.txt
30c68b02161e15435ff52e34f4fe4ab8 c.txt
fe4ab8e15432161f452e345ff30c68b0 a.txt
fe4ab8e15432161f452e345ff30c68b0 d.txt
fe4ab8e15432161f452e345ff30c68b0 e.txt

Upvotes: 3

zjor
zjor

Reputation: 1043

cat ip_addresses | sort | uniq -c | sort -nr | awk '{print $2 " " $1}'

this command would give you desired output

Upvotes: 18

Anonymous
Anonymous

Reputation: 371

for summing up multiple fields, based on a group of existing fields, use the example below : ( replace the $1, $2, $3, $4 according to your requirements )

cat file

US|A|1000|2000
US|B|1000|2000
US|C|1000|2000
UK|1|1000|2000
UK|1|1000|2000
UK|1|1000|2000

awk 'BEGIN { FS=OFS=SUBSEP="|"}{arr[$1,$2]+=$3+$4 }END {for (i in arr) print i,arr[i]}' file

US|A|3000
US|B|3000
US|C|3000
UK|1|9000

Upvotes: 37

Jadu Saikia
Jadu Saikia

Reputation:

I feel awk associative array is also handy in this case

$ awk '{count[$1]++}END{for(j in count) print j,count[j]}' ips.txt

A group by post here

Upvotes: 4

kairouan2020
kairouan2020

Reputation: 151

Solution ( group by like mysql)

grep -ioh "facebook\|xing\|linkedin\|googleplus" access-log.txt | sort | uniq -c | sort -n

Result

3249  googleplus
4211 linkedin
5212 xing
7928 facebook

Upvotes: 7

Francois Wolmarans
Francois Wolmarans

Reputation: 965

The quick and dirty method is as follows:

cat ip_addresses | sort -n | uniq -c

If you need to use the values in bash you can assign the whole command to a bash variable and then loop through the results.

PS

If the sort command is omitted, you will not get the correct results as uniq only looks at successive identical lines.

Upvotes: 75

wzzrd
wzzrd

Reputation: 620

I understand you are looking for something in Bash, but in case someone else might be looking for something in Python, you might want to consider this:

mySet = set()
for line in open("ip_address_file.txt"):
     line = line.rstrip()
     mySet.add(line)

As values in the set are unique by default and Python is pretty good at this stuff, you might win something here. I haven't tested the code, so it might be bugged, but this might get you there. And if you want to count occurrences, using a dict instead of a set is easy to implement.

Edit: I'm a lousy reader, so I answered wrong. Here's a snippet with a dict that would count occurences.

mydict = {}
for line in open("ip_address_file.txt"):
    line = line.rstrip()
    if line in mydict:
        mydict[line] += 1
    else:
        mydict[line] = 1

The dictionary mydict now holds a list of unique IP's as keys and the amount of times they occurred as their values.

Upvotes: 0

Diomidis Spinellis
Diomidis Spinellis

Reputation: 19385

The canonical solution is the one mentioned by another respondent:

sort | uniq -c

It is shorter and more concise than what can be written in Perl or awk.

You write that you don't want to use sort, because the data's size is larger than the machine's main memory size. Don't underestimate the implementation quality of the Unix sort command. Sort was used to handle very large volumes of data (think the original AT&T's billing data) on machines with 128k (that's 131,072 bytes) of memory (PDP-11). When sort encounters more data than a preset limit (often tuned close to the size of the machine's main memory) it sorts the data it has read in main memory and writes it into a temporary file. It then repeats the action with the next chunks of data. Finally, it performs a merge sort on those intermediate files. This allows sort to work on data many times larger than the machine's main memory.

Upvotes: 25

PolyThinker
PolyThinker

Reputation: 5218

You probably can use the file system itself as a hash table. Pseudo-code as follows:

for every entry in the ip address file; do
  let addr denote the ip address;

  if file "addr" does not exist; then
    create file "addr";
    write a number "0" in the file;
  else 
    read the number from "addr";
    increase the number by 1 and write it back;
  fi
done

In the end, all you need to do is to traverse all the files and print the file names and numbers in them. Alternatively, instead of keeping a count, you could append a space or a newline each time to the file, and in the end just look at the file size in bytes.

Upvotes: 3

nicerobot
nicerobot

Reputation: 9235

I'd have done it like this:

perl -e 'while (<>) {chop; $h{$_}++;} for $k (keys %h) {print "$k $h{$k}\n";}' ip_addresses

but uniq might work for you.

Upvotes: 1

Vinko Vrsalovic
Vinko Vrsalovic

Reputation: 340446

It seems that you have to either use a big amount of code to simulate hashes in bash to get linear behavior or stick to the quadratic superlinear versions.

Among those versions, saua's solution is the best (and simplest):

sort -n ip_addresses.txt | uniq -c

I found http://unix.derkeiler.com/Newsgroups/comp.unix.shell/2005-11/0118.html. But it's ugly as hell...

Upvotes: 4

Sudden Def
Sudden Def

Reputation: 10421

Sort may be omitted if order is not significant

uniq -c <source_file>

or

echo "$list" | uniq -c

if the source list is a variable

Upvotes: -9

Joachim Sauer
Joachim Sauer

Reputation: 308239

sort ip_addresses | uniq -c

This will print the count first, but other than that it should be exactly what you want.

Upvotes: 535

Related Questions