使用awk处理多个文件

Question

I've got to process lots of txt files (16 million of rows for each file) using awk. I've got to read for example ten files:

File #1:

en sample_1 200
en.n sample_2 10
en sample_3 10

File #2:

en sample_1 10
en sample_3 67

File #3:

en sample_1 1
en.n sample_2 10
en sample_4 20

...

I would like to have an output like this:

source title f1 f2 f3 sum(f1,f2,f3)

en sample_1 200 10 1 211
en.n sample_2 10 0 10 20
en sample_3 10 67 0 77
en sample_4 0 0 20 20 

Here my first version of code:

#! /bin/bash
clear
#var declaration
BASEPATH=<path_to_file>
YEAR="2014"
RES_FOLDER="processed"
FINAL_RES="2014_06_01"
#results folder creation
mkdir $RES_FOLDER
#processing
awk 'NF>0{a[$1" "$2]=a[$1" "$2]" "$3}END{for(i in a){print i a[i]}}' $BASEPATH/$YEAR/* > $RES_FOLDER/$FINAL_RES

And here my output:

en sample_1 200 10 1
en.n sample_2 10 10
en sample_3 10 67
en sample_4 20

I'm a little bit confused about how to put zero column where no occurrence is found and how to get the sum of all value. I know I've to use this:

{tot[$1" "$2]+=$3} END{for (key in tot) print key, tot[key]}

Hope someone will help. Thank you.

******** EDITED ********

I'm trying to achieve my result in a different kind of way. I create a bash script like this, It produces a sorted file with all of my keys, it's very huge, about 62 millions of record, I slice this file into pieces and I pass each piece to my awk script.

BASH:

#! /bin/bash
clear
FILENAME=<result>
BASEPATH=<base_path>
mkdir processed/slice
cat $BASEPATH/dataset/* | cut -d' ' -f1,2 > $BASEPATH/processed/aggr
sort -u -k2 $BASEPATH/processed/aggr > $BASEPATH/processed/sorted
split -d -l 1000000 processed/sorted processed/slice/slice-
echo $(date "+START PROCESSING DATE: %d/%m/%y - TIME: %H:%M:%S")
for filename in processed/slice/*; do
  awk -v filename="$filename" -f algorithm.awk dataset/* >> processed/$FILENAME
done
echo $(date "+END PROCESSING DATE: %d/%m/%y - TIME: %H:%M:%S")
rm $BASEPATH/processed/aggr
rm $BASEPATH/processed/sorted
rm -rf $BASEPATH/processed/slice

AWK:

BEGIN{
while(getline < filename){
 key=$1" "$2;
 sources[key];
 for(i=1;i<11;i++){
   keys[key"-"i] = "0";
 }
}
close(filename);
}
{
if(FNR==1){
 ARGIND++;
}
key=$1" "$2;
keys[key"-"ARGIND] = $3
}END{
for (s in sources) {
 sum = 0
 printf "%s", s
 for (j=1;j<11;j++) {
   printf "%s%s", OFS, keys[s"-"j]
   sum += keys[s"-"j]
 }
print " "sum
}
}

With awk I preallocate my final array, and reading dataset/* folder I populate its content. I've figured out that my bottleneck came from iterating on dataset folder by awk input (10 files with 16.000.000 lines each). Everything is working on a small set of data, but with real data, RAM (30GB) congested. Does anyone have any suggestions or advices? Thank you.

Answer 1

$ cat tst.awk
{
    key = $1" "$2
    keys[key]
    val[key,ARGIND] = $3
}
END {
    for (key in keys) {
        sum = 0
        printf "%s", key
        for (fileNr=1;fileNr<=ARGIND;fileNr++) {
            printf "%s%s", OFS, val[key,fileNr]+0
            sum += val[key,fileNr]
        }
        print sum
    }
}

$ awk -f tst.awk file1 file2 file3
en sample_4 0 0 2020
en.n sample_2 10 0 1020
en sample_1 200 10 1211
en sample_3 10 67 077

The above uses GNU awk for ARGIND, with other awks just add a line FNR==1{ARGIND++} at the start. Pipe the output to sort if necessary.

Answer 2

awk -vn="<source> <title>" 'function w(m,p){while(split(a[m],t)!=b+2)sub(p," 0&",a[m])}FNR<2{f=FILENAME;o=o?o" <"f">":"<"f">";q=q?q","f:f;++b}{a[$1" "$2]=a[$1" "$2]?a[$1" "$2]" "$NF:$0;w($1" "$2," [^ ]*$");c[$1" "$2]+=$NF}END{print n,o,"sum<("q")>";for(i in a){w(i,"$");print a[i],c[i]|"sort -k2"}}' *
<source> <title> <f1> <f2> <f3> sum<(f1,f2,f3)>
en sample_1 200 10 1 211
en.n sample_2 10 0 10 20
en sample_3 10 67 0 77
en sample_4 0 0 20 20

Answer 3

Since your files are quite large, you might want to use join -- it might be faster and/or use less memory. However it requires the files to be sorted and to have a single join field.

join -a1 -a2 -e0 -o0,1.2,2.2     <(sed $'s/ /\034/' file1 | sort) \
                                 <(sed $'s/ /\034/' file2 | sort) | 
join -a1 -a2 -e0 -o0,1.2,1.3,2.2 - \
                                 <(sed $'s/ /\034/' file3 | sort) | 
awk '{sub(/\034/," "); print $0, $3+$4+$5}' 

Explanation provided upon request