Search in large csv files - mysql

Question

The problem
I have thousands of csv files in a folder. Every file has 128,000 entries with four columns in each line.
From time to time (two times a day) I need to compare a list (10,000 entries) with all csv files. If one of the entries is identical with the third or fourth column of one of the csv files I need to write the whole csv row to an extra file.
Possible solutions
Grep
#!/bin/bash
getArray() {
array=()
while IFS= read -r line
do
array+=("$line")
done < "$1"
}
getArray "entries.log"
for e in "${array[#]}"
do
echo "$e"
/bin/grep $e ./csv/* >> found
done
This seems to work, but it lasts forever. After almost 48 hours the script checked only 48 entries of about 10,000.
MySQL
The next try was to import all csv files to a mysql database. But there I had problems with my table at around 50,000,000 entries.
So I wrote a script which created a new table after 49,000,000 entries and so I was able to import all csv files.
I tried to create an index on the second column but it always failed (timeout). To create the index before the import process wasn't possible, too. It slowed down the import to days instead of only a few hours.
The select statement was horrible, but it worked. Much faster than the "grep" solution but still to slow.
My question
What else can I try to search within the csv files?
To speed things up I copied all csv files to an ssd. But I hope there are other ways.

Answer 1

This is unlikely to offer you meaningful benefits, but some improvements to your script
use the built-in mapfile to slurp a file into an array:
mapfile -t array < entries.log
use grep with a file of patterns and appropriate flags.
I assume you want to match items in entries.log as fixed strings, not as regex patterns.
I also assume you want to match whole words.
grep -Fwf entries.log ./csv/*
This means you don't have to grep the 1000's of csv files 1000's of times (once for each item in entries.log). Actually this alone should give you a real meaningful performance improvement.
This also removes the need to read entries.log into an array at all.

Answer 2

In awk assuming all the csv files change, otherwise it would be wise to keep track of the already checked files. But first some test material:
$ mkdir test # the csvs go here
$ cat > test/file1 # has a match in 3rd
not not this not
$ cat > test/file2 # no match
not not not not
$ cat > test/file3 # has a match in 4th
not not not that
$ cat > list # these we look for
this
that
Then the script:
$ awk 'NR==FNR{a[$1];next} ($3 in a) || ($4 in a){print >> "out"}' list test/*
$ cat out
not not this not
not not not that
Explained:
$ awk ' # awk
NR==FNR { # process the list file
a[$1] # hash list entries to a
next # next list item
}
($3 in a) || ($4 in a) { # if 3rd or 4th field entry in hash
print >> "out" # append whole record to file "out"
}' list test/* # first list then the rest of the files
The script hashes all the list entries to a and reads thru the csv files looking for 3rd and 4th field entries in the hash outputing when there is a match.
If you test it, let me know how long it ran.

Answer 3

You can build a patterns file and then use xargs and grep -Ef to search for all patterns in batches of csv files, rather than one pattern at a time as in your current solution:
# prepare patterns file
while read -r line; do
printf '%s\n' "^[^,]+,[^,]+,$line,[^,]+$" # find value in third column
printf '%s\n' "^[^,]+,[^,]+,[^,]+,$line$" # find value in fourth column
done < entries.log > patterns.dat
find /path/to/csv -type f -name '*.csv' -print0 | xargs -0 grep -hEf patterns.dat > found.dat
find ... - emits a NUL-delimited list of all csv files found
xargs -0 ... - passes the file list to grep, in batches