How do I create large CSVs in seconds? - bash

Question

I am trying to create 1000s of large CSVs rapidly. This function generates the CSVs:
function csvGenerator () {
for ((i=1; i<=$NUMCSVS; i++)); do
CSVNAME=$DIRNAME"-"$CSVPREFIX$i$CSVEXT
HEADERARRAY=()
if [[ ! -e $CSVNAME ]]; then #Only create csv file if it not exist
touch $CSVNAME
echo "file: "$CSVNAME "created at $(date)" >> ../status.txt
fi
for ((j=1; j<=$NUMCOLS; j++)); do
if (( j < $NUMCOLS )) ; then
HEADERNAME=$DIRNAME"-csv-"$i"-header-"$j", "
elif (( j == $NUMCOLS )) ; then
HEADERNAME=$DIRNAME"-csv-"$i"-header-"$j
fi
HEADERARRAY+=$HEADERNAME
done
echo $HEADERARRAY > $CSVNAME
for ((k=1; k<=$NUMROWS; k++)); do
ROWARRAY=()
for ((l=1; l<=$NUMCOLS; l++)); do
if (( l < $NUMCOLS )) ; then
ROWVALUE=$DIRNAME"-csv-"$i"-r"$k"c"$l", "
elif (( l == $NUMCOLS )) ; then
ROWVALUE=$DIRNAME"-csv-"$i"-r"$k"c"$l
fi
ROWARRAY+=$ROWVALUE
done
echo $ROWARRAY >> $CSVNAME
done
done
}
The script takes ~3 mins to generate a CSV with 100k rows and 70 cols. What do I need to do to generate these CSVs at the rate of 1 CSV/~10 seconds?

Answer 1

Let me start by saying that bash and "performant" don't usually go together in the same sentence. As other commentators suggested, awk may be a good choice that's adjacent in some senses.
I haven't yet had a chance to run your code, but it opens and closes the output file once per row — in this example, 100,000 times. Each time it must seek to the end of the file so that it can append the latest row.
Try pulling the actual generation (everything after for ((j=1; j<=$NUMCOLS; j++)); do) into a new function, like generateCsvContents. In that new function, don't reference $CSVNAME, and remove the redirections on the echo statements. Then, in the original function, call the new function and redirect its output to the filename. Roughly:
function csvGenerator () {
for ((i=1; i<=NUMCSVS; i++)); do
CSVNAME=$DIRNAME"-"$CSVPREFIX$i$CSVEXT
if [[ ! -e $CSVNAME ]]; then #Only create csv file if it not exist
echo "file: $CSVNAME created at $(date)" >> ../status.txt
fi
# This will create $CSVNAME if it doesn't yet exist
generateCsvContents > "$CSVNAME"
done
}
function generateCsvContents() {
HEADERARRAY=()
for ((j=1; j<=NUMCOLS; j++)); do
if (( j < NUMCOLS )) ; then
HEADERNAME=$DIRNAME"-csv-"$i"-header-"$j", "
elif (( j == NUMCOLS )) ; then
HEADERNAME=$DIRNAME"-csv-"$i"-header-"$j
fi
HEADERARRAY+=$HEADERNAME
done
echo $HEADERARRAY
for ((k=1; k<=NUMROWS; k++)); do
ROWARRAY=()
for ((l=1; l<=NUMCOLS; l++)); do
if (( l < NUMCOLS )) ; then
ROWVALUE=$DIRNAME"-csv-"$i"-r"$k"c"$l", "
elif (( l == NUMCOLS )) ; then
ROWVALUE=$DIRNAME"-csv-"$i"-r"$k"c"$l
fi
ROWARRAY+=$ROWVALUE
done
echo "$ROWARRAY"
done
}

Answer 2

"Not this way" is I think the answer.
There are a few problems here.
You're not using your arrays as arrays. When you treat them like strings, you affect only the first element in the array, which is misleading.
The way you're using >> causes the output file to be opened and closed once for every line. That's potentially wasteful.
You're not quoting your variables. In fact, you're quoting the stuff that doesn't need quoting, and not quoting the stuff that does.
Upper case variable names are not recommended, due to the risk of collision with system variables. ref
Bash isn't good at this. Really.
A cleaned up version of your function might look like this:
csvGenerator2() {
for (( i=1; i<=NUMCSVS; i++ )); do
CSVNAME="$DIRNAME-$CSVPREFIX$i$CSVEXT"
# Only create csv file if it not exist
[[ -e "$CSVNAME" ]] && continue
touch "$CSVNAME"
date "+[%F %T] created: $CSVNAME" | tee -a status.txt >&2
HEADER=""
for (( j=1; j<=NUMCOLS; j++ )); do
printf -v HEADER '%s, %s-csv-%s-header-%s' "$HEADER" "$DIRNAME" "$i" "$j"
done
echo "${HEADER#, }" > "$CSVNAME"
for (( k=1; k<=NUMROWS; k++ )); do
ROW=""
for (( l=1; l<=NUMCOLS; l++ )); do
printf -v ROW '%s, %s-csv-%s-r%sc%s' "$ROW" "$DIRNAME" "$i" "$k" "$l"
done
echo "${ROW#, }"
done >> "$CSVNAME"
done
}
(Note that I haven't switched the variables to lower case because I'm lazy, but it's still a good idea.)
And if you were to make something functionally equivalent in awk:
csvGenerator3() {
awk -v NUMCSVS="$NUMCSVS" -v NUMCOLS="$NUMCOLS" -v NUMROWS="$NUMROWS" -v DIRNAME="$DIRNAME" -v CSVPREFIX="$CSVPREFIX" -v CSVEXT="$CSVEXT" '
BEGIN {
for ( i=1; i<=NUMCSVS; i++) {
out=sprintf("%s-%s%s%s", DIRNAME, CSVPREFIX, i, CSVEXT)
if (!system("test -e " CSVNAME)) continue
system("date '\''+[%F %T] created: " out "'\'' | tee -a status.txt >&2")
comma=""
for ( j=1; j<=NUMCOLS; j++ ) {
printf "%s%s-csv-%s-header-%s", comma, DIRNAME, i, j > out
comma=", "
}
printf "\n" >> out
for ( k=1; k<=NUMROWS; k++ ) {
comma=""
for ( l=1; l<=NUMCOLS; l++ ) {
printf "%s%s-csv-%s-r%sc%s", comma, DIRNAME, i, k, l >> out
comma=", "
}
printf "\n" >> out
}
}
}
'
}
Note that awk does not suffer from the same open/closer overhead mentioned earlier with bash; when a file is used for output or as a pipe, it gets opened once and is left open until it is closed.
Comparing the two really highlights the choice you need to make:
$ time bash -c '. file; NUMCSVS=1 NUMCOLS=10 NUMROWS=100000 DIRNAME=2 CSVPREFIX=x CSVEXT=.csv csvGenerator2'
[2019-03-29 23:57:26] created: 2-x1.csv
real 0m30.260s
user 0m28.012s
sys 0m1.395s
$ time bash -c '. file; NUMCSVS=1 NUMCOLS=10 NUMROWS=100000 DIRNAME=3 CSVPREFIX=x CSVEXT=.csv csvGenerator3'
[2019-03-29 23:58:23] created: 3-x1.csv
real 0m4.994s
user 0m3.297s
sys 0m1.639s
Note that even my optimized bash version is only a little faster than your original code.

Answer 3

Refactoring your two inner for-loops to loops like this will save time:
for ((j=1; j<$NUMCOLS; ++j)); do
HEADERARRAY+=$DIRNAME"-csv-"$i"-header-"$j", "
done
HEADERARRAY+=$DIRNAME"-csv-"$i"-header-"$NUMCOLS