We Keep Coding

high RAM usage by eval function in python [duplicate]

This question already has an answer here:
eval vs json.loads memory consumption
(1 answer)
Closed 2 years ago.
i implement each line 1 (not line 0) as string from 2 files (1st ~30MB and 2nd ~50MB) where line 0 has just some information which i dont need atm. line 1 is a string array which has around 1.3E6 smaller arrays like that ['I1000009', 'A', '4024', 'A'] as information in it.
[[['I1000009', 'A', '4024', 'A'], ['I1000009', 'A', '6734', 'G'],...],[['H1000004', 'B', '4024', 'A'], ['L1000009', 'B', '6734', 'C'],...],[and so on],...]
both files are in the same way filled. thats the reason why the files are between 30 and 50MB big. i read that files with my .py script to have access to the single information which i need:
import sys
myID = sys.argv[1]
otherID = sys.argv[2]
samePath = '/home/srv/Dokumente/srv/'
FolderName = 'checkArrays/'
finishedFolder = samePath+'finishedAnalysis/'
myNewFile = samePath+FolderName+myID[0]+'/'+myID+'.txt'
otherFile = samePath+FolderName+otherID[0]+'/'+otherID+'.txt'
nameFileOKarray = '_array_goodData.txt'
import csv
import os
import re #for regular expressions
# Text 2 - Start
import operator # zum sortieren der csv files
# Text 2 - End
whereIsMyArray = 1
text_file = open(finishedFolder+myID+nameFileOKarray, "r")
line = text_file.readlines()[whereIsMyArray:];
myGoodFile = eval(line[0])
text_file.close()
text_file = open(finishedFolder+otherID+nameFileOKarray, "r")
line = text_file.readlines()[whereIsMyArray:];
otherGoodFile = eval(line[0])
text_file.close()
print(str(myGoodFile[0][0][0]))
print(str(otherGoodFile[0][0][0]))
the problem what i have is, that if i start my .py script over the shell:
python3 checkarr_v1.py 44 39
the RAM of my 4GB pi server increase to the limit of RAM and Swap and dies. then i tried to start the .py script on a 32Gb RAM server and look at that it worked, but the usage of the RAM is really huge. see pics
(slack mode) overview of normal usage of RAM and CPU:
slackmode
(startsequence) overview in highest usage of RAM ~6GB and CPU: highest point
then it goes up and down after for ~1min: 1.2Gb to 3.6Gb then to 1.7Gb then to 1Gb and then the script finish ~1min and the right output was shown.
can you help me to understand if there is a better way to solve that for an 4Gb raspberry pi? is that a better way to write the 2 files, because the [",] symbols took also there spaces in the file? Is that a better solution as the eval function is to implement that string to an array? sry for that questions, but i cant understand why the 80MB files increase the RAM to around 6Gb. that sounds that i make something wrong. br and thx

1.3E9 arrays is going to be lots and lots of bytes if you read that into your application, no matter what you do.
I don't know if your code does what you actually want to do, but you're only ever using the first data item. If that's what you want to do, then don't read the whole file, just read that first part.
But also: I would advice against using "eval" for deserializing data.
The built-in json module will give data in almost the same format (if you control the input format).
Still, in the end: If you want to hold that much data in your program, you're looking at many GB of memory usage.
If you just want to process it, I'd take a more iterative approach and do a little at the time rather than to swallow the whole files. Especially with limited resources.
Update: I See now that it's 1.3e6, not 1.3e9 entries. Big difference. :-) Then json data should be okay. On my machine a list of 1.3M ['RlFKUCUz', 'A', '4024', 'A'] takes about 250MB.

Python text file read/write optimization

I have been working on this file i/o and have made some progress reading through the site and i am wondering what other ways this can be optimized. I am parsing a test infile of 10GB/30MM lines and writing to an outfile the fields which results in aprog 1.4GB clean file. Initially, it took 40m to run this process and i have reduced it to around 30m. Anyone have any other ideas to reduce this in python. Long term i will be looking to write this in C++ - i just have to learn the language first. thanks in advance.
with open(fdir+"input.txt",'rb',(50*(1024*1024))) as r:
w=open(fdir+"output0.txt",'wb',50*(1024*1024)))
for i,l in enumerate(r):
if l[42:44]=='25':
# takes fixed width line into csv line of only a few cols
wbun.append(','.join([
l[7:15],
l[26:35],
l[44:52],
l[53:57],
format(int(l[76:89])/100.0,'.02f'),
l[89:90],
format(int(l[90:103])/100.0,'.02f'),
l[193:201],
l[271:278]+'\n'
]))
# write about every 5MM lines
if len(wbun)==wsize:
w.writelines(wbun)
wbun=[]
print "i_count:",i
# splits about every 4GB
if (i+1)%fsplit==0:
w.close()
w=open(fdir+"output%d.txt"%(i/fsplit+1),'wb',50*(1024*1024)))
w.writelines(wbun)
w.close()

Try running it in Pypy (https://pypy.org), it will run without changes to your code, and probably faster.
Also, C++ might be an overkill, especially if you don't know it yet. Consider learning Go or D instead.

Python - Checking concordance between two huge text files

So, this one has been giving me a hard time!
I am working with HUGE text files, and by huge I mean 100Gb+. Specifically, they are in the fastq format. This format is used for DNA sequencing data, and consists of records of four lines, something like this:
#REC1
GATTTGGGGTTCAAAGCAGTATCGATCAAATAGTAAATCCATTTGTTCAACTCACAGTTT
+
!''*((((***+))%%%++)(%%%%).1***-+*''))*55CCF>>>>>>CCCCCCC65
#REC2
GATTTGGGGTTCAAAGCAGTATCGATCAAATAGTAAATCCATTTGTTCAACTCACAGTTT
+
!''*((((***+))%%%++)(%%%%).1***-+*''))**55CCF>>>>>>CCCCCCC65
.
.
.
For the sake of this question, just focus on the header lines, starting with a '#'.
So, for QA purposes, I need to compare two such files. These files should have matching headers, so the first record in the other file should also have the header '#REC1', the next should be '#REC2' and so on. I want to make sure that this is the case, before I proceed to heavy downstream analyses.
Since the files are so large, a naive iteration a string comparisson would take very long, but this QA step will be run numerous times, and I can't afford to wait that long. So I thought a better way would be to sample records from a few points in the files, for example every 10% of the records. If the order of the records is messed up, I'd be very likely to detect it.
So far, I have been able to handle such files by estimating the file size and than using python's file.seek() to access a record in the middle of the file. For example, to access a line approximately in the middle, I'd do:
file_size = os.stat(fastq_file).st_size
start_point = int(file_size/2)
with open(fastq_file) as f:
f.seek(start_point)
# look for the next beginning of record, never mind how
But now the problem is more complex, since I don't know how to coordinate between the two files, since the bytes location is not an indicator of the line index in the file. In other words, how can I access the 10,567,311th lines in both files to make sure they are the same, without going over the whole file?
Would appreciate any ideas\hints. Maybe iterating in parallel? but how exactly?
Thanks!

Sampling is one approach, but you're relying on luck. Also, Python is the wrong tool for this job. You can do things differently and calculate an exact answer in a still reasonably efficient way, using standard Unix command-line tools:
Linearize your FASTQ records: replace the newlines in the first three lines with tabs.
Run diff on a pair of linearized files. If there is a difference, diff will report it.
To linearize, you can run your FASTQ file through awk:
$ awk '\
BEGIN { \
n = 0; \
} \
{ \
a[n % 4] = $0; \
if ((n+1) % 4 == 0) { \
print a[0]"\t"a[1]"\t"a[2]"\t"a[3]; \
} \
n++; \
}' example.fq > example.fq.linear
To compare a pair of files:
$ diff example_1.fq.linear example_2.fq.linear
If there's any difference, diff will find it and tell you which FASTQ record is different.
You could just run diff on the two files directly, without doing the extra work of linearizing, but it is easier to see which read is problematic if you first linearize.
So these are large files. Writing new files is expensive in time and disk space. There's a way to improve on this, using streams.
If you put the awk script into a file (e.g., linearize_fq.awk), you can run it like so:
$ awk -f linearize_fq.awk example.fq > example.fq.linear
This could be useful with your 100+ Gb files, in that you can now set up two Unix file streams via bash process substitutions, and run diff on those streams directly:
$ diff <(awk -f linearize_fq.awk example_1.fq) <(awk -f linearize_fq.awk example_2.fq)
Or you can use named pipes:
$ mkfifo example_1.fq.linear
$ mkfifo example_2.fq.linear
$ awk -f linearize_fq.awk example_1.fq > example_1.fq.linear &
$ awk -f linearize_fq.awk example_2.fq > example_2.fq.linear &
$ diff example_1.fq.linear example_2.fq.linear
$ rm example_1.fq.linear example_2.fq.linear
Both named pipes and process substitutions avoid the step of creating extra (regular) files, which could be an issue for your kind of input. Writing linearized copies of 100+ Gb files to disk could take a while to do, and those copies could also use disk space you may not have much of.
Using streams gets around those two problems, which makes them very useful for handling bioinformatics datasets in an efficient way.
You could reproduce these approaches with Python, but it will almost certainly run much slower, as Python is very slow at I/O-heavy tasks like these.

Iterating in parallel might be the best way to do this in Python. I have no idea how fast this will run (a fast SSD will probably be the best way to speed this up), but since you'll have to count newlines in both files anyway, I don't see a way around this:
with open(file1) as f1, open(file2) as f2:
for l1, l2 in zip(f1,f2):
if l1.startswith("#REC"):
if l1 != l2:
print("Difference at record", l1)
break
else:
print("No differences")
This is written for Python 3 where zip returns an iterator; in Python 2, you need to use itertools.izip() instead.

Have you looked into using the rdiff command.
The upsides of rdiff are:
with the same 4.5GB files, rdiff only ate about 66MB of RAM and scaled very well. It never crashed to date.
it is also MUCH faster than diff.
rdiff itself combines both diff and patch capabilities, so you can create deltas and apply them using the same program
The downsides of rdiff are:
it's not part of standard Linux/UNIX distribution – you have to
install the librsync package.
delta files rdiff produces have a slightly different format than diff's.
delta files are slightly larger (but not significantly enough to care).
a slightly different approach is used when generating a delta with rdiff, which is both good and bad – 2 steps are required. The
first one produces a special signature file. In the second step, a
delta is created using another rdiff call (all shown below). While
the 2-step process may seem annoying, it has the benefits of
providing faster deltas than when using diff.
See: http://beerpla.net/2008/05/12/a-better-diff-or-what-to-do-when-gnu-diff-runs-out-of-memory-diff-memory-exhausted/

import sys
import re
""" To find of the difference record in two HUGE files. This is expected to
use of minimal memory. """
def get_rec_num(fd):
""" Look for the record number. If not found return -1"""
while True:
line = fd.readline()
if len(line) == 0: break
match = re.search('^#REC(\d+)', line)
if match:
num = int(match.group(1))
return(num)
return(-1)
f1 = open('hugefile1', 'r')
f2 = open('hugefile2', 'r')
hf1 = dict()
hf2 = dict()
while f1 or f2:
if f1:
r = get_rec_num(f1)
if r < 0:
f1.close()
f1 = None
else:
# if r is found in f2 hash, no need to store in f1 hash
if not r in hf2:
hf1[r] = 1
else:
del(hf2[r])
pass
pass
if f2:
r = get_rec_num(f2)
if r < 0:
f2.close()
f2 = None
else:
# if r is found in f1 hash, no need to store in f2 hash
if not r in hf1:
hf2[r] = 1
else:
del(hf1[r])
pass
pass
print('Records found only in f1:')
for r in hf1:
print('{}, '.format(r));
print('Records found only in f2:')
for r in hf2:
print('{}, '.format(r));

Both answers from #AlexReynolds and #TimPietzcker are excellent from my point of view, but I would like to put my two cents in. You also might want to speed up your hardware:
Raplace HDD with SSD
Take n SSD's and create a RAID 0. In the perfect world you will get n times speed up for your disk IO.
Adjust the size of chunks you read from the SSD/HDD. I would expect, for instance, one 16 MB read to be executed faster than sixteen 1 MB reads. (this applies to a single SSD, for RAID 0 optimization one has to take a look at RAID controller options and capabilities).
The last option is especially relevant to NOR SSD's. Don't pursuit the minimal RAM utilization, but try to read as much as it needs to keep your disk reading fast. For instance, parallel reads of single rows from two files can probably speed down reading - imagine an HDD where two rows of the two files are always on the same side of the same magnetic disk(s).

Python: Read huge number of lines from stdin

I'm trying to read a huge amount of lines from standard input with python.
more hugefile.txt | python readstdin.py
The problem is that the program freezes as soon as i've read just a single line.
print sys.stdin.read(8)
exit(1)
This prints the first 8 bytes but then i expect it to terminate but it never does. I think it's not really just reading the first bytes but trying to read the whole file into memory.
Same problem with sys.stdin.readline()
What i really want to do is of course to read all the lines but with a buffer so i don't run out of memory.
I'm using python 2.6

This should work efficiently in a modern Python:
import sys
for line in sys.stdin:
# do something...
print line,
You can then run the script like this:
python readstdin.py < hugefile.txt

Back in the day, you had to use xreadlines to get efficient huge line-at-a-time IO -- and the docs now ask that you use for line in file.
Of course, this is of assistance only if you're actually working on the lines one at a time. If you're just reading big binary blobs to pass onto something else, then your other mechanism might be as efficient.

Reading an entire binary file into Python

I need to import a binary file from Python -- the contents are signed 16-bit integers, big endian.
The following Stack Overflow questions suggest how to pull in several bytes at a time, but is this the way to scale up to read in a whole file?
Reading some binary file in Python
Receiving 16-bit integers in Python
I thought to create a function like:
from numpy import *
import os
def readmyfile(filename, bytes=2, endian='>h'):
totalBytes = os.path.getsize(filename)
values = empty(totalBytes/bytes)
with open(filename, 'rb') as f:
for i in range(len(values)):
values[i] = struct.unpack(endian, f.read(bytes))[0]
return values
filecontents = readmyfile('filename')
But this is quite slow (the file is 165924350 bytes). Is there a better way?

Use numpy.fromfile.

I would directly read until EOF (it means checking for receiving an empty string), removing then the need to use range() and getsize.
Alternatively, using xrange (instead of range) should improve things, especially for memory usage.
Moreover, as Falmarri suggested, reading more data at the same time would improve performance quite a lot.
That said, I would not expect miracles, also because I am not sure a list is the most efficient way to store all that amount of data.
What about using NumPy's Array, and its facilities to read/write binary files? In this link there is a section about reading raw binary files, using numpyio.fread. I believe this should be exactly what you need.
Note: personally, I have never used NumPy; however, its main raison d'etre is exactly handling of big sets of data - and this is what you are doing in your question.

You're reading and unpacking 2 bytes at a time
values[i] = struct.unpack(endian,f.read(bytes))[0]
Why don't you read like, 1024 bytes at a time?

I have had the same kind of problem, although in my particular case I have had to convert a very strange binary format (500 MB) file with interlaced blocks of 166 elements that were 3-bytes signed integers; so I've had also the problem of converting from 24-bit to 32-bit signed integers that slow things down a little.
I've resolved it using NumPy's memmap (it's just a handy way of using Python's memmap) and struct.unpack on large chunk of the file.
With this solution I'm able to convert (read, do stuff, and write to disk) the entire file in about 90 seconds (timed with time.clock()).
I could upload part of the code.

I think the bottleneck you have here is twofold.
Depending on your OS and disc controller, the calls to f.read(2) with f being a bigish file are usually efficiently buffered -- usually. In other words, the OS will read one or two sectors (with disc sectors usually several KB) off disc into memory because this is not a lot more expensive than reading 2 bytes from that file. The extra bytes are cached efficiently in memory ready for the next call to read that file. Don't rely on that behavior -- it might be your bottleneck -- but I think there are other issues here.
I am more concerned about the single byte conversions to a short and single calls to numpy. These are not cached at all. You can keep all the shorts in a Python list of ints and convert the whole list to numpy when (and if) needed. You can also make a single call struct.unpack_from to convert everything in a buffer vs one short at a time.
Consider:
#!/usr/bin/python
import random
import os
import struct
import numpy
import ctypes
def read_wopper(filename,bytes=2,endian='>h'):
buf_size=1024*2
buf=ctypes.create_string_buffer(buf_size)
new_buf=[]
with open(filename,'rb') as f:
while True:
st=f.read(buf_size)
l=len(st)
if l==0:
break
fmt=endian[0]+str(l/bytes)+endian[1]
new_buf+=(struct.unpack_from(fmt,st))
na=numpy.array(new_buf)
return na
fn='bigintfile'
def createmyfile(filename):
bytes=165924350
endian='>h'
f=open(filename,"wb")
count=0
try:
for int in range(0,bytes/2):
# The first 32,767 values are [0,1,2..0x7FFF]
# to allow testing the read values with new_buf[value<0x7FFF]
value=count if count<0x7FFF else random.randint(-32767,32767)
count+=1
f.write(struct.pack(endian,value&0x7FFF))
except IOError:
print "file error"
finally:
f.close()
if not os.path.exists(fn):
print "creating file, don't count this..."
createmyfile(fn)
else:
read_wopper(fn)
print "Done!"
I created a file of random shorts signed ints of 165,924,350 bytes (158.24 MB) which comports to 82,962,175 signed 2 byte shorts. With this file, I ran the read_wopper function above and it ran in:
real 0m15.846s
user 0m12.416s
sys 0m3.426s
If you don't need the shorts to be numpy, this function runs in 6 seconds. All this on OS X, python 2.6.1 64 bit, 2.93 gHz Core i7, 8 GB ram. If you change buf_size=1024*2 in read_wopper to buf_size=2**16 the run time is:
real 0m10.810s
user 0m10.156s
sys 0m0.651s
So your main bottle neck, I think, is the single byte calls to unpack -- not your 2 byte reads from disc. You might want to make sure that your data files are not fragmented and if you are using OS X that your free disc space (and here) is not fragmented.
Edit I posted the full code to create then read a binary file of ints. On my iMac, I consistently get < 15 secs to read the file of random ints. It takes about 1:23 to create since the creation is one short at a time.