3
\$\begingroup\$

I wrote a log parser in Python and Julia. The Python version takes about 8 seconds to process 350 files, while the Julia version takes about 8 seconds to compile, after which it takes over 20 seconds to process 350 log files.

I wrote the original parser in Python several months ago, with regex, a custom class, and lots of methods for detailed analysis. I recently wrote a Julia version with regex, a custom type, and only a single analysis function (the only one I regularly used). I then wrote a very trimmed-down version of each parser, using only basic string functions. I'm experienced with Python but not with Julia, and I can't figure out why the Julia version is so slow (even discounting the compilation time). If I can AoT compile the Julia script, dramatically reduce its run time, and package it into a native x86/64 executable, it might be worth using.

What am I doing wrong with this Julia script? How can I make it faster?

The Python version:

import datetime
import sys
import os
import tzlocal

min_date = tzlocal.get_localzone().localize(datetime.datetime(2002, 1, 1))
min_date -= min_date.utcoffset()
day = min_date
keys = ['Contraband', 'Dilithium', 'Dilithium Ore', 'Energy Credits']
dct = {key:0 for key in keys}
files = sorted(f for f in os.listdir() if f[:5]=="Chat_")
l = len(files)
try:
    with open('lastlog.txt') as f:
        lastlog = f.read()
except (FileNotFoundError, PermissionError):
    lastlog = ''

print('Date', *keys, sep='\t')
for counter,fname in enumerate(files):
    if fname >= lastlog:
        print('Processing {} out of {}...'.format(counter, l), end='\r', file=sys.stderr)
        with open(fname, encoding='utf-8-sig') as f:
            for line in f:
                if (line[26:70] == ",0,NumericReceived@,@,,,System]You received " and
                    ('Dilithium' in line or 'Energy Credits' in line)):
                    quantity,item = line.strip()[70:].split(None, 1)
                    quantity = int(quantity.replace(',', ''))
                    if item == 'Refined Dilithium':
                        item = 'Dilithium'
                elif line[26:66] == ",0,NumericReceived@,@,,,System]You sold ":
                    item = 'Energy Credits'
                    quantity = int(line.rsplit(None, 3)[1].replace(',', ''))
                elif line[26:78] == ",0,NumericConversionSuccess@,@,,,System]You refined ":
                    item = 'Dilithium'
                    quantity = int(line[78:-11].replace(',', ''))
                elif (line[26:56] == ",0,NumericLost@,@,,,System]You" and
                        (line[56:62] == " lost " or line[56:63] == " spent ") and 
                      ('Dilithium' in line or 'Energy Credits' in line)):
                    quantity,item = line[62:].strip().split(None, 1)
                    quantity = -int(quantity.replace(',', ''))
                    if item == 'Refined Dilithium':
                        item = 'Dilithium'
                elif line[26:83] == ',0,ItemReceived@,@,,,System]Items acquired: Contraband x ':
                    quantity = int(line[70:].rsplit(None, 1)[1].replace(',', ''))
                    item = 'Contraband'
                elif line[26:79] == ",0,ItemReceived@,@,,,System]Item acquired: Contraband":
                    quantity = 1
                    item = 'Contraband'
                else:
                    continue
                dt = tzlocal.get_localzone().localize(datetime.datetime(year=int(line[11:15]), month=int(line[15:17]),
                     day=int(line[17:19]), hour=int(line[20:22]), minute=int(line[22:24]), second=int(line[24:25])))
                dt -= dt.utcoffset()
                if dt.day != day.day and day != min_date:
                    print('{}-{}-{}'.format(day.year,
                                            day.month,
                                            day.day),
                          *(dct[k] for k in keys), sep='\t')
                    dct = {key:0 for key in keys}
                day = dt
                if item in dct:
                    dct[item] += quantity

print('{}-{}-{}'.format(dt.year,
                        dt.month,
                        dt.day),
      *(dct[k] for k in keys), sep='\t')

print('\nDone.', file=sys.stderr)

last = files[-1][5:15]
current = files[-1]
for fname in files[::-1]:
    current = fname
    if fname[5:15] != last:
        break

with open('lastlog.txt', 'w') as f:
    f.write(current)

The Julia version:

import TimeZones

function GetTotals()
    zone = TimeZones.localzone() # get local time zone
    min_date = TimeZones.astimezone(
                TimeZones.ZonedDateTime(
                    DateTime(2002,1,1), zone),
                        TimeZones.TimeZone("UTC")) # set local minimum date as UTC
    day = min_date
    k = String["Contraband", "Dilithium", "Dilithium Ore", "Energy Credits"]
    dct = Dict{String,Int64}(key=>0 for key in k)
    files = sort([f for f in readdir() if f[1:5]=="Chat_"])
    l = length(files)

    if isfile("lastlog.txt")
        lastlog = open("lastlog.txt") do f
            readstring(f)
        end
    else
        lastlog = ""
    end

    println("Date", '\t', join(k, '\t'))
    for (counter,fname) in enumerate(files)
        if fname >= lastlog # "Chat_2016-11-04" < fname < "Chat_2016-11-12" 
            print(STDERR, "Processing ", counter, " out of ", l, "...\r")
            open(fname) do f
                for line in eachline(f)
                    if length(line) > 70 && line[27:70] == ",0,NumericReceived@,@,,,System]You received " &&
                        (contains(line, "Dilithium") || contains(line, "Energy Credits"))
                        quantity,item = split(strip(line)[71:end], ' '; limit=2)
                        quantity = parse(Int, replace(quantity, ",", ""))
                        if item == "Refined Dilithium"
                            item = "Dilithium"
                        end
                    elseif length(line) > 66 && line[27:66] == ",0,NumericReceived@,@,,,System]You sold "
                        item = "Energy Credits"
                        quantity = parse(Int, replace(rsplit(line, ' '; limit=4)[2], ",", ""))
                    elseif length(line) > 78 && line[27:78] == ",0,NumericConversionSuccess@,@,,,System]You refined "
                        item = "Dilithium"
                        quantity = parse(Int, replace(line[79:end-13], ",", ""))
                    elseif length(line) > 63 && (line[27:56] == ",0,NumericLost@,@,,,System]You" &&
                            (line[57:62] == " lost " || line[57:63] == " spent ") && 
                          (contains(line, "Dilithium") || contains(line, "Energy Credits")))
                        quantity,item = split(strip(line[63:end]), ' '; limit=2)
                        quantity = -parse(Int, replace(quantity, ",", ""))
                        if item == "Refined Dilithium"
                            item = "Dilithium"
                        end
                    elseif length(line) > 83 && line[27:83] == ",0,ItemReceived@,@,,,System]Items acquired: Contraband x "
                        quantity = parse(Int, replace(rsplit(line[71:end], ' ', limit=2)[2], ",", ""))
                        item = "Contraband"
                    elseif length(line) > 79 && line[27:79] == ",0,ItemReceived@,@,,,System]Item acquired: Contraband"
                        quantity = 1
                        item = "Contraband"
                    else
                        continue
                    end
                    dt = TimeZones.astimezone(TimeZones.ZonedDateTime(DateTime(line[12:19]*line[21:26], "yyyymmddHHMMSS"), zone), TimeZones.TimeZone("UTC"))

                    current = TimeZones.floor(dt, Dates.Day)
                    if day != current && day != min_date
                        println(Dates.format(day, "yyyy-mm-dd"),
                                '\t', join((dct[key] for key in k), '\t'))
                        dct = Dict{String,Int64}(key=>0 for key in k)
                    end
                    day = current
                    if item in keys(dct)
                        dct[item] += quantity
                    end
                end
            end
        end
    end
    println(Dates.format(day, "yyyy-mm-dd"), '\t', join((dct[key] for key in k), '\t'))
    println(STDERR, "\nDone.")
    last = files[end][6:15]
    current = files[end]
    for fname in files[end:-1:1]
        current = fname
        if fname[6:15] != last
            break
        end
    end
    open("lastlog.txt", "w") do f
        write(f, current)
    end
end

GetTotals()
\$\endgroup\$
  • \$\begingroup\$ Are you sure it's type-stable? Have you checked @code_warntype? Also, v[a:b] in Julia creates a copy. You may want to reduce allocations by using views: @view v[a:b]. To handle compilation times, put the function in a module and precompile the module. \$\endgroup\$ – Chris Rackauckas Jan 2 '17 at 7:43
  • \$\begingroup\$ @ChrisRackauckas - I tried @code_warntype but didn't understand the output (I can post or link it if you'd like to see it). @view doesn't work on strings. I added __precompile__() and module MyModule to the top of the script (and end at the bottom), and I'm getting the impression from the doc that it'll only make a difference if this module is imported into another script, because it still has a long delay before producing output. \$\endgroup\$ – TigerhawkT3 Jan 2 '17 at 9:00
  • \$\begingroup\$ Please post the @code_warntype. (I notified some people in the Gitter channel about this too since strings aren't really my thing. But many people there could help you) \$\endgroup\$ – Chris Rackauckas Jan 2 '17 at 16:06
  • \$\begingroup\$ Could you put a tar of at least a sample of the log files you used, so that people can try different approaches and do profiling? \$\endgroup\$ – Scott Jones Jan 2 '17 at 18:49
  • \$\begingroup\$ @ChrisRackauckas - I've put the @code_warntype output into a zip file here, along with some sample log files. \$\endgroup\$ – TigerhawkT3 Jan 3 '17 at 1:10

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.