Bigbadwolf

30 agosto 20071

?Who's afraid

[email protected]0830 DADA Spa., Firenze, Italy

of the

Big Bad Wolf

mailto:[email protected]

30 agosto 20073

Debugging web applications● Debugging is hard● Complex, legacy applications...● Superstition● Panic● A metaphor involving pigs and wolves to the

rescue!

30 agosto 20074

This metaphor is brought to you by

● firenze.pm● perl.it● beer

– (Belhaven Best, James Pub, Firenze)

30 agosto 20075

What happens when a little piggy goes in the forest?

The Forest

30 agosto 20076


The ForestCGI

30 agosto 20077


The ForestCGI

mod_perl

30 agosto 20078


The ForestCGI

mod_perl

Perl

30 agosto 20079


The ForestCGI

mod_perlPerl

DBI

30 agosto 200710


The ForestCGI

mod_perlPerl

DBI

modules

30 agosto 200711


The ForestCGI

mod_perlPerl

DBI

modules

ourcode

30 agosto 200712

The Forest● And more!

– Network– HTML– XML– 3rd party services– AJAX– Configuration– Unix permissions

30 agosto 200713

Werewolves!

30 agosto 200714

Werewolves!

There!Wolves!

30 agosto 200715


30 agosto 200716


The Forestpig

16

30 agosto 200717


The Forest

YARGGH!

30 agosto 200718


The Forest

404

30 agosto 200719


The Forest another pig

30 agosto 200720


The Forest

30 agosto 200721


● It never came out!● It came out limping and wounded● Dead! Gored by werewolves!● Takes a long time to come out

30 agosto 200722


● It never came out!● It came out limping and wounded

– Part of the functionality is broken– Information wrong

● Dead! Gored by werewolves!– User sees raw error on screen

● Takes a long time to come out– optimization needed

30 agosto 200723

Werewolves!● Round up the villagers for a lynchmob!● (Bring a pitchfork!)● Burn them!

– (maybe burn a few innocent villagers along the way too!)

30 agosto 200724

Engineers● Luckily things are (a little) better if we know

what we're doing

30 agosto 200725

We shouldn't see this

Scary Forest(eeek!)

30 agosto 200726

But this

30 agosto 200727

Logging● Trace the piggy as he goes through the forest

27

30 agosto 200728

Logging

The Forestpig

PIG1: I'm scared

30 agosto 200729

Logging

The Forest

PIG1: I'm scaredPIG1: I'm very scared

30 agosto 200730

Logging

The Forest

PIG1: I'm scaredPIG1: I'm very scaredPIG1: This is Cruelty to Animals!

30 agosto 200731

Logging

The Forest

PIG1: I'm scaredPIG1: I'm very scaredPIG1: This is Cruelty to Animals!PIG1: Speaking to the Database

30 agosto 200732

Logging

The Forest

PIG1: I'm scaredPIG1: I'm very scaredPIG1: This is Cruelty to Animals!PIG1: Speaking to the DatabasePIG1: Look over there, a wol...

30 agosto 200733

Logging (print statements)● Print statements

– print “I AM HERE!”;

30 agosto 200734

Logging (warnings)● warn

– warn “THIS HAPPENED!”;

– (this will end up in the Apache error log● /var/log/apache2/error.log(.yyyy.mm.dd)

● errors also go to Apache error log– always check the error log

30 agosto 200735

Logging (apache error log)● $ tail f error.log.20070411

● | grep “Media” # your module● | grep “HC” # your identifier

● Point to the right machine– edit /etc/hosts ( ...\drivers\etc\hosts on Windows)– send to a server you have an account on

– (restart Firefox, sigh)

30 agosto 200736

Logging (apache error log)● $ tail f error.log.20070411

● | grep v “uninitialized”

● Too many warnings!● Hard to see what are real errors● (Also annoys the Sysadmins)

30 agosto 200737

Logging (print statements)● Print to HTML

– contenttype: text/html (or text/plain)– <pre> $info </pre>

– You can wrap this in a call● $this>printDebug(“pippo”);

● But this messes up headers!● Label your own piggy!

– $this>printDebug( “HC(contacts): pippo”);

30 agosto 200738

Logging (data structures)● Complex data structures

– Data::Dumper

– $VAR1 = { 'y' => 2, 'x' => 1 };

30 agosto 200739

Logging (manual trace)● When did the script crash?

– $this>printDebug(“HC 1 got here”);

– do something– $this>printDebug(“HC 2 got here”);



30 agosto 200740

Logging (manual trace)● Yes, that's quite stupid

– (but it beats using a pitchfork)

30 agosto 200741

Logging (end of html debug)

● To avoid messing up headers you can get your framework to print a development footer at the bottom of the HTML page instead

30 agosto 200742

Logging (to file)● save to a file

– open file– write– close file

● Lots of people recommend Log::Log4perl

30 agosto 200743

Logging ● use Log::Log4Perl qw(:easy);

● Log::Log4Perl>easy_init($ERROR);

● my $log = Log::Log4perl>get_logger('foo');

● $log>debug(“Only show this to logs”);

● $log>info(“This is more important”);

● $log>warn(“Oh dear”);

● $log>error(“Eek! This error on screen”);

● $log>fatal(“Werewolves!”);

30 agosto 200744

Logging● Log files eventually get really big● Sysadmins will want to rotate them● This may mean appending

– .yyyy.mm.ddto the filenames

30 agosto 200745

Error handling

The Forest

another pig

If a pig screams in the forest and

there's noone there to hear it... did it

really die?

45

30 agosto 200746

Error handling● Sometimes it's OK to die

– use CGI::Carp qw( fatalsToBrowser );

– die “HC: eeeeeek! The piggy is dead”;

30 agosto 200747

Error handling (CSI!)● But how did the piggy die?

– Bad parameter at TestModule.pm line 10.● If only we knew who had called TestModule!

30 agosto 200748

Error handling (Carp)● package TestModule;

● use Carp qw( croak );

● ... croak “Bad parameter”;...

– Bad parameter at test.pl line 7

30 agosto 200749

Error handling (stacktraces)● What if it was another subroutine or module

called?● use Carp qw(confess);

● confess “Ooops!”;

● Ooops! at Pippo.pm line 13 Pippo::three() called at Pluto.pm line 10 Pluto::two() called at Pinco.pm line 11 Pinco::one() called at test.pl line 7

30 agosto 200750

Error handling (framework)● If your framework handles this, you may get

stacktraces by default– They aren't always easy to read, but they contain

lots of useful information!

– Catalyst– In house framwork

30 agosto 200751

Error handling (Database)● We don't want the user seeing DB errors in live!

– BUT– We probably want to see them in staging!– classic idiom

● $sth>execute() or die $sth>errstr;

– You may want to:● have a “fatal” function which either dies informatively (in

staging) or produces an appropriate error document (in live)

30 agosto 200752

Error handling (Database)● Do you really want to fail silently...?

– $sth>execute()or warn $sth>errstr;

– (or $logger>log($sth>errstr));

30 agosto 200753

Testing

The Forest

I hope someone

tested this parachute

54

30 agosto 200754

Burn the Witch!● Sometimes we think we know where the

problem is– Sometimes we're wrong

30 agosto 200755

How do we check● A test!

– does it weigh more or less than a duck?● use Test::More tests => 1;

– is( $possible_witch>weight(), $duck>weight(), “It's the witch!” );

30 agosto 200756

Testing in the forest● Sometimes we need to test:

– user clicks on link in browser, logs in– user puts in information– the script talks to the Database– the script talks to a 3rd party API– the script talks to a module which returns some

information– the HTML is processed

● (e.g. the forest)

30 agosto 200757

Testing in the forest● But...

– what if we suspect that the problem is with formatting of display name?

– MyModule:: createDisplayName();

– Just test that!

30 agosto 200758

See the trees...● Write a simple test

– use Test::More;

– use MyModule;

– is ($m>createDisplayName('robin'), “Robin Hood”, “display name returned correctly” );

● If the test returns correctly, then maybe the problem is before (or after)

● (we could use logging to help find out which!)

30 agosto 200759

See the trees...

● MVC!

30 agosto 200760

Testing● Helps clarify what the problem is● Helps clarify where the problem is

– (or isn't)

● Also– vampires are repelled by tests!

30 agosto 200761

Interactive Testing● Perl has a debugger● Handy for running little snippets of code to

check how it works

30 agosto 200762

Interactive Testing~$ perl debug

Enter h or `h h' for help, or `man perldebug' for more help.

main::(e:1): bug DB<1> @a = (1,2,3,4,5)

DB<2> splice(@a, 2,1) = ("one", "two");Can't modify splice in scalar assignment at (eval 18)[/usr/share/perl/5.8/perl5db.pl:628] line 2, at EOF

DB<3> splice(@a, 2,1, "one", "two");

DB<4> x @a 0 1 1 2 2 'one' 3 'two' 4 4 5 5

30 agosto 200763

Interactive Debugging● Perl's debugger can also do much more

– Load modules– Set breakpoints– Step through file, etc.

● It could also be connected to remotely, for example in Apache sessions

● see also Devel::ebug● ... and Devel::REPL

30 agosto 200764

Mapping the Forest

64

30 agosto 200765

Mapping the Forest● Documentation!

– not just a boring task– it's a way to make sure the piggies don't get eaten

by wolves next time

● POD is good!● but

# Watch out! We're doing this because # otherwise {...} would happen!

is good too.

30 agosto 200766

Timing (bonus)

The Forest

what's the

hurry?

66

30 agosto 200767

Timing the piggies● By hand

– my $timer = time;

– ...do stuff...– my $delta = time $timer;

30 agosto 200768

Timing the piggies (accurately)

● By hand– use Time::HiRes qw(time);

– my $timer = time;

– ...do stuff...– my $delta = time $timer;

30 agosto 200769

What to optimize● Sometimes a script/subroutine may not be the

guilty party● We should measure...

– everything?– by hand?

30 agosto 200770

What to optimize● Devel::DProf

– Show which subroutines took most time to run in your script

30 agosto 200771

test.pl#!/usr/bin/perluse strict; use warnings;

print_hello();do_do_sleep();

sub print_hello { for (1..100_000) { print "HELLO"; }}

sub do_do_sleep { for (1..2) { do_sleep(); }}sub do_sleep { sleep 1;}

30 agosto 200772

running test.pl under DProf● $ perl d:DProf test.pl

● HELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLOHELLO

30 agosto 200773

dprofpp$ dprofpp Total Elapsed Time = 2.379776 Seconds User+System Time = 0.059776 SecondsExclusive Times%Time ExclSec CumulS #Calls sec/call Csec/c Name 66.9 0.040 0.040 1 0.0400 0.0400 main::print_hello 16.7 0.010 0.010 3 0.0033 0.0033 main::BEGIN 0.00 0.000 1 strict::bits 0.00 0.000 1 strict::import 0.00 0.000 1 warnings::import 0.00 0.000 1 bytes::import 0.00 0.000 1 DynaLoader::dl_load_fi 0.00 0.000 1 DynaLoader::dl_undef_s 0.00 0.000 1 DynaLoader::dl_find_sy 0.00 0.000 1 DynaLoader::dl_install 0.00 0.000 1 warnings::BEGIN 0.00 0.000 1 Data::Dumper::bootstra

30 agosto 200774

dprofpp timings● Note that sleep wasn't counted

– real times not counted, just CPU of local machine– sleep doesn't affect local CPU– neither does remote DB call !

30 agosto 200775

dprofpp -r (real)$ dprofpp rTotal Elapsed Time = 2.379776 Seconds Real Time = 2.379776 SecondsExclusive Times%Time ExclSec CumulS #Calls sec/call Csec/c Name 84.4 2.010 2.010 2 1.0050 1.0050 main::do_sleep 14.7 0.350 0.350 1 0.3500 0.3500 main::print_hello 0.42 0.010 0.010 1 0.0100 0.0100 warnings::BEGIN 0.42 0.010 0.010 1 0.0100 0.0100 overload::BEGIN 0.00 0.000 1 strict::bits 0.00 0.000 1 strict::import 0.00 0.000 1 warnings::import 0.00 0.000 1 bytes::import 0.00 0.000 1 DynaLoader::dl_load_fi 0.00 0.000 1 DynaLoader::dl_undef_s 0.00 0.000 1 DynaLoader::dl_find_sy 0.00 0.000 1 DynaLoader::dl_install 0.00 0.000 1 warnings::BEGIN

30 agosto 200776

dprofpp -r timings● Note that do_do_sleep wasn't counted

– inclusive times not counted (child calls). Just the time a single subroutine takes to run.

– Usually, you want to know what all of your subroutine was doing!

30 agosto 200777

dprofpp -rI (real, inclusive)

$ dprofpp rITotal Elapsed Time = 2.379776 Seconds Real Time = 2.379776 SecondsInclusive Times%Time ExclSec CumulS #Calls sec/call Csec/c Name 84.4 2.010 2.010 2 1.0050 1.0050 main::do_sleep 84.4 2.010 1 2.0100 main::do_do_sleep 14.7 0.350 0.350 1 0.3500 0.3500 main::print_hello 0.84 0.020 3 0.0066 main::BEGIN 0.42 0.010 0.010 1 0.0100 0.0100 warnings::BEGIN 0.42 0.010 0.010 1 0.0100 0.0100 overload::BEGIN 0.00 0.000 1 strict::bits 0.00 0.000 1 strict::import 0.00 0.000 1 warnings::import 0.00 0.000 1 bytes::import 0.00 0.000 1 DynaLoader::dl_load_fi 0.00 0.000 1 DynaLoader::dl_undef_sy 0.00 0.000 1 DynaLoader::dl_find_sym

30 agosto 200778

DProf stability● Perl debugging hooks are... a little fragile● dprofpp is particularly so● Depending on the host/perl you may find

perl d:Whatever will crash up to ~20% of time

30 agosto 200779

DProf in web scripts● Not run from command line...● but we can add most command line flags to

shebang!● Change

– #!/usr/bin/perl– #!/usr/bin/perl d:DProf

● This works– (except the 20% of time that script will just crash)

30 agosto 200780

DProf in web scripts● By default, DProf will create a file

– $current_dir/tmon.out

● (This may be an issue if you are developing without individual sandboxes)

30 agosto 200781

DProf in web scripts (tmon.out)

– #!/usr/bin/perl – #!/usr/bin/perl d:DProf

30 agosto 200782

DProf in web scripts● There are other modules!

– Devel::SmallProf

– Devel::FastProf

– Devel::DProfLB (“Less Bad”)

– Apache::DProf (If running on mod_perl)

● Search “prof” on http://search.cpan.org– look at documentation– look at reviews!

30 agosto 200783

Profiling: what next?● You found a hotspot

– takes lots of time...– ... or called lots of times

● What next?– make the hotspot faster– find a way to rewrite it– PROFIT?

30 agosto 200784

Profiling: what next?● You found a hotspot

– takes lots of time...– ... or called lots of times

● What next?– make the hotspot faster– find a way to rewrite it

– make sure everything still works!

30 agosto 200785

testing optimization● things to test

– 1) does it still work– 2) do other things that used the same code still

work– 3) is it faster

● 1. you can do by hand● 2. it's nicer to have automated tests

30 agosto 200786

testing optimization● old version of conv

sub conv { my $id = shift;

my %hash = ( 0 => 'a', 1 => 'b', 2 => 'c', 3 => 'd', 4 => 'e', 5 => 'f', 6 => 'g', 7 => 'h', 8 => 'i', 9 => 'l' ); my @single = split //, $id; my $res; foreach (@single){ $res = $hash{$_}.$res; } return $res;}

30 agosto 200787

testing optimization● new version of conv

sub conv { my $id = shift; $id=~tr/0123456789//cd;

# delete things not in the range $id=~tr/0123456789/abcdefghil/d;

# do the transliteration return reverse $id;}

30 agosto 200788

testing optimization● The testuse Test::More tests=>1;

# test the "conv" function in sn_contacts/main.chm

is (conv( "0123" ), "dcba" );is (conv( "hello 0123 world" ), "dcba" ); # sanity checkis (conv( "890" ), "ali" );is (conv( "0123456789" ), "lihgfedcba" );

is (conv2( "0123" ), "dcba" );is (conv2( "hello 0123 world" ), "dcba" ); # sanity checkis (conv2( "890" ), "ali" );is (conv2( "0123456789" ), "lihgfedcba" );

30 agosto 200789

testing optimization (3. is it faster?)

use Benchmark qw(:all);use Test::More tests => 1;

my @LIST = '0000'..'9999';

is_deeply( [ conv1(@LIST) ], [ conv2(@LIST) ], "structures are identical",);

sub do_conv1 { return map { conv( $_ ); } @_;}sub do_conv2 { return map { conv2( $_ ); } @_;}

30 agosto 200790


my $results = timethese( 1_000_000, { conv1 => \&conv1, conv2 => \&conv2, }, "none");cmpthese( $results );

30 agosto 200791


$ perl bench.pl 1..1ok 1 structures are identical

Benchmark: timing 1000000 iterations of conv1, conv2... conv1: 1 wallclock secs ( 0.16 usr + 0.00 sys = 0.16 CPU) @ 6250000.00/s (n=1000000) (warning: too few iterations for a reliable count) conv2: 0 wallclock secs ( 0.16 usr + 0.00 sys = 0.16 CPU) @ 6250000.00/s (n=1000000) (warning: too few iterations for a reliable count)

Rate conv2 conv1conv2 6250000/s 0%conv1 6250000/s 0%

30 agosto 200792


$ perl bench.pl 1..1ok 1 structures are identical

Benchmark: timing 1000000 iterations of conv1, conv2... conv1: 1 wallclock secs ( 0.16 usr + 0.00 sys = 0.16 CPU) @ 6250000.00/s (n=1000000) (warning: too few iterations for a reliable count) conv2: 0 wallclock secs ( 0.16 usr + 0.00 sys = 0.16 CPU) @ 6250000.00/s (n=1000000) (warning: too few iterations for a reliable count)

Rate conv2 conv1conv2 6250000/s 0%conv1 6250000/s 0%

30 agosto 200793


● Benchmark compares on a micro level, running thousands (or millions) of times

● Now test on macro (single script)– Redo dprofpp, compare results

– (Or even save the old tmon.out)

Grazie

Hakim CassimallySenior Dev, dada.net

[email protected]

...

Grazie

Web Developer @ Dada GroupDada Group is seeking Perl and/or PHP Developers to join our passionate & international Production Team.Multiple locations are available in Italy, Spain, the United States, Brazil & China, for on-site work.

Your profile: you have passion for the Internet, new technologies and systems integration. Moreover you have extensive experience developing applications for the web, especially in the consumer entertainment (web & mobile) field.

If you think you are up to the challenge, please send us your resume !!!

Contact: [email protected]: www.dada.net

mailto:[email protected]

Bigbadwolf

Technology