Ein Stall voller Trüffelschweine - (PHP-)Profiling-Tools im Überblick

EIN STALL VOLLER
TRÜFFELSCHWEINE

(PHP)-Profiling-Tools im Überblick

René Bruns - TNG Big Techday

ÜBER MICH

René Bruns

 Entwickler seit 2001

 seit 2009 bei Travian Games

 seit 2011 Technical Director

2

UND WER SEID IHR?

3

UND WER SEID IHR?

Entwickler?

3

UND WER SEID IHR?

Entwickler?

Systemadministratoren?

3

UND WER SEID IHR?

Entwickler?


QA?

3

UND WER SEID IHR?

Entwickler?


C*O? QA?

3

UND WER SEID IHR?

Entwickler? Projektmanager?


C*O? QA?

3

PROFILING?

Profiler

Es gibt unterschiedliche Problembereiche in der
Softwareentwicklung, die durch ineffiziente Programmierung
ausgelöst werden. Ein Profiler hilft dem Entwickler durch
Analyse und Vergleich von laufenden Programmen die
Problembereiche aufzudecken.
http://de.wikipedia.org/wiki/Profiler_(Programmierung)

5

AUSGANGSSITUATION

SLOW!!11!

6

PROFILING FRÜHER(?)

class Main
{
public static function run() {
$otherClass = new OtherClass();
$otherClass->doNothingImportant();
$dummy = new SlowClass();
$dummy->slowFunction(10);
}
}

Main::run();

7


class Main
{
$start = microtime(true);
printf ("Before OC: %f" . PHP_EOL, microtime(true) - $start);
printf ("After OC: %f" . PHP_EOL, microtime(true) - $start);
printf ("Before SC: %f" . PHP_EOL, microtime(true) - $start);
printf ("After SC: %f" . PHP_EOL, microtime(true) - $start);
}
}

Main::run();

8


class Main
{
$start = microtime(true);
printf ("Before OC: %f" . PHP_EOL, microtime(true) - $start);
printf ("After OC: %f" . PHP_EOL, microtime(true) - $start);
printf ("Before SC: %f" . PHP_EOL, microtime(true) - $start);
printf ("After SC: %f" . PHP_EOL, microtime(true) - $start);
}
} $ php Main.php
Before OC: 0.000625
Main::run();
After OC: 0.004919
Before SC: 0.005099
..........
After SC: 10.013932
8

PROFILING HEUTE

 Viele verschiedene Fehlerquellen
 Browser
 Frontend
 Backend
 Datenbank
 Services

 Schwachstelle suchen

 Vom Groben in‘s Feine

9

SCHRITT 1: MELDUNG ÜBERPRÜFEN

He‘s
dead,
Jim!
10

SCHRITT 2: URSACHE EINKREISEN

Browser

11


Browser

Frontend

11


Browser

Frontend

Backend

11


Browser

Frontend

Backend

DB

11

BROWSER

 Langsamer JS-Code

 Zu hoher Speicherverbrauch

 Zu viele CSS-Selektoren

 Langsame Ladezeiten
 Zu viele parallele Downloads
 Offen gehaltene Verbindungen
 Netzwerk
 Internet

 Antwort vom Server zu langsam -> Backend

12

FRONTEND / BACKEND

 Unnötige Schreib- / Lesezugriffe in Schleifen

 Log-Ausgaben

 Schleifen sind unnötig geschachtelt

 Unnötig langsame Befehle benutzt
 RegExp statt ==

 Externer Service reagiert langsam

 Datenbankabfrage dauert zu lange --> DB

20

XDEBUG

 PHP Extension by Derick Rethans

 Kann per INI-Direktive aktiviert werden

 breites Einsatzspektrum
 Debugger
 Profiler
 Tracer
 StackTrace-Formatter
 ...

 hoher Memory-/ CPU-/ IO-Footprint

 Niemals(!) dauerhaft im Produktivbetrieb verwenden

21

XDEBUG

$ php -d xdebug.profiler_enable=1 -d xdebug.profiler_output_dir=. Main.php
..........

22

XDEBUG

..........

$ ls -l ca*
-rwxrwxrwx 1 rene staff 4461 11 Jun 13:53 cachegrind.out.69408

22

XDEBUG

..........

$ ls -l ca*
-rwxrwxrwx 1 rene staff 4461 11 Jun 13:53 cachegrind.out.69408

$ cat cachegrind.out.69408
version: 1
creator: xdebug 2.2.0
cmd: /Volumes/TCDATA/Sourcen/git/ProfilingTalk/Main.php
part: 1
positions: line

events: Time

fl=/Volumes/TCDATA/Sourcen/git/ProfilingTalk/OtherClass.php
fn=require_once::/Volumes/TCDATA/Sourcen/git/ProfilingTalk/OtherClass.php
1 42

fl=/Volumes/TCDATA/Sourcen/git/ProfilingTalk/SlowClass.php
fn=require_once::/Volumes/TCDATA/Sourcen/git/ProfilingTalk/SlowClass.php
10

22

XDEBUG
 Cachegrind-Files können via KCachegrind, WinCachegrind, WebGrind oder PhpStorm-IDE
ausgewertet werden.

 Bsp. PhpStorm:

23

XDEBUG
 Cachegrind-Files können via KCachegrind, WinCachegrind, WebGrind oder PhpStorm-IDE
ausgewertet werden.

 Bsp. PhpStorm:

24

XHPROF

 PHP Extension by Facebook

 Kann codeseitig aktiviert werden

 Profiler only

 sehr schlanker Footprint, daher auch im Livebetrieb
einsetzbar (Monte Carlo-Algorithmus)

26

XHPROF

header.php:

include_once '/usr/lib/php/pear/xhprof_lib/utils/xhprof_lib.php';
include_once '/usr/lib/php/pear/xhprof_lib/utils/xhprof_runs.php';
xhprof_enable(XHPROF_FLAGS_CPU + XHPROF_FLAGS_MEMORY);

footer.php:
include_once '/usr/lib/php/pear/xhprof_lib/utils/xhprof_lib.php';
include_once '/usr/lib/php/pear/xhprof_lib/utils/xhprof_runs.php';

$xhprof_data = xhprof_disable();
$xhprof_runs = new XHProfRuns_Default();
$run_id = $xhprof_runs->save_run($xhprof_data, 'testrene');
echo "http://localhost/~rene/xhprof_html/index.php?run={$run_id}&source=testrenen";

27

XHPROF
$ php -d auto_append_fileo_prepend_file="header.php" -d auto_append_file="footer.php" Main.php
..........
http://localhost/~rene/xhprof_html/index.php?run=4fd61a7e92aeb&source=testrene

28

XHPROF
$ php -d auto_append_fileo_prepend_file="header.php" -d auto_append_file="footer.php" Main.php
..........
http://localhost/~rene/xhprof_html/index.php?run=4fd61a7e92aeb&source=testrene

$ cat 4fd61a7e92aeb.testrene
a:14:{s:23:"main()==>load::Main.php";a:5:{s:2:"ct";i:1;s:2:"wt";i:459;s:3:"cpu";i:1220;s:
2:"mu";i:10408;s:3:"pmu";i:0;}s:43:"main()==>load::ProfilingTalk/OtherClass.php";a:5:{s:
2:"ct";i:1;s:2:"wt";i:798;s:3:"cpu";i:113;s:2:"mu";i:6880;s:3:"pmu";i:0;}s:
47:"main()==>run_init::ProfilingTalk/OtherClass.php";a:5:{s:2:"ct";i:1;s:2:"wt";i:14;s:3:"cpu";i:
10;s:2:"mu";i:1104;s:3:"pmu";i:0;}s:42:"main()==>load::ProfilingTalk/SlowClass.php";a:5:{s:
2:"ct";i:1;s:2:"wt";i:467;s:3:"cpu";i:87;s:2:"mu";i:9648;s:3:"pmu";i:0;}s:
46:"main()==>run_init::ProfilingTalk/SlowClass.php";a:5:{s:2:"ct";i:1;s:2:"wt";i:9;s:3:"cpu";i:
7;s:2:"mu";i:1104;s:3:"pmu";i:0;}s:36:"OtherClass::doNothingImportant==>pow";a:5:{s:
2:"ct";i:1;s:2:"wt";i:1655;s:3:"cpu";i:926;s:2:"mu";i:1176;s:3:"pmu";i:0;}[...]}

28

DATENBANK

 Unnötige Daten geladen

 Kein/falscher Index benutzt

 Daten werden nach der Abfrage noch sortiert

 Langsame IOs

 Falsche Systemeinstellungen (Buffer Sizes, ...)

31

DB - EXPLAIN

select LastName, FirstName, sum(UnitPrice * Quantity) from employees e inner join
orders o on e.EmployeeID = o.EmployeeID inner join "Order Details" od on o.OrderID
= od.OrderID group by LastName, FirstName;

32

DB - EXPLAIN


sqlite> explain query plan [...];

0|0|0|SCAN TABLE employees AS e (~1000000 rows)
0|1|1|SEARCH TABLE orders AS o USING AUTOMATIC COVERING INDEX (EmployeeID=?) (~7 rows)
0|2|2|SEARCH TABLE Order Details AS od USING INDEX sqlite_autoindex_Order Details_1
(OrderID=?) (~10 rows)
0|0|0|USE TEMP B-TREE FOR GROUP BY

32

DB - EXPLAIN


sqlite> explain query plan [...];

0|0|0|SCAN TABLE employees AS e (~1000000 rows)
0|1|1|SEARCH TABLE orders AS o USING AUTOMATIC COVERING INDEX (EmployeeID=?) (~7 rows)
0|2|2|SEARCH TABLE Order Details AS od USING INDEX sqlite_autoindex_Order Details_1
(OrderID=?) (~10 rows)
0|0|0|USE TEMP B-TREE FOR GROUP BY

sqlite> explain [...];
0|Trace|0|0|0||00|
1|SorterOpen|3|5|0|keyinfo(2,BINARY,BINARY)|00|
2|Integer|0|7|0||00|
3|Integer|0|6|0||00|
4|Null|0|10|11||00|
5|Gosub|9|74|0||00|
6|Goto|0|81|0||00|
7|OpenRead|0|120|0|3|00|

32

SCHRITT 3: ERFOLG PRÜFEN

 Hat die Optimierung etwas gebracht?

 Vorher-Nachher-Check

 Lasttests

 Tools
 Apache Benchmark (AB)
 JMeter

33

APACHE BENCHMARK (AB)
$ ab -n 100 -c 10 http://localhost/~rene/talk/Main.php
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Copyright 1996 Adam Twiss, Zeus Technology Ltd, http://www.zeustech.net/
Licensed to The Apache Software Foundation, http://www.apache.org/

Benchmarking localhost (be patient).....done

Server Software: Apache/2.2.21
Server Hostname: localhost
Server Port: 80

Document Path: /~rene/talk/Main.php
Document Length: 11 bytes

Concurrency Level: 10
Time taken for tests: 56.692 seconds
Complete requests: 100
Failed requests: 0
Write errors: 0
Total transferred: 22300 bytes
HTML transferred: 1100 bytes
Requests per second: 1.76 [#/sec] (mean)
Time per request: 5669.198 [ms] (mean)
Time per request: 566.920 [ms] (mean, across all concurrent requests)
Transfer rate: 0.38 [Kbytes/sec] received

Connection Times (ms)
min mean[+/-sd] median max
Connect: 0 3660 2329.9 5022 6356
Processing: 0 1758 3060.0 0 11378
Waiting: 0 1757 3059.9 0 11377
Total: 5010 5417 1403.8 5030 11378

Percentage of the requests served within a certain time (ms)
50% 5030
66% 5040
75% 5052
80% 5063
90% 6316
95% 11378 vorher
98% 11378
99% 11378
100% 11378 (longest request)

34

APACHE BENCHMARK (AB)
This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Benchmarking localhost (be patient).....done This is ApacheBench, Version 2.3 <$Revision: 655654 $>
Server Software: Apache/2.2.21
Server Hostname: localhost Benchmarking localhost (be patient).....done
Server Port: 80

Document Path: /~rene/talk/Main.php Server Software: Apache/2.2.21
Document Length: 11 bytes Server Hostname: localhost
Server Port: 80
Concurrency Level: 10
Time taken for tests: 56.692 seconds Document Path: /~rene/talk/Main.php
Complete requests: 100 Document Length: 11 bytes
Failed requests: 0
Write errors: 0 Concurrency Level: 10
Total transferred: 22300 bytes Time taken for tests: 5.395 seconds
HTML transferred: 1100 bytes Complete requests: 100
Requests per second: 1.76 [#/sec] (mean) Failed requests: 0
Time per request: 5669.198 [ms] (mean) Write errors: 0
Time per request: 566.920 [ms] (mean, across transferred:
Total all concurrent requests)
22300 bytes
Transfer rate: 0.38 [Kbytes/sec] received
HTML transferred: 1100 bytes
Requests per second: 18.54 [#/sec] (mean)
Connection Times (ms) Time per request: 539.458 [ms] (mean)
min mean[+/-sd] median max Time per request: 53.946 [ms] (mean, across all concurrent requests)
Connect: 0 3660 2329.9 5022 6356 Transfer rate: 4.04 [Kbytes/sec] received
Processing: 0 1758 3060.0 0 11378
Waiting: 0 1757 3059.9 0 11377 Connection Times (ms)
Total: 5010 5417 1403.8 5030 11378 min mean[+/-sd] median max
Connect: 0 395 217.3 531 577
Percentage of the requests served within a certain time (ms) 0 142 214.4
Processing: 1 559
50% 5030 Waiting: 0 142 214.6 0 559
66% 5040 Total: 512 538 14.9 535 577
75% 5052
80% 5063
vorher
Percentage of the requests served within a certain time (ms)
90% 6316 50% 535
95% 11378 66% 541
98% 11378 75% 553

nachher
99% 11378 80% 555
100% 11378 (longest request) 90% 559
95% 559
98% 566
99% 577
100% 577 (longest request)

34

APACHE JMETER

 verteilte Lasttests

 Java-Programm

 Client-Server-Architektur

 Testen von
 HTTP/S
 SOAP
 DB‘s via JDBC
 SMTP, POP3, IMAP
 ...

35

JMETER - TESTPLAN ERSTELLEN

 Manuell oder per HTTP Proxy Server

36

JMETER - TESTPLAN ERSTELLEN

 Manuell oder per HTTP Proxy Server

37

JMETER - USER EINLOGGEN

38

JMETER - USER EINLOGGEN

39

JMETER - PARALLELE ZUGRIFFE

40

JMETER - PARALLELE ZUGRIFFE

41

JMETER - BERICHTE

42

LINKLISTE

 Google Chrome: https://www.google.com/chrome
 Chrome DevTools: https://developers.google.com/chrome-
developer-tools/?hl=de

 Xdebug: http://xdebug.org/
 PHPStorm: http://www.jetbrains.com/phpstorm/
 KCachegrind: http://kcachegrind.sourceforge.net
 WinCacheGrind: http://sourceforge.net/projects/wincachegrind/
 Webgrind: http://code.google.com/p/webgrind/

 XHProf: http://php.net/xhprof

 AB: http://httpd.apache.org/docs/2.0/programs/ab.html

 Jmeter: http://jmeter.apache.org/

43

FRAGEN?
GERNE AUCH SPÄTER:
MAIL: R.BRUNS@TRAVIANGAMES.COM
TWITTER: @RENEBRUNS

Ein Stall voller Trüffelschweine - (PHP-)Profiling-Tools im Überblick

More Related Content

What's hot (20)

Viewers also liked (18)

Similar to Ein Stall voller Trüffelschweine - (PHP-)Profiling-Tools im Überblick (20)

Recently uploaded (20)

Ein Stall voller Trüffelschweine - (PHP-)Profiling-Tools im Überblick

Editor's Notes