Top to Bottom PHP Profiling

PHP applications are becoming more complex than ever before. We've moved away from the simple LAMP stack to having many different layers and pieces that our application interacts with. There are many places where things can slow down or break. We'll discuss the different parts of a PHP application including the browser, php application, database, and the server. We'll show different tools for each part and how to use them. By the end of this presentation you should have a whole arsenal of tools to help you profile & understand your application much better.

Transcript

1. Profiling PHP
   Applications Justin Carmony - Lone Star PHP `14 @JustinCarmony
2. About Me •Director of Development
 @ Deseret Digital Media •President of the Utah PHP Usergroup •Make (and Break) Web Stuff ~10 years
3. Experience •Lots of work with “Middle-Scale” websites. •Lucky to Work on Fun Projects •Lots of Web Service Stuff •Learned most of this from other really smart people.
4. Lets Start With a Story!
5. You Work for an Awesome Tech Company
6. Team Is Working Hard to Build New Things!
7. You launch your awesome product!
8. A Few More Features…
9. … and next thing you know…
11. Awesome Job Team, We Rock!
12. We Need ! Real-Time XYZ Feature! ASAP! &#$%!
14. “I fixed it by turning
 off caching…” - Dev @ 80th Hour This Week
15. “I’m sure this will work…”
19. “Our servers are melting!”
26. Let’s Learn How to Profile
27. We’ll Use An Example and an awesome one at that...
29. We Have Similar Jobs Awesome, huh?
30. MythBusters • They get a Myth • Break it down to its different parts • They Conduct a Bunch of Tests • They Draw Conclusions • and have a Fun Time
32. Developers • We get Myths
 (Hey, this is slow) • We Break it down to the different pieces • We Conduct a Bunch of Tests. • We Draw Conclusions
 (and normally make changes) • and Have a Fun Time...
34. So Lets Get Started! php /lets/get/started.php now
35. First Off, Some
    PHP Myths and there are a lot of them!
36. Common Performance 
 Claims (Myths) • echo instead of print() • echo commas, not periods • don’t use custom functions/ classes • pass by reference instead of by value • single quotes, not double quotes • include instead of include_once • === faster than == • for faster than foreach Busted
37. These “Performance Tips” 
 Rarely Make a Difference
38. Thats Like Debating the Windshield Wipers’s Effect on this Car’s Performance
39. Immaterial
40. Intuition Based 
 Optimizations i.e. how to not effectively improve performance Phrase Coined by Dave Smith http://thesmithfam.org/blog/2011/02/
41. WRONG 90% of the time Your Intuitions are
42. Example: Fuel Efficiency Myths
43. MythBusters Tested • Keeping Your AC Off vs Windows Down • Idling Better than Stop/Start • Magical Aerodynamics • Dirt-Free Filters • Special Fuel Additives to Slow Burn
44. BUSTED They Didn’t Make a Difference, Or Worsened Fuel Efficiency All Were
45. After Testing, Tory & Grant Used 33% More Fuel While Driving Angry Driving Angry? Tory & Grant used 33% more fuel while driving angry.
46. Effective Profiling
    is based on
    Actual Results
47. Two “Profiling” Modes
48. Normal Profiling Lets Try this Configuration Change and Measure the Performance Difference
49. Emergency Profiling Aaaaaaaaaaaaah the Website is Down!!!! Why is it so slow??? Fix it!!!!!!
50. Same Techniques
    Apply to Both • Some you’ll want to chose first depending on your situation. • You’ll want to be careful when profiling in Production, you can make things worse.
51. The Full Picture
52. Browser Web Server Static Files PHP App Server OS Hardware Web Services & Resources Cache Database Data Store
53. Don’t Panic!
54. Goal: Find Bottlenecks
55. Tools!
56. Browser Web Server Static Files PHP App Server OS Hardware Web Services & Cache Database Data Store
57. Browser Dev Tools
58. Google Page Speed Web Site & Browser Plugin
61. HTTP Debugging Lower Level Fun Stuff
62. WireShark Windows / Mac / Linux
64. Charles
 Web Debugging Proxy Windows / Mac / Linux
66. Other Tools • YSlow
 http://developer.yahoo.com/yslow/ • Pingdom Tools
 http://tools.pingdom.com/ • Windows HTTP Debugger: Fiddler2
67. Browser Web Server Static Files PHP App Server OS Hardware Web Services & Cache Database Data Store
68. apachetop apachetop -f /var/log/apache/access.log
70. Apachetop works
    with nginx
71. Apache’s mod_status <Location /server-status> SetHandler server-status ! Order Deny,Allow Deny from all Allow from 25.131.42.122 </Location> ! # ExtendedStatus On
73. siege
74. siege • Command Line Tool • Run Concurrent HTTP Requests • Runs on Linux & Mac OS • Windows Users: Run a VM • Great Way to Test End Result
75. • Create txt file with lists of URLs to hit • Run Command:
 siege -c 10 -r 10 -f urls.txt • c = concurrent
 r = # of requests
 f = path to URL file siege
78. Other Tools • nginx stub status (similar to mod_status)
 http://wiki.nginx.org/HttpStubStatusModule • Apache Bench & http_load -- CLI • DDM crawlr
 https://github.com/deseretdigital/crawlr • JMeter -- Java Desktop App
79. Browser Web Server Static Files PHP App Server OS Hardware Web Services & Resources Cache Database Data Store
80. XHProf / XHGui My Favorite PHP Profiler
82. XHProf • Developed by Facebook • Works well for both Development 
 & Production* Profiling • pecl extension • Decent UI for viewing Results http://mirror.facebook.net/facebook/xhprof/doc.html * - Use Sampling & Special CPU Flags for Production
83. XHGui 1.0 • Better GUI • Easy to Setup • Built In Sampling • Advanced Configuration • MySQL Backend https://github.com/preinheimer/xhprof
84. XHGui 2.0 • Rewrite of XHGUI 1.0 • MongoDB Backend • Better UI • Use this unless Mongo won’t work in your setup. • I recommend using this! https://github.com/perftools/xhgui
85. Demo Time
86. Xdebug Profiler • Install Xdebug • Enable Xdebug Profiling • Outputs a Cachegrind Log • Use KCachegrind / WinCachegrind / Webgrind to view • MacCallGrind for $40
87. Enabling Xdebug Profiling xdebug.profiler_enable=1 
 xdebug.profiler_output_dir=/tmp 
 xdebug.profiler_output_name=cachegrind.out.%p Xdebug Profiling Not for Production (unless you have a 100TB HDD laying around)
88. Webgrind
89. Timing Points • Most Frameworks have Built-In Profiling / Timing Points • Most ORMs have them as well • You can do them yourself • A must for Database Queries
90. Timing Points $start = microtime(true); // true to get as float ! /* Do your Crazy Code, i.e. query */ ! $end = microtime(true); ! $time = round($end - $start, 4);
91. Other Tools • Inclued
 http://php.net/manual/en/book.inclued.php • Memtrack
 http://php.net/manual/en/book.memtrack.php
92. Browser Web Server Static Files PHP App Server OS Hardware Web Services & Cache Database Data Store
93. Databases • Typically, First Thing to Slow Down • Things to that will Kill the DB: • Missing Indexes • Nested Queries • Large Joins • Locked Queries
94. Jet Profiler • MySQL Profiler • Free Version (Okay) & Paid (Awesome) • Not Cheap ($399) • But Worth It • Analytics Over Time
95. Jet Profiler
96. MySQL Commands • explain <query> • show processlist • show variables • show status
97. mtop
98. Browser Web Server Static Files PHP App Server OS Hardware Web Services & Cache Database Data Store
99. vmstat command: vmstat 1
101. vmstat • swap > 0 means swapping
 Memory Issue • cpu sys + us really high
 CPU / Code / PHP Problem • cpu wa > 10
 Disk IO Problem* * - Technically could be Network IO as well, but typically one of the last and more rare bottlenecks to hit
102. vnstat command: vnstat -l -i eth0
103. top
104. htop
105. strace
106. iotop
107. Other Tools • netstat • iostat • mpstat • pidstat • (on Debian, install via sysstat package)
108. Other Tools • grep, awk, sed • IPs Connected:
 netstat -plan | grep :80 | awk '{print $5}' | sed 's/::ffff://g' | awk -F: '{print $1}' | sort | uniq -c | sort -n
109. Monitoring
110. Graphite & StatsD
111. Munin http://munin-monitoring.org/
112. Wormly http://wormly.com/
113. Whew, Lots of Tools and a lot more out there not in this talk Find any cool ones, let me know!
114. Normal Profiling • Start with XHProf/XHGui and FireBug • Avoid Premature Optimization • Complicated Change • Little Reward • Use siege, or alternative, to simulate load.
115. Emergency Profiling • Start with OS Level Testing Tools (htop, vmstat, vnstat) to check Server Performance • Determine which Resource(s) are being over utilized • Finding the bottleneck is key
116. Emergency Profiling • What Changed? • Increased Traffic? • New Feature? • Something Failed/Down? • Don’t Panic & Start Wildly Guessing
117. Few Final Thoughts
118. The Better You
     Understand
     The Problem
     The Better You Can
     Fix It
119. Don’t
     Put Off Profiling
     Until there is an
     Emergency
120. “Throwing Hardware” at it typically Masks the Problem
121. Ask Others for Ideas
     ! aka
     Brainstorming
122. Good Luck!
123. Checkout Dustin Whittle’s Talk https://speakerdeck.com/dustinwhittle/performance-testing-crash-course-1
124. Questions?
125. Thanks! Twitter: JustinCarmony ! IRC: carmony #uphpu #salt #phpmentoring ! Website: http://www.justincarmony.com/blog ! Email: justin@justincarmony.com