High Performance Rails (long edition)

)JHI 1FSGPSNBODF 3BJMT Issei Naruta RubyKaigi 2013 (long edition)

*TTFJ/BSVUB VP of DevOps at COOKPAD Inc. @mirakui Image delivering
&& site performance improvement

“Speed up Rails”

COOKPAD is Japan’s top recipe site allowing visitors to upload
and search through original, user- created recipes. 20 million UU/month 1.38 million recipes *Jan 2013

$00,1"% • AWS • Rails 3.2 • Ruby 2.0.0

4J[FPG 0VS3BJMT"QQ • 1003 models • 236 controllers • 2871
view templates • 1978 lines in routes.rb • 3383 assets in manifest.yml *26 May 2013

8IBUJTUIF lSFTQPOTFUJNFz

Started GET "/articles/1" ... : Article Load (0.1ms) SELECT articles.*
FROM ... : Rendered articles/show.html.erb within layouts/application (0.7ms) Completed 200 OK in 100ms (Views: 70.1ms | ActiveRecord: 10.0ms) 3FTQPOTFUJNFJO SBJMTMPH

3FTQPOTFUJNFJO SFTQPOTFIFBEFS X-Runtime: 0.123456

routing before_ﬁlter after_ﬁlter rendering action Rails Rack "Completed" time X-Runtime
)5513FRVFTU )5513FTQPOTF

*T3BJMT4MPX

http://www.techempower.com/benchmarks/ Rails (slow) Servlet (fast) 8FC'SBNFXPSL#FODINBSLT

.BKPSQSFNJTFPG IJHIQFSGPSNBODF 3BJMTBQQT

%PO`UQSPDFTT SVCZDPEF JGQPTTJCMF

/HJOY 6OJDPSO QSPYZ CVGGFSJOH 3BJMT

/HJOY 6OJDPSO serve static files

/HJOY 6OJDPSO 7BSOJTI "QBDIF TUBUJDpMFT QBHFDBDIJOH QSPYZ CVGGFSJOH 3BJMT

#SPXTFSTJEF $BDIJOH

6TJOHCSPXTFSTJEFDBDIJOH

/FWFSVTFCSPXTFSTJEFDBDIJOH

$POEJUJPOBM(&5

1st request 2nd request

GSFTI@XIFO • Set ETag and Last-Modified to response header •
Return “304 Not Modified” if; •ETag == If-None-Match • Last-Modified <= If-Modified-Since • Skip rendering if 304 (fast!)

is equal to

FUBH\^ 3BJMT ETag is made of <!BSUJDMF DVSSFOU@VTFSJE>

3VCZ7FSTJPO

0VSSVCZIJTUPSZ • Sep 2012 • REE (Ruby Enterprise Edition) 1.8.7
• Feb 2013 • MRI 1.9.3 • Apr 2013 • MRI 2.0.0

8IBU`T3&& • MRI 1.8.7 + “MBARI patch” • “100% compatible
with MRI 1.8.7” • Copy-on-Write Friendly GC • tcmalloc

REE 1.8.7 MRI 1.9.3 MRI 2.0.0 3FTQPOTFUJNF CZSVCZWFSTJPOT -100ms -20ms

#FODINBSLT JOUIJTQSFTFOUBUJPOBSF 3VCZ 3BJMTSD

1FSGPSNBODF #PUUMFOFDLT JO3BJMT

0% 20% 40% 60% 80% 100% SQL Ruby 3FTQPOTFUJNF CSFBLEPXO
SFDJQFTIPX DPPLQBEDPN

4MPXTUVGG *FYQFSJFODFE • ActiveRecord objects creation • Resolving routes

"DUJWF3FDPSE`T PWFSIFBE "SUJDMFMBTU UP@B "SUJDMF-PBE NT 4&-&$5ABSUJDMFTA '30.

"DUJWF3FDPSE`T PWFSIFBE #FODINBSLNT\"SUJDMFMBTU UP@B^ "SUJDMF-PBE NT 4&-&$5ABSUJDMFTA '30.
The query is fast enough but creating 100 AR objects is slow

3FTPMWJOHSPVUFT • Fat routes slow down url_for • e.g. link_to
helper

4J[FPG 0VS3BJMT"QQ • 1003 models • 236 controllers • 2871
view templates • 1978 lines in routes.rb • 3383 assets in manifest.yml *26 May 2013

MJOL@UPIFMMP IFMMP@JOEFY@VSM MJOL@UPbIFMMP` DPOUSPMMFSbIFMMP` BDUJPObJOEFY` 8IJDIJTGBTUFS

DPOpHSPVUFTSC HFUbIFMMPJOEFY` UPbIFMMPJOEFY` UJNFTEPcJc HFUlIFMMP\J^z UPbIFMMPJOEFY` FOE .BLFGBUSPVUFT

MJOL@UPIFMMP IFMMP@JOEFY@VSM NT MJOL@UPbIFMMP` DPOUSPMMFSbIFMMP` BDUJPObJOEFY` NT 3FTVMU

4MPXSBDL NJEEMFXBSF w3BDL$BDIF • Caches response with Rails.cache • Default
in Rails 3.2 • Not default in Rails 4

$BDIJOH

'SBHNFOU$BDIF

DBDIFEP USFOE@LFZXPSET FOE

DBDIFEP controller action cache digest (Rails4) views/localhost:3000/hello/cache_test/3e9258928c27f4ffc5135520c3a976c5 %FGBVMUDBDIFLFZ

DBDIFGSBHNFOUbLFZ`EP views/localhost:3000/hello/cache_test?fragment=key1 /3e9258928c27f4ffc5135520c3a976c5 fragment key 'SBHNFOUDBDIFLFZ

DBDIFbLFZ`EP views/key1/116839051556390fb4d5b25362cfe6eb global key (MPCBMDBDIFLFZ

DBDIF!BSUJDMFEP table name @article.cache_key (id, updated_at) views/articles/1-20130527190532189241000 /116839051556390fb4d5b25362cfe6eb "3DBDIFLFZ

$BDIF%JHFTUT BLBl3VTTJBOEPMMDBDIJOHz

projects documents comments 1 * 1 * has_many :documents has_many
:comments belongs_to :document id updated_at id project_id updated_at id document_id updated_at belongs_to :documents

projects/show.html.erb _document.html.erb _comment.html.erb DBDIFQSPKFDUEP DBDIFEPDVNFOUEP DBDIFDPNNFOUEP

comment.update!

projects/show.html.erb _document.html.erb _comment.html.erb DBDIFQSPKFDUEP DBDIFEPDVNFOUEP DBDIFDPNNFOUEP still old... 3BJMT ʜʜ
LFZOPUDIBOHFE ʜʜ

projects documents comments 1 * 1 * has_many :documents has_many
:comments belongs_to :document, touch: true id updated_at id project_id updated_at id document_id updated_at belongs_to :documents, touch: true update

projects/show.html.erb _document.html.erb _comment.html.erb DBDIFQSPKFDUEP DBDIFEPDVNFOUEP DBDIFDPNNFOUEP updated 3BJMT8BZ LFZDIBOHFE LFZDIFOHFE
LFZDIBOHFE

projects/show.html.erb _document.html.erb _comment.html.erb DBDIFQSPKFDUEP DBDIFEPDVNFOUEP DBDIFDPNNFOUEP updated 3BJMT8BZ LFZDIBOHFE .PEJGZpMF
LFZDIBOHFE

made of [id, updated_at] MD5 of the template file itself
and all of its dependencies views/documents/1-20130527190532189241000 /116839051556390fb4d5b25362cfe6eb

"DUJPO$BDIF

caches_action :show, if: -> { !current_user.staff? }, cache_path: -> {
custom_cache_path }, expires_in: 1.hour

expires_in Sweeper expires_in expires_in expires_in Sweeper Sweeper Sweeper Sweeper Sweeper
Sweeper expires_in expires_in expires_in Sweeper

Action Caching is not Rails 4 Way Use Russian-doll Caching

action cache page cache sweeper actionpack-action_caching actionpack-page_caching rails-observers &YUSBDUFEBTHFNT GPS3BJMT
3BJMT

5FNQMBUF &OHJOFT

FSC IBNM TMJN )PX3BJMTSFOEFS UFNQMBUFT

FSC IBNM TMJN 3VCZ )PX3BJMTSFOEFS UFNQMBUFT "DUJPO7JFX5FNQMBUFDPNQJMF BUpSTUUJNFPG SFOEFSJOHFBDIUFNQMBUFT

&3# FSVCJT

SFOEFSbFSC@UFTU` @BQQ@WJFXT@IFMMP@FSC@UFTU@IUNM@FSC@@@ Calls

Compiling time is not critical matter for Rails apps, but
rendering time is.

#FODINBSL UJNFTEP SFOEFS@UP@TUSJOH FOE ERB (erubis) Haml Slim

3FOEFSJOH5JNF ERB Haml Slim 1,032.49 ms 1,866.02 ms 1,689.26 ms

6OJDPSOXJUI ($EJTBCMF

)5513FRVFTUT time

6OJDPSO%FGBVMU GC

($EJTBCMF GC.start GC.disable GC.disable

# unicorn config after_fork do |server, worker| GC.disable end #
config.ru require “unicorn/oob_gc” use Unicorn::OobGC, 10 Run GC each 10 requests

# config.ru # Unicorn self-process killer require 'unicorn/worker_killer' # Max
requests per worker use Unicorn::WorkerKiller::MaxRequests, 3072, 4096 # Max memory size (RSS) per worker use Unicorn::WorkerKiller::Oom, (192*(1024**2)), (256*(1024**2)) VOJDPSOXPSLFSLJMMFS IUUQTHJUIVCDPNL[LVOJDPSOXPSLFSLJMMFS

1SPpMJOH

SVCZQSPG

SVCZQSPG • Call-tree is too complex to profile Rails apps

"DUJWF4VQQPSU#FODINBSLBCMF 3BJMTPSMBUFS <% benchmark "Process data files" do %> <%=
expensive_files_operation %> <% end %> # production.log Process data files (123.45ms)

'JHIUJOHBHBJOTU TMPXRVFSJFT

"SQSPYZ IUUQTHJUIVCDPNDPPLQBEBSQSPYZ "DUJWF3FDPSE %#"EBQUFS SELECT * FROM ... Query

"SQSPYZ IUUQTHJUIVCDPNDPPLQBEBSQSPYZ "DUJWF3FDPSE %#"EBQUFS 1SPYZ 1SPYZ logging rewrite query

"SQSPYZFYBNQMF

"SQSPYZFYBNQMF "DUJWF3FDPSE %#"EBQUFS SELECT * FROM ... $PNNFOU"EEFS SELECT *
FROM ... /*this_is_comment*/

"DUJWF3FDPSE %#"EBQUFS "SQSPYZ qVFOUE .POHP%# -PHHJOHTMPXRVFSJFT VTJOH"SQSPYZ

$PODMVTJPO

5SBEFPGG JOQFSGPSNBODFUVOJOH

%FWFMPQFST 6TFST 4FSWFST

8F`SFIJSJOH http://info.cookpad.com/jobs/ [en, ja]

5JNFUPIBDL HVZT http://info.cookpad.com/24contest4 http://info.cookpad.com/24contest4_en 24-hour hack-a-thon June 15 - 16
[en] [ja]

5IBOLZPVGPS MJTUFOJOH

High Performance Rails (long edition)

High Performance Rails (long edition)

More Decks by Issei Naruta

Other Decks in Programming

Featured

Transcript