Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Git infrastructure with Go
Search
David Calavera
February 21, 2015
Programming
3
570
Git infrastructure with Go
Talk for Gophercon India 2015 about using Go to manage large git data system.
David Calavera
February 21, 2015
Tweet
Share
More Decks by David Calavera
See All by David Calavera
The language of Chatops
calavera
0
140
How GitHub develops, ships and supports GitHub Enterprise
calavera
2
460
Project Warbird
calavera
1
160
Trinidad
calavera
0
32
JRuby hacking guide
calavera
0
57
My name is Trinidad
calavera
0
54
Other Decks in Programming
See All in Programming
250830 IaCの選定~AWS SAMのLambdaをECSに乗り換えたときの備忘録~
east_takumi
0
390
Testing Trophyは叫ばない
toms74209200
0
880
Rancher と Terraform
fufuhu
2
550
Laravel Boost 超入門
fire_arlo
3
220
為你自己學 Python - 冷知識篇
eddie
1
350
私の後悔をAWS DMSで解決した話
hiramax
4
210
RDoc meets YARD
okuramasafumi
4
170
testingを眺める
matumoto
1
140
Namespace and Its Future
tagomoris
6
700
「待たせ上手」なスケルトンスクリーン、 そのUXの裏側
teamlab
PRO
0
530
FindyにおけるTakumi活用と脆弱性管理のこれから
rvirus0817
0
520
Android 16 × Jetpack Composeで縦書きテキストエディタを作ろう / Vertical Text Editor with Compose on Android 16
cc4966
2
230
Featured
See All Featured
Unsuck your backbone
ammeep
671
58k
Being A Developer After 40
akosma
90
590k
Large-scale JavaScript Application Architecture
addyosmani
512
110k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
194
16k
Measuring & Analyzing Core Web Vitals
bluesmoon
9
580
Building a Scalable Design System with Sketch
lauravandoore
462
33k
Connecting the Dots Between Site Speed, User Experience & Your Business [WebExpo 2025]
tammyeverts
8
530
Keith and Marios Guide to Fast Websites
keithpitt
411
22k
Why You Should Never Use an ORM
jnunemaker
PRO
59
9.5k
Bash Introduction
62gerente
615
210k
Art, The Web, and Tiny UX
lynnandtonic
303
21k
ReactJS: Keep Simple. Everything can be a component!
pedronauck
667
120k
Transcript
High performance git infrastructure with Go David Calavera Code Climate
None
2005
(╯°□°)╯︵ ┻━┻
2008
2010…
How do we make git faster?
* system(“git log …”) * Ruby for brevity
git fetch origin
None
github.com/libgit2/git2go Disclaimer: examples do not handle errors
// Load with side effects. // Initialize libgit2’s TLS: //
// func init() { // C.git_libgit2_init() // } // // Import package “git”, which is // not very goimports friendly. import “github.com/libgit2/git2go”
// Create a new repository. // Do not use a
working directory. path := “/var/git/repository” bare := true r, _ := git.InitRepository(path, bare) ! // Clone a repository. url := “git://github.com/golang/go” opts := git.CloneOptions{Bare: bare} r, _ := git.Clone(url, path, &opts)
// Create a new remote ref. name := “my-fork” url
:= “git://github.com/wadus/go” rm, _ := r.CreateRemote(name, url) ! // Fetch all refs from a remote. var refspecs []string rm.Fetch(refspecs, nil, nil)
// Search for objects. sha := “4c279186e24f7b3a59aa682a870747df6eaca013” oid := git.NewOid(sha)
! c, _ := r.LookupCommit(oid) b, _ := r.LookupBlob(oid) t, _ := r.LookupTree(oid) ! o, _ := r.Lookup(oid) fmt.Printf(“ %v\n”, o.Type())
// Read commit data. sha := “4c279186e24f7b3a59aa682a870747df6eaca013” oid := git.NewOid(sha)
path := “src/os/exec.go” ! c, _ := r.LookupCommit(oid) t, _ := c.Tree() e, _ := t.EntryByPath(path) ! b, _ := r.LookupBlob(e.Id()) fmt.Printf(“ %q\n”, b.Contents())
// Commit new changes. idx, _ := r.Index() idx.AddByPath(“src/os/exec.go”) t,
_ := idx.WriteTree() idx.Write() ! h, _ := r.Head() c, _ := r.LookupCommit(h) ! s := &git.Signature{“me”, “
[email protected]
”, time.Now()} m := “Add moar changes” r.CreateCommit(“”, s, s, m, t, c)
None
None
Designing a distributed git storage
Application Git Service
Constraint your data model
// protocol buffers schema. message Branch { required string name
= 1; } ! message Repository { optional string name = 1; repeated Branch branches = 2; }
// Read branches. var branches []*pb.Branch ! f := func(b
*git.Branch, t git.BranchType) error { n, _ := b.Name() p := &pb.Branch{ Name: &n, } branches = append(branches, p) return nil } ! b, _ := r.NewBranchIterator(git.BranchRemote) b.ForEach(f)
// Read branches via http. h := func(w http.ResponseWriter, r
*http.Request) { pbBranches := readBranches(r) pbRepo := &pb.Repository{ Branches: pbBranches, } ! data, _ := proto.Marshal(pbRepo) w.Write(data) } ! http.HandleFunc(“/r/foo/branches”, h)
Design from first principles
A shared-data system can have at most two of the
three following properties: Consistency, Availability, and tolerance to network Partitions Dr. Eric Brewer
You Can’t Sacrifice Partition Tolerance Coda Hale
github.com/afex/hystrix-go" github.com/rubyist/circuitbreaker" github.com/eapache/go-resiliency/braker
// Read branches via http. import “github.com/rubyist/circuitbreaker” ! out :=
5 * time.Second url := “http://git-server/r/foo/branches” c := circuit.NewHTTPClient(out, 10, nil) ! c.BreakerTripped = func() { // Handle partition error response. } ! resp, _ := c.Get(url)
Replication Consistency Vs Availability
Application Git Service Git Service Primary Replica
Application Git Service Git Service Primary Replica Fetch!
Application Git Service Git Service Primary Replica Fetch!
Application Git Service Git Service Primary Replica Done!
Application Git Service Git Service Primary Replica Done!
// Handle fetch requests. h := func(w http.ResponseWriter, r *http.Request)
{ peerChannel := replicateRequest(r) rm, _ := repo.LoadRemote(“origin”) var refspecs []string rm.Fetch(refspecs, nil, nil) ! if peersChannel != nil { waitForPeers(peerChannel) } w.WriteHeader(201) } ! http.HandleFunc(“/r/foo/fetch”, h)
// Replicate request. func replicateRequest(r *http.Request) chan int { if
req.Header.Get(“X-GIT-REPLICATE”) != “” { return nil } peerChannel := make(chan int) replicaURL, err := url.Parse(replicaHost) replicaURL.Path = r.Path replicaURL.Header.Set(“X-GIT-REPLICATE”, “true") req, _ := http.NewRequest(“POST”, replicaURL.String(), nil) ! go func() { resp, _ := httpClient.Do(req) peerChannel <- resp.StatusCode }() return peerChannel }
// Wait for replica response. func waitForPeers(channel chan int) error
{ replicaStatus := <- channel switch replicaStatus { case 201: // default: // ☔️ ☔️ } ! return nil }
git architectures are fun
Thank you! @calavera