GitHub Pages on Riak and Webmachine

GitHub Pages, a feature allowing users to publish content to the web by simply pushing content to one of their GitHub hosted repositories, has had lackluster performance and uptime in the recent years. In this talk, Jesse will discuss the core requirements of the GitHub Pages application, why Erlang, Riak, and Webmachine were chosen for the development, and how they were used to fulfill those requirements now and for years to come with minimal development and operational maintenance.

Transcript

1. riak &
   webmachine
   on
   github:pages
   

   View Slide

2. github.com/jnewland
   @jnewland
   

   View Slide

3. riak &
   webmachine
   on
   github:pages
   

   View Slide

4. what
   is
   pages?
   

   View Slide

5. simple
   static
   file
   hosting*
   

   View Slide

6. View Slide

7. github.com/user/repo/gh-pages
   „
   http://user.github.com/repo
   

   View Slide

8. View Slide

9. github.com/user/user.github.com
   „
   http://user.github.com
   

   View Slide

10. View Slide

11. View Slide

12. $ cat CNAME
    example.com
    custom domain support
    

    View Slide

13. what
    is wrong with
    pages?
    

    View Slide

14. SLOW
    

    View Slide

15. 1 node
    ext3
    gigs and gigs of HTML
    

    View Slide

16. IO bound
    

    View Slide

17. fscking downtime
    

    View Slide

18. not HA
    

    View Slide

19. LOTS
    TO
    FIX
    THROW
    IT ALL
    OUT
    

    View Slide

20. what
    can power
    pages 2.0?
    

    View Slide

21. 1. grab content from git
    2. run through jekyll
    3. write somewhere
    4. serve over HTTP
    

    View Slide

22. what
    currently powers
    pages?
    

    View Slide

23. 1. ruby
    2. ruby
    3. ext3
    4. nginx
    

    View Slide

24. $ wc -l pages_map.conf
    57615 pages_map.conf
    

    View Slide

25. 1. ruby
    2. ruby
    3. riak_kv
    4. riak_kv*
    

    View Slide

26. *
    read-only
    transactional builds
    CNAMEs & redirects
    index.html fallback
    custom 404.html per-repo
    almost
    

    View Slide

27. 1. ruby
    2. ruby
    3. riak_kv
    4. webmachine resource
    

    View Slide

28. why
    not
    x?
    

    View Slide

29. squid
    +
    nginx/apache
    +
    filesystem
    

    View Slide

30. should work fine
    what happens when you need N > 1?
    just shard ‘em
    how do you populate new partitions?
    just rsync stuff around
    

    View Slide

31. building a
    distributed
    system
    ASS
    FIRST
    

    View Slide

32. remember, I do ops
    low maintenance, resilient
    systems make lazy
    sysadmins love you
    

    View Slide

33. View Slide

34. schema
    design
    

    View Slide

35. 2 buckets
    hostspages
    

    View Slide

36. hosts
    key: HTTP Host Header
    value: redirect or repo/sha map
    use: data key prefix lookup
    index: user_id
    

    View Slide

37. hosts
    {
    “repos”: {
    “jnewland.github.com”: “deadbeef”,
    “pages”: “beadfeed”
    }
    }
    jnewland.github.com
    

    View Slide

38. pages
    key: sha/URI
    value: HTML / other data
    use: data storage
    index: repo_id
    

    View Slide

39. 
    
    
    content="text/html; charset=utf-8" />
    Jesse Newland
    

    View Slide

40. how does
    data get in?
    

    View Slide

41. riak-ruby-client
    read files from disk
    write to riak
    update sites object
    gc old builds
    

    View Slide

42. how does
    data get out?
    

    View Slide

43. curl jnewland.github.com/raptor.gif
    GET /riak/hosts/jnewland.github.com
    GET /riak/pages/deadbeef/raptor.gif
    

    View Slide

44. curl jnewland.github.com/pages/flow.png
    GET /riak/hosts/jnewland.github.com
    GET /riak/pages/beadfeed/flow.png
    

    View Slide

45. what
    about
    CNAMEs?
    

    View Slide

46. hosts
    example.github.com
    {
    “redirect”: “example.com”
    }
    

    View Slide

47. hosts
    example.com
    {
    “repos”: {
    “example.com”: “deadbeef”
    }
    }
    

    View Slide

48. curl example.github.com/raptor.gif
    GET /riak/hosts/example.github.com
    301 example.com/raptor.gif
    GET /riak/hosts/example.com
    GET /riak/pages/deadbeef/raptor.gif
    

    View Slide

49. pages_wm_resource.erl
    

    View Slide

50. webmachine
    is
    SO DAMN COOL
    

    View Slide

51. View Slide

52. %% webmachine resource exports
    -export([
    init/1,
    service_available/2,
    malformed_request/2,
    content_types_provided/2,
    resource_exists/2,
    previously_existed/2,
    moved_permanently/2,
    last_modified/2,
    generate_etag/2,
    produce_doc_body/2
    ]).
    

    View Slide

53. grab local riak client
    check app config var
    service_available/2
    

    View Slide

54. service_available(RD, Ctx=#ctx{riak=RiakProps,req_id=ReqId}) ->
    IdRD = wrq:set_resp_header("X-Request-Id", ReqId, RD),
    BrandedRD = wrq:set_resp_header(
    "X-GitHub-Pages-Version",
    release_handler_util:app_version(pages),
    IdRD),
    case application:get_env(pages, disabled) of
    {ok, true} ->
    {false, BrandedRD, Ctx};
    _ ->
    case riak_kv_wm_utils:get_riak_client(
    RiakProps,
    riak_kv_wm_utils:get_client_id(RD)) of
    {ok, C} ->
    {true, BrandedRD, Ctx#ctx{client=C}};
    _Error ->
    {false, BrandedRD, Ctx}
    end
    end.
    

    View Slide

55. parse Host header, URI
    malformed_request/2
    

    View Slide

56. malformed_request(RD, Ctx) ->
    try
    Host = wrq:get_req_header("Host", RD),
    HostWithoutPort = re:replace(
    Host,
    "\:.*",
    "",
    [{return,list}]),
    Tokens = [
    riak_kv_wm_utils:maybe_decode_uri(RD, X) ||
    X <- wrq:path_tokens(RD)],
    ParsedCtx = Ctx#ctx{tokens=Tokens,host=HostWithoutPort},
    {false, RD, ParsedCtx}
    catch
    Exception:Reason ->
    log_error({exception, Exception, Reason}, RD, Ctx),
    {true, RD, Ctx}
    end.
    

    View Slide

57. guess mime type from path
    assume text/html for / URIs
    content_types_provided/2
    

    View Slide

58. content_types_provided(RD, Ctx) ->
    Filename = lists:last(Ctx#ctx.tokens),
    Extension = filename:extension(Filename),
    case mochiweb_mime:from_extension(Extension) of
    undefined ->
    {[{"text/html", produce_doc_body}], RD, Ctx};
    Mime ->
    {[{Mime, produce_doc_body}], RD, Ctx}
    end.
    

    View Slide

59. hit hosts bucket (r=1)
    stash redirect or sha
    404 if no hosts data
    resource_exists/2
    

    View Slide

60. resource_exists(RD, Ctx) ->
    RedirectOrSha = redirect_or_sha(Ctx),
    case RedirectOrSha of
    {redirect, Redirect} ->
    {true, RD, Ctx#ctx{redirect={redirect, Redirect}}};
    {sha, Sha} ->
    page_data_exists(RD, Ctx#ctx{sha={sha, Sha}});
    _ ->
    {false, RD, Ctx}
    end.
    

    View Slide

61. previously_existed/2
    moved_permanently/2
    

    View Slide

62. previously_existed(RD, Ctx) ->
    case Ctx#ctx.redirect of
    {redirect, _} ->
    {true, RD, Ctx};
    _ ->
    {false, RD, Ctx}
    end.
    moved_permanently(RD, Ctx) ->
    case Ctx#ctx.redirect of
    {redirect, RedirectHost} ->
    MovedURI = list_join(lists:append(
    [RedirectHost],
    Ctx#ctx.tokens),
    "/"),
    {{true}, MovedURI, RD, Ctx};
    _ ->
    {false, RD, Ctx}
    end.
    

    View Slide

63. hit pages bucket (r=1)
    fallback for index.html
    fallback for 404.html
    page_data_exists/2
    

    View Slide

64. curl foo.github.com/
    GET /riak/hosts/foo.github.com
    GET /riak/pages/f0f0f0f0/
    GET /riak/pages/f0f0f0f0/index.html
    GET /riak/pages/f0f0f0f0/index.htm
    GET /riak/pages/f0f0f0f0/index.xhtml
    GET /riak/pages/f0f0f0f0/index.xml
    GET /riak/pages/f0f0f0f0/404.html
    

    View Slide

65. < 300 lines of erlang
    simple
    

    View Slide

66. {webmachine, [{
    dispatch_list, [{
    %% riak_kv stuff
    {["pages",'*'],pages_wm_resource,[]},
    {["pages"],pages_wm_resource,[]}]}}
    ]}
    nginx proxies to / /pages
    

    View Slide

67. remember, I do ops
    one system service
    data store and api
    predictable performance
    busy ops best friend
    

    View Slide

68. what’s
    next
    

    View Slide

69. metrics with folsom
    graphite / gaug.es
    logging with lager
    HTTP caching
    ???
    

    View Slide

70. private beta soon
    turn on / off with DNS
    repo access
    @jnewland
    

    View Slide

71. erlang
    ruby
    ops
    c
    work with me
    

    View Slide

72. thanks
    

    View Slide