summaryrefslogtreecommitdiff
path: root/src/fabric_view_reduce.erl
blob: 1a619877ab897389e3595f61c916bfab4bbeae52 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
-module(fabric_view_reduce).

-export([go/5]).

-include("fabric.hrl").

go(DbName, {GroupId, VName}, Args, Callback, Acc0) ->
    {ok, DDoc} = fabric:open_doc(DbName, <<"_design/", GroupId/binary>>, []),
    #group{def_lang=Lang, views=Views} = Group =
        couch_view_group:design_doc_to_view_group(#db{name=DbName}, DDoc),
    {NthRed, View} = fabric_view:extract_view(nil, VName, Views, reduce),
    {VName, RedSrc} = lists:nth(NthRed, View#view.reduce_funs),
    Workers = lists:map(fun(#shard{name=Name, node=N} = Shard) ->
        Ref = rexi:cast(N, {fabric_rpc, reduce_view, [Name,Group,VName,Args]}),
        Shard#shard{ref = Ref}
    end, partitions:all_parts(DbName)),
    BufferSize = couch_config:get("fabric", "reduce_buffer_size", "20"),
    #view_query_args{limit = Limit, skip = Skip} = Args,
    State = #collector{
        query_args = Args,
        callback = Callback,
        buffer_size = list_to_integer(BufferSize),
        counters = fabric_dict:init(Workers, 0),
        skip = Skip,
        limit = Limit,
        lang = Group#group.def_lang,
        os_proc = couch_query_servers:get_os_process(Lang),
        reducer = RedSrc,
        rows = dict:new(),
        user_acc = Acc0
    },
    try fabric_util:receive_loop(Workers, #shard.ref, fun handle_message/3,
        State, infinity, 5000) of
    {ok, NewState} ->
        {ok, NewState#collector.user_acc};
    Error ->
        Error
    after
        fabric_util:cleanup(Workers),
        catch couch_query_servers:ret_os_process(State#collector.os_proc)
    end.

handle_message({rexi_DOWN, _, _, _}, nil, State) ->
    % TODO see if progress can be made here, possibly by removing all shards
    % from that node and checking is_progress_possible
    {ok, State};

handle_message({rexi_EXIT, Reason}, Worker, State) ->
    ?LOG_ERROR("~p rexi_EXIT ~p", [?MODULE, Reason]),
    #collector{callback=Callback, counters=Counters0, user_acc=Acc} = State,
    Counters = fabric_dict:erase(Worker, Counters0),
    case fabric_view:is_progress_possible(Counters) of
    true ->
        {ok, State#collector{counters = Counters}};
    false ->
        Callback({error, dead_shards}, Acc),
        {error, dead_shards}
    end;

handle_message(#view_row{key=Key} = Row, {Worker, From}, State) ->
    #collector{counters = Counters0, rows = Rows0} = State,
    case fabric_dict:lookup_element(Worker, Counters0) of
    undefined ->
        % this worker lost the race with other partition copies, terminate it
        gen_server:reply(From, stop),
        {ok, State};
    % first ->
    %     gen_server:reply(From, ok),
    %     Rows = dict:append(Key, Row#view_row{worker=Worker}, Rows0),
    %     C1 = fabric_dict:store(Worker, 1, Counters0),
    %     C2 = fabric_view:remove_overlapping_shards(Worker, C1),
    %     NewState = State#collector{counters=C2, rows=Rows},
    %     case fabric_dict:any(first, C2) of
    %     true ->
    %         {ok, NewState};
    %     false ->
    %         fabric_view:maybe_send_row(State#collector{counters=C2, rows=Rows})
    %     end;
    _ ->
        Rows = dict:append(Key, Row#view_row{worker=Worker}, Rows0),
        C1 = fabric_dict:update_counter(Worker, 1, Counters0),
        C2 = fabric_view:remove_overlapping_shards(Worker, C1),
        State1 = State#collector{rows=Rows, counters=C2},
        State2 = fabric_view:maybe_pause_worker(Worker, From, State1),
        fabric_view:maybe_send_row(State2)
    end;

handle_message(complete, Worker, State) ->
    Counters = fabric_dict:update_counter(Worker, 1, State#collector.counters),
    fabric_view:maybe_send_row(State#collector{counters = Counters}).