优先时效性，全量落后

pq压缩

cb：term召回

近线召回：预测用户下次刷新，将没进入精算的结果加入缓存。用户发起请求之前就会进行预计算

grg:多样性， buffer, context dnn 与模型相关的服务

强id & 离散dnn，模型退化严重

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16

    DemoProcessor processor;
    BthreadGraphExecutor executor;
    GraphBuilder builder;
    builder.executor(executor);
    {
        auto& v = builder.add_vertex(processor);
        v.named_depend("a").to("A");
        v.named_emit("B").to("B");
        v.named_emit("C").to("C");
    }
    builder.finish();
    auto graph = builder.build();
    auto a = graph->find_data("A");
    auto b = graph->find_data("B");
    auto c = graph->find_data("C");
}