起因

在做标签引擎的时候,咱们在采纳了bitmap存储对象id,根底的构造如下

标签类型标签值对象id bitmap
性别[1,2,3]
性别[8,9,10]

表如下:

create table if not exists label_string_local on cluster clickhouse_cluster(    label_type  String comment '标签id',    label_value String comment '标签值',    object_bitmap AggregateFunction(groupBitmap, UInt32) comment '标签值')    engine = AggregatingMergeTree PARTITION BY label_type        ORDER BY (label_type, label_value)        SETTINGS index_granularity = 8192;

到前面需要要求对对象id分页返回,问题就来了,clickhouse的官网没有bitmap的分页函数,最原始的解决方案就是把bitmap整个返回,在应用层对bitmap进行切割,这样导致接口的性能急剧下降。开始萌发了个大胆的想法,给clickhouse增加bitmap分页函数

开干

通过浏览Clickhouse的源码,步骤如下:

  1. 实现分页

在Clickhouse中bitmap指向的class是RoaringBitmapWithSmallSet ,bitmap底层应用的是RoaringBitmap,github地址:https://github.com/RoaringBitmap/CRoaring.git ,RoaringBitmapWithSmallSet对rb进行了包装,在这个类下增加分页函数

   UInt64 rb_offset_limit(UInt64 offset, UInt64 limit, RoaringBitmapWithSmallSet & r1) const    {        if (limit == 0 || offset >= size())            return 0;        if (isSmall())        {            UInt64 count = 0;            UInt64 offset_count = 0;            auto it = small.begin();            for (;it != small.end() && offset_count < offset; ++it)                ++offset_count;            for (;it != small.end() && count < limit; ++it, ++count)                r1.add(it->getValue());            return count;        }        else        {            UInt64 count = 0;            UInt64 offset_count = 0;            auto it = rb->begin();            for (;it != rb->end() && offset_count < offset; ++it)                ++offset_count;            for (;it != rb->end() && count < limit; ++it, ++count)                r1.add(*it);            return count;        }    }
  1. Clickhouse函数定义

FunctionsBitmap.h定义Clickhouse函数

struct BitmapSubsetOffsetLimitImpl{public:    static constexpr auto name = "subBitmap";    template <typename T>    static void apply(        const AggregateFunctionGroupBitmapData<T> & bitmap_data_0,        UInt64 range_start,        UInt64 range_end,        AggregateFunctionGroupBitmapData<T> & bitmap_data_2)        {        bitmap_data_0.rbs.rb_offset_limit(range_start, range_end, bitmap_data_2.rbs);        }};using FunctionBitmapSubsetOffsetLimit = FunctionBitmapSubset<BitmapSubsetOffsetLimitImpl>;
  1. Clickhouse函数注册

FunctionsBitmap.cpp注册函数

#include <Functions/FunctionFactory.h>// TODO include this last because of a broken roaring header. See the comment inside.#include <Functions/FunctionsBitmap.h>namespace DB{void registerFunctionsBitmap(FunctionFactory & factory){    ...    factory.registerFunction<FunctionBitmapSubsetOffsetLimit>();    ...}}

这样就完事了,最终这部分的代码提交到了Clickhosue仓库,最终失去了合并,https://github.com/ClickHouse/ClickHouse/pull/27234

后续

前面又来了个需要,要求标签可能批改,这又炸了,Clickhosue是不反对批改的,bitmap采纳的数据结构是AggregateFunction(groupBitmap, UInt32),groupBitmap的合并逻辑是或运算,外部Clickhosue开发了一种新的数据结构xor_groupBitmap,反对合并逻辑异或运算,变相反对删除操作,思考这部分并不通用,所以没有开源进去

广州4年java求内推,https://dhbin.cn