Support hll_raw_agg in Aggregate Function #832
Conversation
be/src/exprs/aggregate_functions.cpp
Outdated
| DCHECK(!src.is_null); | ||
| DCHECK_EQ(src.len, HLL_SETS_BYTES_NUM); | ||
|
|
||
| StringVal result(ctx, src.len+1); |
There was a problem hiding this comment.
| StringVal result(ctx, src.len+1); | |
| StringVal result(ctx, src.len + 1); |
| FunctionContext* ctx, | ||
| StringVal* dst) { | ||
| hll_union_agg_init(ctx, dst); | ||
| } |
There was a problem hiding this comment.
All calculations can be based on full type, so dst will be initted to a hll full type .
zhaidongbo
force-pushed
the
feature-HllRawAgg
branch
from
261759f to
0ac20a6
Compare
be/src/exprs/aggregate_functions.cpp
Outdated
| @@ -25,6 +23,7 @@ | |||
| #include "runtime/string_value.h" | |||
| #include "runtime/datetime_value.h" | |||
| #include "exprs/anyval_util.h" | |||
| #include "exprs/aggregate_functions.h" | |||
There was a problem hiding this comment.
You'd better not move #include "exprs/aggregate_functions.h", we put this first to make sure that all need includes is written in header files
| select HLL_CARDINALITY(uv_set) from test_uv; | ||
| select dt, HLL_CARDINALITY(uv_set) from test_uv; | ||
| select dt, HLL_CARDINALITY(uv) from (select dt, HLL_RAW_AGG(set1) as uv from test group by dt) tmp; | ||
| select dt, HLL_UNION_AGG(set1) as uv from test group by dt; |
There was a problem hiding this comment.
I think that these three queries don't lead to same result.
1 would return many rows for one dt value
2 and 3 queries would group by dt column and return one row for each value
So, you should split them into different examples
be/src/udf/udf.h
Outdated
| return (doris::HllDataType)(ptr[0]); | ||
| } | ||
|
|
||
| uint8_t &getHllData(int idx) const { |
There was a problem hiding this comment.
Don't return a reference value, you can add a function named setHllData
be/src/exprs/aggregate_functions.h
Outdated
| return hll_algorithm(dst.ptr, dst.len); | ||
| } | ||
| static int64_t hll_algorithm(const HllVal &dst) { | ||
| return hll_algorithm(dst.ptr+1, dst.len-1); |
There was a problem hiding this comment.
| return hll_algorithm(dst.ptr+1, dst.len-1); | |
| return hll_algorithm(dst.ptr + 1, dst.len - 1); |
zhaidongbo
force-pushed
the
feature-HllRawAgg
branch
from
0ac20a6 to
6276230
Compare
be/src/udf/udf.h
Outdated
| @@ -762,6 +766,45 @@ struct LargeIntVal : public AnyVal { | |||
| } | |||
| }; | |||
|
|
|||
| struct HllVal : public StringVal { | |||
|
|
|||
There was a problem hiding this comment.
I think HllFullVal is a better name, because your HllVal only represent the Full type.
be/src/exprs/aggregate_functions.cpp
Outdated
| } | ||
|
|
||
| void AggregateFunctions::hll_union_parse_and_cal(HllSetResolver& resolver, StringVal* dst) { | ||
| void AggregateFunctions::hll_union_parse_and_cal(HllSetResolver& resolver, HllVal* dst) { | ||
|
|
There was a problem hiding this comment.
hll_union_parse_and_cal can be a member of HllVal， and it support a function that merge other type or src StringVal, so we will get a better encapsulation for hll.
be/src/exprs/aggregate_functions.cpp
Outdated
|
|
||
| for (int i = 0; i < src.len; ++i) { | ||
| dst->ptr[i] = std::max(dst->ptr[i], src.ptr[i]); | ||
| dst->setHllData(i, std::max(dst->getHllData(i), src.getHllData(i))); | ||
| } |
There was a problem hiding this comment.
HLLVal can support a member of merge other HllVal.
be/src/udf/udf.h
Outdated
| DCHECK(!is_null); | ||
|
|
||
| return (doris::HllDataType)(ptr[0]); | ||
| } |
There was a problem hiding this comment.
return doris::HllDataType::HLL_DATA_FULL.
hll_raw_agg Function aggregates the HLL type value, and return the HLL type value
zhaidongbo
force-pushed
the
feature-HllRawAgg
branch
from
6276230 to
f62f7c1
Compare
…he#832) Also, change some config variable names of file cache.
Support hll_raw_agg in Aggregate Function
hll_raw_agg Function aggregates the HLL type value, and return the HLL type value