The precision returned by uniqCombined is too low about clickhouse

[serxa]

It is quite unusual. Could you provide more information about your dataset? How many strings, what is the average length, high or low cardinality? Also, it would be interesting to see the results of queries comparing how uniqExact, uniq, uniqCombined, and uniqCombined64 behave on your dataset.

[nickitat]

any specific example when the problem reproduces?
have you tried changing HLL_precision parameter?

[den-crane]

uniqCombined is for int32 numbers and strings (with strings it works as uniqCombined64). Try uniqCombined64 instead.

uniqCombined(2, xx)

Haha, in this case you pass a tuple(2,xx) and it starts to behave as uniqCombined64. Just use uniqCombined64.

• #63066

[den-crane]

@nickitat Maybe uniqCombined should throw an error if not 32 bit argument is passed and show an explanation with a suggestion to use uniqCombined64? And the setting allow_unsafe_uniqCombined=false for the backward compatibility in queries and states saved with materialized views and projections.

[crazyleeyang]

It is quite unusual. Could you provide more information about your dataset? How many strings, what is the average length, high or low cardinality? Also, it would be interesting to see the results of queries comparing how uniqExact, uniq, uniqCombined, and uniqCombined64 behave on your dataset.

It is quite unusual. Could you provide more information about your dataset? How many strings, what is the average length, high or low cardinality? Also, it would be interesting to see the results of queries comparing how uniqExact, uniq, uniqCombined, and uniqCombined64 behave on your dataset.

The statistics of column length arre as follows

┌─col_len─┬──count()─┐
│       8 │        2 │
│      11 │       13 │
│      13 │       48 │
│      14 │     1078 │
│      15 │    14925 │
│      16 │  2017052 │
│      17 │        2 │
│      19 │        1 │
│      24 │        5 │
│      32 │  8241286 │
│      33 │        1 │
│      34 │       22 │
│      35 │      407 │
│      36 │  7784682 │
│      40 │        1 │
│      42 │        2 │
│      50 │        7 │
│      64 │ 10931446 │
│      65 │        2 │
│      71 │        7 │
│      82 │      245 │
│     116 │        2 │
│     146 │       76 │
└─────────┴──────────┘

[crazyleeyang]

any specific example when the problem reproduces? have you tried changing HLL_precision parameter?

i have test with HLL_precision, it works ok