uniqCombined is wrong

[alexey-milovidov]

Describe the unexpected behaviour

clickhouse-cloud :) SELECT count(), uniq(*) AS a, uniqCombined(*) AS b FROM planes_mercator \G

SELECT
    count(),
    uniq(*) AS a,
    uniqCombined(*) AS b
FROM planes_mercator

Query id: 5202ef0c-5f71-455c-af0f-95f305fd06a4

Row 1:
──────
count(): 50738937576
a:       48882212795
b:       6195774276

1 row in set. Elapsed: 56.120 sec. Processed 50.74 billion rows, 16.27 TB (904.11 million rows/s., 289.99 GB/s.)
Peak memory usage: 5.05 GiB.

WTF?

[UnamedRus]

Probably, it's using 32bit based state, because of data types in *

SELECT uniqCombined64(number)
FROM numbers_mt(20000000000)

Query id: 33ae687b-2426-4c39-a8a1-9252ce613d7f

┌─uniqCombined64(number)─┐
│            20058723685 │ -- 20.06 billion
└────────────────────────┘

1 row in set. Elapsed: 19.479 sec. Processed 19.96 billion rows, 159.71 GB (1.02 billion rows/s., 8.20 GB/s.)
Peak memory usage: 1.83 MiB.

SELECT uniqCombined(number)
FROM numbers_mt(20000000000)

Query id: 9ffb671e-f417-41bc-b933-288e7b770cf4

┌─uniqCombined(number)─┐
│           6138591160 │ -- 6.14 billion
└──────────────────────┘

1 row in set. Elapsed: 19.841 sec. Processed 20.00 billion rows, 160.00 GB (1.01 billion rows/s., 8.06 GB/s.)
Peak memory usage: 1.38 MiB.