Ken Krugler
07/16/2021, 9:59 PMselect sum(metric), key, min(date) as firstSeen, max(date) as lastSeen from table where date >= <lowDate> AND date <= <highDate> group by key order by firstSeen desc limit 1
. I get a single row as expected, but with a firstSeen
and lastSeen
both equal to <highDate>
. I was expecting the firstSeen
result to be equal to <lowDate>
. If I then run the exact same query, but add in AND key = '<key value from the previous result>'
, I get the a single row with the requested key value, but now the firstSeen
result is equal to <lowDate>
(as expected), and sum(metric)
is larger (also as expected). Any ideas what is going on?Jackie
07/16/2021, 10:06 PMselect date from table where date >= <lowDate> AND date <= <highDate> AND key = '<key value from the previous result>'
?key
a very high cardinality column? If so, some groups might be dropped due to Pinot's approximation algorithm for group-by queriesKen Krugler
07/16/2021, 10:09 PMnumGroupsLimitReached: false
in response. And the difference is for a specific key value, not a missing key value.select date,metric from table WHERE date >= <lowDate> AND date <= <highDate> AND key = '<key value from previous result>'
, I get four rows. The sum of the metric values is the larger (expected) value. The one row with date = <highDate>
has the metric
value = the sum(metric)
returned from the first query.order by lastSeen
(instead of by firstSeen
), the results are more correct. firstSeen
seems right for the various rows I get back, and the sum(metric)
almost matches what I get when I add the AND key = 'specific key value'
filter.