Influxdb 1.8.10 Some Queries are Super Slow

[tomcheney]

I have an influx database where we have around 20,000 measurements with only a few fields being written at quite high rate (10-200hz typically) for periods of the day. I am starting to run into an issue where some simple queries (ie mean group by 10s for a day) that should be taking less than a second are taking more than a minute. It's odd because this is happening to only a few measurements and then often a few days later the issue seems to resolve itself.

I have tried rebuilding indexes and manually compacting series with the influx_inspect buildtsi helpers but this has no effect. Are there some compaction/sharding settings that I need to tweak? Could this be caused by writing data out of order or large blocks of historical data?

[tomcheney]

When I find one of these corrupted/broken measurements I am able to resolve the issue by doing a SELECT INTO to copy fields to a new measurement, then dropping the old measurement and then reversing the copy from my temporary measurement, but this is very time consuming.