1

My understanding of estimated droppable tombstones value is = (Number of columns where tombstones passed gc_grace_seconds/Number of columns).

But in some sstablemetadata output, I could see the value is more than 1. How come it is possible? At a specific point in time, let's assume all columns become tombstone, which passed gc_grace_seconds then the value would be 1. Then how is it more than 1 subsequently...is that time dependant? i.e. as time passes, the value will be increased?

1 Answer 1

1

The inputs used to calculate the ratio of droppable tombstones are based on estimates, not absolute values, so it is not unusual for it be more than 1.0.

You may think of "number of columns" as being a fixed number but the formula for calculating the droppable ratio uses an estimated column count -- not a fixed count. The column count itself is an estimate because it is based on the estimated count of cells per partition so it is not absolute.

If you're interested, you can see how the ratio is calculated in StatsMetadata.getEstimatedDroppableTombstoneRatio(). In it you'll see that even the estimatedCellPerPartitionCount is not a single quantity but a histogram of estimates (see the EstimatedHistogram.java class). Cheers!

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.