After some research and debugging:

• Mapper creates prepared statement first and uses it in the next executions
• prepare statement span is created and has a proper parent
• consecutive execute operations on Mapper work and have proper parent span
• the problem is with the first operation which creates prepared statement and executes it straight away. execute operation is reported with no parent. I wasn't able to make it work.

@tmszdmsk let's start from beginning.
Could you create a test which demonstrate problem you found? (Test should fail).
Then we will try to find solution.

@malafeev, sorry for the delay

I wrote a test that demonstrates the problem, it is actually why the tests are failing:
https://travis-ci.org/opentracing-contrib/java-cassandra-driver/builds/404851120#L3429

@tmszdmsk I mean what is the initial problem?
I don't think that we need to build span for prepared statement. (prepare* methods should not be changed).
Only when prepared statement is executed span should be created and finished. But it is already implemented in public ResultSet execute(Statement statement).

@malafeev then this is the test case you want (in the first commit)
aa6bf2c

I'm closing this PR because it doesn't fix the issue.
There is no proper way to propagate active span into mapper's futures.
Most likely we need to create issue against Cassandra Mapper to add such kind support.
Test is added to demonstrate issue.