I guess you're right. But the strange thing is that the canceled query shows in job browser as 'Running' and the percents go up - 0%, 50%, then the job is failed.
How does the cancelling actually work? Is it like the hadoop kill command? It seems to me like it works until certain phase of map reduce is done.
And another thing - after cancelling the job in Hue I can kill it with hadoop job -kill job_<id>. If it was killed already, it would show "no such job".