Today I had to kill quite a large hive generated MR job. The progress on the
mappers was reversed halfway (so actually declining). When I got to the
local mapped logs from the TT I saw that there were up to 33 task attempts
on the mappers that stalled and it looked like this would go on forever. 33
task attempts and more per mapper will completely flood your JT with admin
The strange thing however is that in my hadoop config the mapper max task
attempts is set to 4 (before failing the whole job).
Why is this Hive MR job not responding to this setting and how can I enforce