anarcat authored May 31, 2010 and anarcat committed May 31, 2010

we are already saving the task in hosting-task. the reason why we were
saving in both places was to avoid a race condition where a task would
be started *while* the dispatcher would restart, which is very unlikely,
and was still possible with the existing code anyways.

now the checks are done only in hosting-task. it's more possible to have
a conflicting concurrent run of a task, but both runs would need to
start at about the same time so that the _validate() drush hooks would
run in parallel and both would check and save the task status at once.

i consider this too unlikely to desserve the current hack.

furthermore, the problem with the current code was that if the
hosting-task command was interrupted (for example through
provision_init() load checks) before it was able to save the status, we
end up with zombie tasks that are marked as running but that are gone
from the process table (because they never started properly).