'every' us a utility commonly used by workers in their core loops. It tells
them to sleep for a while if they've not done anything. During those sleeps the
worker will read data from its parent and update internal stats.
There are bug(s):
- No way for a worker to tell every to loop immediately and look for more work.
- If a tracker has spent more than the specified time doing work, it does not
fetch waiting data from the parent process.
If both are fixed the JobMaster can speed up fsck/rebalance quite a bit, and
other workers can speed through batch jobs a lot better.
Original issue reported on code.google.com by dorma...@rydia.net on 20 Sep 2010 at 12:30
Original issue reported on code.google.com by
dorma...@rydia.net
on 20 Sep 2010 at 12:30