Closed thalesmg closed 1 year ago
Thanks!
Hi @uwiger , do you have plans to tag a new version that would contain this fix?
We are interested in using the upstream version if possible.
Thanks! :smile_cat:
Published version 0.9.1
Thank you! :beers:
Since
rand:uniform/1
's return value is a number in a closed interval (0 =< N =< Sz
), we shouldn't call it with the pool size + 1, otherwise we will get a non-uniform distribution of results where one of the workers receive substantially more work than all others.To verify this, I've started a pool with 8 workers using the
random
load-balancing algorithm and measured the frequencies of each PID ingproc_pool
.Results prior to the fix (note how
<0.172.0>
has almost double the frequency of other PIDs):After the fix: