After https://github.com/apache/mesos/commit/447d814ac80e67f30a0ffe2ee6047d85dc8fc383 the refuse_seconds should not be as dangerous as it used to be. This was backported to 0.24 and above recently. Without the backported patch the filter may never take effect when the allocation takes longer than the actual filter.
On Tue, Apr 26, 2016 at 3:34 PM, Erb, Stephan <stephan....@blue-yonder.com> wrote: > Hi everyone, > > within this Aurora review request https://reviews.apache.org/r/46603/ we > are wondering about the current Filter.refuse_seconds default value of 5 > seconds [1]. > > Is there any reason why this particular value was chosen? Would we have to > expect increased load for the leading Mesos master of extremely large > clusters if we set it to 0? > > Thanks for your feedback!, > Stephan > > > [1] > https://github.com/apache/mesos/blob/3d2d3edb96c9c428f988653cdd4428a05690e747/include/mesos/mesos.proto#L1423