> On 29 May 2015, at 02:58, 杨浩 <[email protected]> wrote: > > Thanks for sharing the blog. > > In the placement phase 3, will anti-affinity works with restrict policy, > i.e. if we distribute two component instances in node1 and node2, and then > the instance on node1 fails, can it restart on node1?
By default: yes. That is, if node1 itself is up. In slider 0.80, slider will request the instance on node 1, and eventually, when the escalation timeout is reached, re-issue the container request, this time saying "anywhere in the cluster". That may result in it coming up on node 2, if that is where space is found > > 2015-05-12 20:24 GMT+08:00 Steve Loughran <[email protected]>: > >> >> Given that 0.80-incubating is up for vote, please can people try it and >> give more feedback! we need more people giving votes + or -, be they >> binding or not. And yes: please be rigorous. >> >> The maven artifacts are being served up in a temporary repo; using them in >> your own build is another way to test how well things work. >> >> Please check! >> >> >> I've just posted something on what's in Slider 0.80-incubating related to >> placement >> >> >> http://steveloughran.blogspot.co.uk/2015/05/dynamic-datacentre-applications.html >> >> The new placement work is just one of the reasons you need to move up to >> slider-0.80-incubating when released -and we can't release it without those >> reviews and votes >>
