Hey, I'm not sure I understood your question correctly. But AFAIK recovery_agent_removal_limit flag is intended to limit the number of agents that will be marked unreachable after the re-registration timeout. If the master sees that it has to remove more agents than the limit allows, it will failover. Otherwise, agents that have not yet re-registered will be marked unreachable at slave_removal_rate_limit. Here's the code that does that: https://github.com/apache/mesos/blob/master/src/master/master.cpp#L1946
We no longer shutdown agents if they try to re-register after being marked unreachable, so we can safely remove those agents from the registry. However, it still might be a good signal for the operator to investigate why a lot of agents did not re-register. On Fri, Sep 1, 2017 at 6:46 AM, tommy xiao <xia...@gmail.com> wrote: > toady i have a curious to read mesos source code for > --recovery_agent_removal_limit. how does it working from source code. i > have not found any useful logic for recovery_agent_removal_limit. anyone > can do me favor? > > -- > Deshi Xiao > Twitter: xds2000 > E-mail: xiaods(AT)gmail.com > -- Ilya Pronin
