Hi, For scale testing we've implemented a containerizer and an executor that consume no resources. The containerizer runs an executor that sends `TASK_RUNNING` on `launchTask()` and `TASK_KILLED` on `killTask()`. That way any amount of resources can be offered by the agent (through `--resources` option) and any number of tasks can be run.
On Sat, Jan 7, 2017 at 1:07 AM, Vinod Kone <vinodk...@apache.org> wrote: > Great to hear! > > Haven't looked at the doc yet, but I know some folks from Twitter were also > interested this. https://issues.apache.org/jira/browse/MESOS-6768 > > Probably worth to see if the ideas can be consolidated? > > On Fri, Jan 6, 2017 at 6:57 PM, Zhitao Li <zhitaoli...@gmail.com> wrote: > > > (sending this again since previous attempt seemed bumped back) > > > > Hi folks, > > > > As all of you we are super excited to use Mesos to manage thousands of > > different applications on our large-scale clusters. When the application > > and host amount keeps increasing, we are getting more and more curious > > about what would be the potential scalability limit/bottleneck to Mesos' > > centralized architecture and what is its robustness in the face of > various > > failures. If we can identify them in advance, probably we can manage and > > optimize them before we are suffering in any potential performance > > degradations. > > > > To explore Mesos' capability and break the knowledge gap, we have a > > proposal to evaluate Mesos scalability and robustness through stress > test, > > the draft of which can be found at: draft_link > > <https://docs.google.com/document/d/10kRtX4II74jfUuHJnX2F5te > > qpXzHYFQAZGWjCdS3cZA/edit?usp=sharing>. > > Please > > feel free to provide your suggestions and feedback through comment on the > > draft. > > > > Probably many of you have similar questions as we have. We will be happy > to > > share our findings in these experiments with the Mesos community. Please > > stay tuned. > > > > -- > > Cheers, > > > > Ao Ma & Zhitao Li > > >
