> On May 14, 2016, 2:01 a.m., Qian Zhang wrote:
> > src/slave/slave.cpp, line 1020
> > <https://reviews.apache.org/r/47209/diff/3/?file=1382234#file1382234line1020>
> >
> >     If we do the link here, that means we still establish TCP connection 
> > with master right after a new master is detected, right? But I think we 
> > want to do the link after the initial backoff to avoid SYN flood.
> 
> Cong Wang wrote:
>     Authentication code path doesn't have a backoff, so we don't want to 
> touch it in this patch.
> 
> Qian Zhang wrote:
>     Thanks, so that means if there are a large number of agents in the 
> cluster and each agent has authentication enabled, when new master is 
> detected, it is still possible all those agents try to connect with master at 
> the same time?

If they use TCP, I think so. Also, there is already a TODO saying we need a 
backoff for authentication too.


- Cong


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/47209/#review133243
-----------------------------------------------------------


On May 13, 2016, 3:37 a.m., David Robinson wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/47209/
> -----------------------------------------------------------
> 
> (Updated May 13, 2016, 3:37 a.m.)
> 
> 
> Review request for mesos, Benjamin Mahler, Ian Downes, and Cong Wang.
> 
> 
> Bugs: MESOS-5330
>     https://issues.apache.org/jira/browse/MESOS-5330
> 
> 
> Repository: mesos
> 
> 
> Description
> -------
> 
> Establish TCP connection after backing off.
> 
> 
> Diffs
> -----
> 
>   src/slave/slave.cpp 116ea59b72950db4a7cd721b7ba5bfbb2e1c1454 
> 
> Diff: https://reviews.apache.org/r/47209/diff/
> 
> 
> Testing
> -------
> 
> make check
> 
> Also started a master and agent and captured a tcpdump to verify the 
> connection is only established _after_ the registration backoff.
> 
> 
> Thanks,
> 
> David Robinson
> 
>

Reply via email to