> On June 10, 2014, 1:22 a.m., Ben Mahler wrote:
> > Please do a grep for 'quorum', I believe there are a few other places that 
> > need to be modified (mesos-local and tests).
> > 
> > What if both flags are specified? We should either reject this or validate 
> > the values (e.g. quorum=1 masters=3 is not safe, quorum=4 masters=3 is 
> > invalid).
> 
> Chengwei Yang wrote:
>     I see, there is another bug in auto initialization code, which get 
> clusterSize as quorum * 2 - 1, this is not correct, say below examples.
>     
>     num-masters quorum clusterSize
>     1           1      1
>     2           2      3 (wrong)
>     3           2      3
>     4           3      5 (wrong)
>     5           3      5
>     ...
>     
>     So need more changes to code. I'd like to describe my idea before submit 
> a new version of patch.
>     
>     1. since 0.19.0 is freezing, this patch will go 0.20.x
>     
>     There are two options I'd like to hear from you.
>     
>     1. --num_masters will replace --quorum, means overwrite --quorum for 
> several releases and finally remove '--quorum', so no non-standard quorum 
> will be used. quorum = num_master / 2 + 1
>     2. both --num_masters and --quorum are used, user can specify 
> non-standard quorum if he like, specify a larger quorum may introduce 
> performance regression while a smaller quorum may introduce inconsistency.
>     
>     Generally, I vote the first option, how do you think?

I think it would be nice to get to your suggestion (2):

--num_masters (required)
--quorum (optional, for those looking for stronger replication at the cost of 
performance)

If we pass in --num_masters to the log code as the clusterSize, is there 
anything you don't like about this option?


How about the following phased transition:

0.20.0: At least one of --quorum or --num_masters is required. This lets those 
using --quorum upgrade without effort and allows them to add --num_masters 
while running this version. We pass --num_masters to the relevant log 
initialization code.

0.21.0: --num_masters (required), --quorum (optional)

How does that sound to you?


- Ben


-----------------------------------------------------------
This is an automatically generated e-mail. To reply, visit:
https://reviews.apache.org/r/22369/#review45186
-----------------------------------------------------------


On June 10, 2014, 1:03 a.m., Chengwei Yang wrote:
> 
> -----------------------------------------------------------
> This is an automatically generated e-mail. To reply, visit:
> https://reviews.apache.org/r/22369/
> -----------------------------------------------------------
> 
> (Updated June 10, 2014, 1:03 a.m.)
> 
> 
> Review request for mesos.
> 
> 
> Bugs: https://issues.apache.org/jira/browse/MESOS-1465
>     
> https://issues.apache.org/jira/browse/https://issues.apache.org/jira/browse/MESOS-1465
> 
> 
> Repository: mesos-git
> 
> 
> Description
> -------
> 
> Use '--nr_of_masters' instead of '--quorum' because user may confused
> with zookeeper quorum and may specify an incorrect quorum if he/she
> doesn't know quorum well.
> 
> Use '--nr_of_masters' just simple and straight-forward, and quorum will
> be nr_of_masters / 2 + 1, since specify a larger quorum doesn't make
> sense in pratice.
> 
> Review: https://reviews.apache.org/r/22369
> 
> 
> Diffs
> -----
> 
>   src/master/flags.hpp 4863359 
>   src/master/main.cpp 8ceaae6 
> 
> Diff: https://reviews.apache.org/r/22369/diff/
> 
> 
> Testing
> -------
> 
> done.
> 
> 
> Thanks,
> 
> Chengwei Yang
> 
>

Reply via email to