[ 
https://issues.apache.org/jira/browse/GIRAPH-259?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13427834#comment-13427834
 ] 

Jan van der Lugt commented on GIRAPH-259:
-----------------------------------------

Ah, now I understand. Yes, this is an interesting issue. I would argue for an 
approach where the master can't add to a SumAggregator (so it cannot use the 
aggregate() call), but only use setAggregatedValue(). The vertices, in turn, 
should only use aggregate() and never setAggregatedValue(), since the latter 
would cause non-deterministic executions. I guess it's better to guard people 
from making these mistakes. Of course the master could still read the 
aggregated value, add something to it and set that as the new value, but that 
simply counts as setting a new value. In this way, the aggregator can reset to 
the default value if the master does not set a new value. This way we 
circumvent the 15/16 issue, since the master has to explicitly set the value 
every step. We would have to have 2 aggregator interfaces for this... Does this 
all make sense?

I still argue for an API without a WorkerContext handling aggregators, since I 
believe all of this can be handled by a MasterCompute at a much nicer level of 
abstraction. Only registerAggregator() and useAggregator() need to be handled 
by the WorkerContext, but isn't it easier to just let all workers register and 
use all aggregators by default? I don't think this really adds much of an 
overhead, since the state of each aggregator is really small (usually just an 
object with one primitive value). Would like to hear your thoughts on this.
                
> TestBspBasic.testBspPageRank is broken
> --------------------------------------
>
>                 Key: GIRAPH-259
>                 URL: https://issues.apache.org/jira/browse/GIRAPH-259
>             Project: Giraph
>          Issue Type: Bug
>            Reporter: Maja Kabiljo
>            Assignee: Maja Kabiljo
>         Attachments: GIRAPH-259-1.patch, GIRAPH-259-2.patch, 
> GIRAPH-259-3.patch
>
>
> Test crashes on line 152 in class SimplePageRankVertex in distributed mode.

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators: 
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira

        

Reply via email to