[ 
https://issues.apache.org/jira/browse/MESOS-1571?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14121590#comment-14121590
 ] 

Alexander Rukletsov commented on MESOS-1571:
--------------------------------------------

I wouldn't increase the number of slave flags for one reason: if we have 
several timeouts configurable through slave flags we impose on the user the 
responsibility to know how they are related and to make sure there are no 
anomalies (e.g. command executor signal escalation timeout is bigger than 
executor shutdown grace period on the slave). I would rather propagate the 
current flag value ({{flags.executor_shutdown_grace_period}}) and let executors 
calculate their timeouts based on it.

The task is now to propagate this value from slave to executor and command 
executor (and possibly custom executors). I would like to discuss several 
options:
* Set an environment variable and let the command executor read it (thanks 
[~nnielsen]). This seems to be command executor specific, but doesn't require 
changing {{.proto}} definitions. 
* Add information about slave's grace period to {{SlaveInfo}} and let all 
interested parties extract it. This approach requires updating protobuf 
definitions, which is nevertheless the case if we introduce the timeout to the 
{{TaskInfo}}. To make it more generic, we can introduce a repeated key-value 
field into the {{SlaveInfo}} (reuse {{Parameter}} message or make a new 
{{Flag}} message with optional value?).

Any other ideas?

> Signal escalation timeout is not configurable
> ---------------------------------------------
>
>                 Key: MESOS-1571
>                 URL: https://issues.apache.org/jira/browse/MESOS-1571
>             Project: Mesos
>          Issue Type: Bug
>            Reporter: Niklas Quarfot Nielsen
>            Assignee: Alexander Rukletsov
>
> Even though the executor shutdown grace period is set to a larger interval, 
> the signal escalation timeout will still be 3 seconds. It should either be 
> configurable or dependent on EXECUTOR_SHUTDOWN_GRACE_PERIOD.
> Thoughts?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to