Billie Rinaldi commented on YARN-8060:

Thanks for the reviews, [~shaneku...@gmail.com] and [~gsaha]!

bq. do you think it would make sense to change readiness_check to 
readiness_checks and support multiple checks
This idea has crossed my mind before, but I don't think it is necessary for the 
fixes made in this patch. I like how each of the basic checks has optional DNS 
checking. It's kind of difficult to design a feature without having a specific 
use case for it, but since the idea keeps coming up, let's open a separate 
ticket to discuss allowing multiple checks.

bq. Should we honor CONTAINER_FAILURE_THRESHOLD specified at the system level?
It's not exactly that the service/component-level property is being specified 
at the system level, it's more that it is allowing you to adjust the 
system-wide default instead of having all systems stuck with the default of 10. 
That's how I view the YarnServiceConf getters; they use hardcoded defaults 
unless different defaults are specified in the yarn-site.xml file.

bq. The component instance name is not getting logged in the fail message.
That is handled in the log line, LOG.info("Readiness check failed for {}: {}", 
instance.getCompInstanceName(), status). Although, now that you mention it, I 
should be using our convention of "[COMPONENT {}]" at the beginning of the log 
line text.

Thanks again. I'll make the rest of the changes and upload a new patch.

> Create default readiness check for service components
> -----------------------------------------------------
>                 Key: YARN-8060
>                 URL: https://issues.apache.org/jira/browse/YARN-8060
>             Project: Hadoop YARN
>          Issue Type: Sub-task
>          Components: yarn-native-services
>            Reporter: Billie Rinaldi
>            Assignee: Billie Rinaldi
>            Priority: Major
>         Attachments: YARN-8060.1.patch, YARN-8060.2.patch, YARN-8060.3.patch
> It is currently possible for a component instance to have READY status before 
> the AM retrieves an IP for the container. We should make sure the IP has been 
> retrieved before marking the instance as READY.
> This default probe could also have an option to check for a DNS entry for the 
> instance's hostname if a DNS address is provided.

This message was sent by Atlassian JIRA

To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org

Reply via email to