vvraskin commented on issue #3855: Add invoker container name to health protocol
URL: 
https://github.com/apache/incubator-openwhisk/pull/3855#issuecomment-404179507
 
 
   Unique name[1] is used for dynamic kafka topic assignment while display 
name[2] is for the mapping between invoker container name and its kafka topic. 
   For the case of Kube based deployments, we use worker node ip addresses as 
[1]. This allows invoker topic stickiness for the case of worker OS reload. 
   [2] is the invoker docker container name in kube. If one deploys Invokers as 
k8s "Deployment" each node gets it container name dynamically in a form 
`invoker-j8ptk`, currently each invoker reports its status via health protocol 
in a form `invoker0`. And we cannot say which invoker topic belongs to which 
invoker container in the deployment. Thus [2] is a conventional mapping between 
invoker topic and actual container name we see in kube.
   
   In theory I can pack [2] into [1], but then each time the worker node gets 
OS reloaded, the invoker container will receive a new container name, which 
means that a new kafka topic will be created. This could potentially cause an 
issue, e.g. if Kafka topic of the invoker was not empty before the reload its 
activations will hang on that topic. 

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org


With regards,
Apache Git Services

Reply via email to