This repository has been archived by the owner on Apr 27, 2022. It is now read-only.
-
Notifications
You must be signed in to change notification settings - Fork 16
Issues: radanalyticsio/oshinko-cli
Author
Label
Projects
Milestones
Assignee
Sort
Issues list
[prometheus monitoring ] expose metrics port for pod workers.
bug
#102
opened Feb 23, 2018 by
zak-hassan
Consider making cluster name optional in the API and generating a name
enhancement
#93
opened Nov 3, 2017 by
elmiko
Prevent scaling up of existing cluster with a different spark image
bug
question
#92
opened Nov 3, 2017 by
elmiko
If master pod is destroyed and recreated, it takes ages for worker pods to timeout
#84
opened Oct 4, 2017 by
bleggett
Allow User to set memory limits that a single container in a pod can request
enhancement
#83
opened Oct 2, 2017 by
dfeddema
Assign Spark Workers Pods and Spark Master Pods to Specific Nodes
#60
opened Jul 7, 2017 by
dfeddema
ProTip!
Type g i on any issue or pull request to go back to the issue listing page.