Rank allocation to -1 when the underlying node is not ready

[ykale]

Rank allocation to -1 when the underlying node is not ready

Typically when running the xgboost operator in autoscale mode and more nodes are required, the operator allocates a negative rank to the pods instead of waiting till some threshold time and fails. This is observed in Azure Kubernetes Service.

[issue-label-bot]

Issue-Label Bot is automatically applying the labels:

Label	Probability
kind/bug	0.85
area/front-end	0.77
area/operator	0.98

Please mark this comment with 👍 or 👎 to give our bot feedback!
Links: app homepage, dashboard and code for this bot.

[merlintang]

this is a good point, can you send a PR to fix this ?