Kernels get stuck in "starting" execution state for weeks / months.
Typically, we end up manually deleting these apps in the Sagemaker interface when we notice they've been running for weeks/months without use.
From our experience, this seems to happen with the GPU instances (i.e ml.g4dn.xlarge) but it's not a common event.
Aside from billing concerns, it causes confusion around if we end up with many Kernels stuck in "Starting" and Apps running. This is not too common but we end up accumulating them over time.
There may be a different solution here but it seems that it would be reasonable to check on "starting" execution state in addition to "idle"
This issue looks to be related to:
#14
Kernels get stuck in "starting" execution state for weeks / months.
Typically, we end up manually deleting these apps in the Sagemaker interface when we notice they've been running for weeks/months without use.
From our experience, this seems to happen with the GPU instances (i.e ml.g4dn.xlarge) but it's not a common event.
Aside from billing concerns, it causes confusion around if we end up with many Kernels stuck in "Starting" and Apps running. This is not too common but we end up accumulating them over time.
There may be a different solution here but it seems that it would be reasonable to check on "starting" execution state in addition to "idle"
This issue looks to be related to:
#14