Does the CSI driver automatically detach volumes from nodes if it goes down and schedules it on another node #360
-
|
I'm looking to build a service that might be needing more performant storage than Longhorn RWX for some usecases, so I was looking to understand if the current Hetzner-CSI driver will allow me to attach a volume to a node, let kubernetes schedule a pod on that node which for example could be a database, and if that said node goes down then the CSI driver would automatically detach the volume from the dead node and attach it to a new node and schedule the database on then ew node alongside the same volume? |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 7 replies
-
|
Yes, that's the default behaviour although there is some time delay before Kubernetes decides the node is not coming back up and therefore it attempts to reattach the volume on another node. A side note: while performance with Hetzner volumes is pretty good (around 300MB/sec sequential both read and write, and around 7500 IOPS both read and write), if you need the highest performance for databases I would recommend you use the local path provisioner so to use the local disk directly. |
Beta Was this translation helpful? Give feedback.
-
|
Local storage on cloud instance is very fast NVME storage, so that's plenty of IOPS for databases. So shared compute instances are fine. Of course if you want more predictable performance also on the compute side, you can use instances with dedicated cores. |
Beta Was this translation helpful? Give feedback.
-
|
Predictability of performance is nothing that i'm worrying about. I'm more looking to understand for example in cases where there is no replication at the application level, how can I make sure that the correct volume is then scheduled to the correct node if using local path provisioner? |
Beta Was this translation helpful? Give feedback.
-
|
If there is no replication at application level I don't recommend using the local path provisioner. |
Beta Was this translation helpful? Give feedback.
-
|
Got it. Thanks for the tips! Much appreciated :-) |
Beta Was this translation helpful? Give feedback.
-
|
np :) |
Beta Was this translation helpful? Give feedback.
Yes, that's the default behaviour although there is some time delay before Kubernetes decides the node is not coming back up and therefore it attempts to reattach the volume on another node.
A side note: while performance with Hetzner volumes is pretty good (around 300MB/sec sequential both read and write, and around 7500 IOPS both read and write), if you need the highest performance for databases I would recommend you use the local path provisioner so to use the local disk directly.