Joshua Moody

@derekbit correct implementing our own recovery backend either based on etcd or via an api on longhorn/api server extension, seems like a good idea once we have all the other...

@derekbit great job on the investigation and testing :) I wonder though if the connection was via `cni0` before the failure, what happens to the connection after the failover? Since...

@derekbit while working on this, can you also update the underlying ganesha version :) It's a good way to get comfortable with it, there is a custom patch applied to...

@PhanLe1010 I agree if we can make the deployment toggle + user configurable during helm / rancher chart installation and it does not impact upgrade process, it would be a...

Just to elaborate on todays meeting, you can deal with this on the application side. Since if a config map changes and it's used in volume mount mode, kubernetes will...

Requires investigation first, possibly related to the open kubernetes bug around terminating pods.

@innobead @derekbit @shuo-wu this is the issue I was thinking about in regard to dereks snapshot data missmatch issue. This might potentially be related if it's indeed true that different...

Great job on the investigation and testing so far. Below are some ideas to help narrow down the root cause. - check if this is a regression by testing with...

Read this part of the documentation and see if this helps for the down node case: - https://longhorn.io/docs/1.3.0/high-availability/node-failure/ - https://longhorn.io/docs/1.3.0/references/settings/#pod-deletion-policy-when-node-is-down

@jenting If I remember right we did some modification to the replica selection process, i.e. pass the replica information. Perhabs our fallback for the replica search does not work correctly?...