Testing on juicefs+elasticsearch, some shards will be in the unassigned state

[yanming-zhang]

When testing juicefs+elasticsearch, I encountered a problem, not sure if it is a bug?
It is to write to a certain extent (about 5G or more), some shards will be in the unassigned state

[yanming-zhang]

The version I use is v1.0.0-beta2

[davies]

@yanming-zhang Have you found something related to it in the logging of Elasticsearch or JuiceFS?

[yanming-zhang]

@yanming-zhang Have you found something related to it in the logging of Elasticsearch or JuiceFS?

need to set some parameters, then work normally
attr-cache=0
entry-cache=0
dir-entry-cache=0

[davies]

@yanming-zhang We have hit the issue before, and setting attr-cache=0.1 may be enough. It could be caused by a bug in kernel, have not tested it in recent kernel yet.