Issue about Full Ejection of Cache Metadata


For some reason, I have to change the settings of “Cache Metadata” for a bucket from “Value Ejection” to “Full Ejection”. After I made such change, all the nodes are in the ‘pending warm up’ mode. It has been over 12 hours. The RAM usage and disk usage hasn’t been changing.

I assume under ‘full ejection’, the in-memory index and meta data will be transferred to disk.However, it doesn’t seem to do so. The whole cluster seems to be at ‘pending’ state.

Does anyone have the similar experience?
How can I tell the progress of the change I made, and how long shall I expect the completion of the change?

Any suggestions would be appreciated!


As per the Change Bucket Settings documentation, changing the eviction policy requires the cluster to go through warmup:

Important: Changing the ejection policy will invoke a bucket restart, resulting in temporary inaccessibility of data while the bucket warms up.

You can use the cbstats warmup command to monitor warmup progress. See the cbstats documentation for details.

Thank you @drigby!

I ran the cbstats on all the nodes. Only 2 return results and both indicated that ep_warmup_state is done.

May I ask how many nodes should return the stats? I guess all the nodes shall have results, right?

What could be the problem if some nodes don’t respond to the warm up? Does it mean the node has trouble initialize the warmup? And that cause the cluster inaccessible?