XDCR exactly half of the documents are not being replicated

Hi there,

I encountered a problem with bidirectional XDCR - 1 of 3 buckets are not replicated fully - I mean exactly 50% of all documents stored in one bucket are ignored during replication, when other 50% are replicated. Other buckets are not affected.

How do I know? See the screenshots below.
Cluster 1:

Cluster 2:

And more screenshots:
Sets from Cluster 1:

incoming XDCR ops from Cluster 2:

You can notice the pattern - Cluster 2 has exacly 50% of what Cluster 1 has.

Cluster 1 is the only one point of input so there is no chance for Cluster 2 to create a new document inside this bucket. The flow is simple Cluster 1 -> Cluster 2 (bidirectional is just in case…).

Do you need some details? How can I debug it? Which logs to check?
I guess restarting both clusters (in this case - instances) will solve the problem but I think it’s a bug on Couchbase side).

BTW I’m using Couchbase Server 4.5.1 CE.

Best, Maciej

Hi, I solved the problem by changing bucket configuration (switched from Value Ejection to Full Ejection) on Cluster 2 but I guess it’s because memory cleaning or sth… Reboot would do the same job I think.