Join GitHub today
GitHub is home to over 28 million developers working together to host and review code, manage projects, and build software together.Sign up
DCP Feed does not seed from all nodes #3756
In non-import mode, when we start the DCP feed, we first do a stats call to each node in the CB cluster to get
We then process this here:
Of particular concern is the
The results from a single node only contain sequence numbers for vBuckets present on that node. It does include both Active and Replica vBuckets, so we actually get
This still leaves us with some of the vBucket sequence numbers unseeded as they were only provided from the skipped nodes, meaning that we start those streams from 0. This causes a massive backlog for Sync Gateway to work through (and cache). The impact goes up as the number of nodes in the cluster increases, as the one node we're seeding from has a smaller fraction of vBuckets.