Release notes (beta)

This release note is for the 3.0.0 BETA REFRESH release. It contains a list of known and fixed issues.

Couchbase .NET incompatibility

An incompatibility between Microsoft Framework 4.0 or earlier, the Couchbase .NET SDK 1.3.7 or earlier, and Couchbase Server 3.0 has been identified where the URL generated for a View operation on a Bucket is improperly encoded. This causes view requests to fail with the following message in the body:

"Design document not found, body: {"error":"not_found","reason":"missing"}"
The affected Microsoft .NET versions are 4.0 or earlier, Couchbase .NET SDK 1.3.7 or earlier, and Couchbase Server 3.0 (including Beta) >= 3.0. Microsoft .NET version 4.5 changes the way Uri.ToString handles it’s Unicode encoding, which resolves the issue, thus clients running on that version or higher of the CLR are not impacted. Additionally, Couchbase Server 2.5 and lower are not impacted.
Note: Couchbase .NET 1.3.8 has a patch which resolves the issue for all versions of Couchbase Server and Microsoft .NET Frameworks.
In addition to the patch released with Couchbase .NET SDK version 1.3.8, add the following element to your App.Config or Web.Config to enable support for all SDK and Server versions.
    <iriParsing enabled="true"/>

Customers who are not ready or cannot upgrade their Couchbase .NET SDK’s to 1.3.8 or greater are strongly advised to add this element to their deployments if they intend to use the Couchbase .NET SDK with Couchbase Server 3.0 or greater.


Known issues

Online upgrade from 2.0 to 3.0 on Ubuntu 12.04 does not work
Rebalance with delta recovery exited with message "{badmatch, {error,{failed_nodes,['ns_1@']}}}"
Upgrade to R16 causes 20% regression in view query latency and throughput
There is a 3-5x increase in index size during rebalance with views
Indexing progress statistic goes backwards
Beer-sample loading is stuck in a crashed state
Incoming workload suffers when XDCR enabled
Intra-replication falling behind under moderate-heavy workload

Under a moderate replication workload (20-30K sets per second), the backlog becomes too large resulting in high memory usage.

Sliding endseqno during initial index build or DCP reading from disk snapshot results in longer stale=false query latency and index startup time
UPR replica streams cannot send items from partial snapshots

Fixed issues

wbeam.smp RSS suddenly increases to 45GB after an unexpected_binary error
Backfilling the entire vBucket can starve other streams that also need to backfill
Error displayed when node did not mark to collect
IP address does not show the same in each step of cluster-wide collect info.
Shared thread pool: high CPU overhead due to OS level context switches / scheduling
warmup_min_items_threshold setting is not honored correctly in 3.0 warmup