16.3. 2.0.x Branch¶
- Native clustering is now supported. Rather than use CouchDB replication
between multiple, distinct CouchDB servers, configure a cluster of CouchDB
nodes. These nodes will use an optimized Erlang-driven ‘internal replication’
to ensure data durability and accessibility. Combine a clustered CouchDB with
a load balancer (such as
haproxy) to scale CouchDB out horizontally. More details of the clustering feature are available in the Cluster Reference.
- Futon replaced by brand-new, completely re-engineered Fauxton interface. URL remains the same.
- Multiple queries can be made of a view with a single HTTP request.
- Views can be queried with sorting turned off (
sorted=false) for a performance boost.
- The global changes feed has been enhanced. It is now resumable and persistent.
- New endpoints added (documentation forthcoming):
- /_membership shows all nodes in a cluster
/_bulk_getspeeds up the replication protocol over low-latency connections
/_node/api to access individual nodes’ configuration and compaction features
/_cluster_setupapi to set up a cluster from scratch.
/_upapi to signal health of a node to a load-balancer
/_logendpoint was removed.
- “Backend” interface on port 5986 used for specific cluster admin tasks. Of
interest are the
_dbsdatabases visible only through this interface.
- Support added for Erlang/OTP 17.x, 18.x and 19
- New streamlined build system written for Unix-like systems and Microsoft Windows
- Configuration has moved from
instance_start_timenow always reports
- The update sequences returned by the /db/_changes feed are no longer integers. They can be any JSON value. Applications should treat them as opaque values and return them to CouchDB as-is.
- Temporary views are no longer supported.
- It is possible to have multiple replicator databases.
replicator/dbconfig option has been removed. Instead
_replicatorand any database names ending with the
/_replicatorsuffix will be recognized as replicator databases by the system.
- Note that the semantics of some API calls have changed due to the introduction
of the clustering feature. Specifically, make note of the difference between
202when storing a document.
all_or_nothingis no longer supported by the bulk_docs API
- After updating a design document containing a
show, an immediate GET to that same
showfunction may still return results from the previous definition. This is due to design document caching, which may take a few seconds to fully evict, or longer (up to ~30s) for a clustered installation.
All known issues filed against the 2.0 release are contained within the official CouchDB JIRA instance or CouchDB GitHub Issues.
The following are some highlights of known issues for which fixes did not land in time for the 2.0.0 release:
COUCHDB-2980: The replicator (whether invoked via
_replicateor a document stored in the
_replicatordatabase) understands two kinds of source and target:
- A URL (e.g.,
https://foo:firstname.lastname@example.org/db1), called a “remote” source or target
- A database name (e.g.,
db1), called a “local” source or target.
Whenever the latter type is used, this refers to a local unclustered database, not a clustered one.
In a future release we hope to support “local” source or target specs to clustered databases. For now, we recommend always using the URL format for both source and target specifications.
- A URL (e.g.,
COUCHDB-3034: CouchDB will occasionally return 500 errors when multiple clients attempt to PUT or DELETE the same database concurrently.
COUCHDB-3119: Adding nodes to a cluster fails if the Erlang node name is not
couchdb(of the form
COUCHDB-3050: Occasionally the
dev/runscript used for development purposes to start a local 3-node cluster will fail to start one or more nodes.
COUCHDB-2817: The compaction daemon will only compact views for shards that contain the design document.
COUCHDB-2804: The fast_view optimization is not enabled on the clustered interface.
#656: The OAuth 1.0 support is broken and deprecated. It will be removed in a future version of CouchDB.