-
Notifications
You must be signed in to change notification settings - Fork 8
Release Notes
Major changes since version 1.6.1 include:
-
Allow creation and deletion of databases when a node is down.
-
Wait for read repair before responding to the client. See commit 0b59b5 for more details.
-
Modify get_missing_revs to return possible ancestor revisions. Needed to support the _revs_diff endpoint (and the replicator in the upcoming 1.2 release of Apache CouchDB, which uses that endpoint).
-
Modify the clustered update sequence so that it sorts correctly. This improves replication checkpointing.
-
Improve detection of errors during request processing by uniformly processing rexi_DOWN and rexi_EXIT errors.
-
Reply 'accepted' instead of 'ok' if a document is written but we fail to meet the requested quorum. The HTTP layer will translate this into a 202 status code in the case of a single-document PUT request.
-
Fix a few classes of timeouts and hangs when subscribing to continuous _changes feeds, especially when filters are applied.
-
Use a simple majority as the default quorum.
-
Handle missing docs when using "linked documents", and support requests to link against a specific revision.
-
Support stale=update_after in view queries.
- Send 100 Continue when appropriate
- Fail gracefully on _security if database is missing
- Fix limit=0 timing bug dropping total/offset from response
- Read from same shards every time on stale=ok view request
- Include some fields that were missing in view group _info
Bugs Squashed
- Ignore attachment offsets when deciding whether to run read repair
- Send group=true result when group_level is set but key is not a list
- Cleanly terminate view responses on timeout
- Fix behavior of _changes with include_docs and deleted documents
- Fix handling of reader lists in _security
Features Added
- Support linked documents in views
- Improve rapid DB re-creation by tagging shards with a timestamp
- Improve performance under constant concurrent load
- Improve formatting of and increase detail level in error messages
- Allow keylists instead of records in view query API
- Added type specs to API functions
- Fix quorum calculation when reading docs with attachments
- Fix heartbeat setting for _changes
- Add support for view index cleanup
- Add support for filtered _changes
- Asynchronously repair inconsistent copies of a document on read
- Fix race condition causing some reduce views to hang
- Fix timeout when calling update_docs with an empty list
- Display full cluster update sequence in DB info