COUCHDB-1557: Upgrade some code to use BIFs bring good improvements for
replication.
COUCHDB-1363: Fix rarely occurred, but still race condition in changes feed
if a quick burst of changes happens while replication is starting the
replication can go stale. #573a7bb9
COUCHDB-1389: Improved tracebacks printed by the JS CLI tests.
COUCHDB-1339: Use shell trap to catch dying beam processes during test runs.
#2921c78
COUCHDB-1338: Start CouchDB with port=0. While CouchDB might be already
running on the default port 5984, port number 0 let the TCP stack figure out a
free port to run. #127cbe3
OAuth secrets can now be stored in the users system database
as an alternative to key value pairs in the .ini configuration.
By default this is disabled (secrets are stored in the .ini)
but can be enabled via the .ini configuration key use_users_db
in the couch_httpd_oauth section.
Documents in the _users database are no longer publicly
readable.
Confidential information in the _replication database is no
longer publicly readable.
Password hashes are now calculated by CouchDB. Clients are no
longer required to do this manually.
Cookies used for authentication can be made persistent by enabling
the .ini configuration key allow_persistent_cookies in the
couch_httpd_auth section.
cURL is no longer required to build CouchDB as it is only
used by the command line JS test runner. If cURL is available
when building CouchJS you can enable the HTTP bindings by
passing -H on the command line.
Temporarily made make check pass with R15B. A more thorough
fix is in the works (COUCHDB-1424).
The _active_tasks API now offers more granular fields. Each
task type is now able to expose different properties.
Added built-in changes feed filter _view.
Fixes to the _changes feed heartbeat option which caused
heartbeats to be missed when used with a filter. This caused
timeouts of continuous pull replications with a filter.
Properly restart the SSL socket on configuration changes.
A new replicator implementation. It offers more performance and
configuration options.
Passing non-string values to query_params is now a 400 bad
request. This is to reduce the surprise that all parameters
are converted to strings internally.
Added optional field since_seq to replication objects/documents.
It allows to bootstrap a replication from a specific source sequence
number.
Simpler replication cancellation. In addition to the current method,
replications can now be canceled by specifying the replication ID
instead of the original replication object/document.
Added optional database and view index file compression (using Google’s
snappy or zlib’s deflate). This feature is enabled by default, but it
can be disabled by adapting local.ini accordingly. The on-disk format
is upgraded on compaction and new DB/view creation to support this.
Several performance improvements, most notably regarding database writes
and view indexing.
Computation of the size of the latest MVCC snapshot data and all its
supporting metadata, both for database and view index files. This
information is exposed as the data_size attribute in the database and
view group information URIs.
The size of the buffers used for database and view compaction is now
configurable.
Added support for automatic database and view compaction. This feature
is disabled by default, but it can be enabled via the .ini configuration.
Performance improvements for the built-in changes feed filters _doc_ids
and `_design’.
Fix leaking file handles after compacting databases and views.
Fix databases forgetting their validation function after compaction.
Fix occasional timeout errors after successfully compacting large databases.
Fix ocassional error when writing to a database that has just been compacted.
Fix occasional timeout errors on systems with slow or heavily loaded IO.
Fix for OOME when compactions include documents with many conflicts.
Fix for missing attachment compression when MIME types included parameters.
Preserve purge metadata during compaction to avoid spurious view rebuilds.
Fix spurious conflicts introduced when uploading an attachment after
a doc has been in a conflict. See COUCHDB-902 for details.
Fix for frequently edited documents in multi-master deployments being
duplicated in _changes and _all_docs. See COUCHDB-968 for details on how
to repair.
Significantly higher read and write throughput against database and
view index files.
Don’t trigger view updates when requesting _design/doc/_info.
Fix for circular references in CommonJS requires.
Made isArray() function available to functions executed in the query server.
Documents are now sealed before being passed to map functions.
Force view compaction failure when duplicated document data exists. When
this error is seen in the logs users should rebuild their views from
scratch to fix the issue. See COUCHDB-999 for details.
JSON strings are now displayed as-is in the document view, without the
escaping of new-lines and quotes. That dramatically improves readability of
multi-line strings.
Same goes for editing of JSON string values. When a change to a field value is
submitted, and the value is not valid JSON it is assumed to be a string. This
improves editing of multi-line strings a lot.
Hitting tab in textareas no longer moves focus to the next form field, but
simply inserts a tab character at the current caret position.
Added optional ‘raw’ binary collation for faster view builds where Unicode
collation is not important.
Improved view index build time by reducing ICU collation callouts.
Improved view information objects.
Bug fix for partial updates during view builds.
Move query server to a design-doc based protocol.
Use json2.js for JSON serialization for compatiblity with native JSON.
Major refactoring of couchjs to lay the groundwork for disabling cURL
support. The new HTTP interaction acts like a synchronous XHR. Example usage
of the new system is in the JavaScript CLI test runner.
Fix for replication problems where the write queues can get backed up if the
writes aren’t happening fast enough to keep up with the reads. For a large
replication, this can exhaust memory and crash, or slow down the machine
dramatically. The fix keeps only one document in the write queue at a time.
Fix for databases sometimes incorrectly reporting that they contain 0
documents after compaction.
CouchDB now uses ibrowse instead of inets for its internal HTTP client
implementation. This means better replication stability.
Fix for chunked responses where chunks were always being split into multiple
TCP packets, which caused problems with the test suite under Safari, and in
some other cases.
Fix for an invalid JSON response body being returned for some kinds of
views. (COUCHDB-84)
Fix for connections not getting closed after rejecting a chunked request.
(COUCHDB-55)
CouchDB can now be bound to IPv6 addresses.
The HTTP Server header now contains the versions of CouchDB and Erlang.
The view engine has been completely decoupled from the storage engine. Index
data is now stored in separate files, and the format of the main database
file has changed.
Databases can now be compacted to reclaim space used for deleted documents
and old document revisions.
Support for incremental map/reduce views has been added.
To support map/reduce, the structure of design documents has changed. View
values are now JSON objects containing at least a map member, and
optionally a reduce member.
View servers are now identified by name (for example javascript) instead of
by media type.
Automatically generated document IDs are now based on proper UUID generation
using the crypto module.
The field content-type in the JSON representation of attachments has been
renamed to content_type (underscore).
When adding a field to a document, Futon now just adds a field with an
autogenerated name instead of prompting for the name with a dialog. The name
is automatically put into edit mode so that it can be changed immediately.
Fields are now sorted alphabetically by name when a document is displayed.
Futon can be used to create and update permanent views.
The maximum number of rows to display per page on the database page can now
be adjusted.
Futon now uses the XMLHTTPRequest API asynchronously to communicate with the
CouchDB HTTP server, so that most operations no longer block the browser.
View results sorting can now be switched between ascending and descending by
clicking on the Key column header.
Fixed a bug where documents that contained a @ character could not be
viewed. (COUCHDB-12)
The database page now provides a Compact button to trigger database
compaction. (COUCHDB-38)
Fixed portential double encoding of document IDs and other URI segments in
many instances. (COUCHDB-39)
Improved display of attachments.
The JavaScript Shell has been removed due to unresolved licensing issues.
SpiderMonkey is no longer included with CouchDB, but rather treated as a
normal external dependency. A simple C program (_couchjs) is provided that
links against an existing SpiderMonkey installation and uses the interpreter
embedding API.
View functions using the default JavaScript view server can now do logging
using the global log(message) function. Log messages are directed into the
CouchDB log at INFO level. (COUCHDB-59)
The global map(key, value) function made available to view code has been
renamed to emit(key, value).
Fixed handling of exceptions raised by view functions.
CouchDB now uses MochiWeb instead of inets for the HTTP server
implementation. Among other things, this means that the extra configuration
files needed for inets (such as couch_httpd.conf) are no longer used.
The HTTP interface now completely supports the HEAD method. (COUCHDB-3)
Improved compliance of Etag handling with the HTTP specification.
(COUCHDB-13)
Etags are no longer included in responses to document GET requests that
include query string parameters causing the JSON response to change without
the revision or the URI having changed.
The bulk document update API has changed slightly on both the request and the
response side. In addition, bulk updates are now atomic.
CouchDB now uses TCP_NODELAY to fix performance problems with persistent
connections on some platforms due to nagling.
Including a ?descending=false query string parameter in requests to views
no longer raises an error.
Requests to unknown top-level reserved URLs (anything with a leading
underscore) now return a unknown_private_path error instead of the
confusing illegal_database_name.
The Temporary view handling now expects a JSON request body, where the JSON
is an object with at least a map member, and optional reduce and
language members.
Temporary views no longer determine the view server based on the Content-Type
header of the POST request, but rather by looking for a language member
in the JSON body of the request.
The status code of responses to DELETE requests is now 200 to reflect that
that the deletion is performed synchronously.