OpenSearch/docs/reference/migration/migrate_1_0.asciidoc

350 lines
10 KiB
Plaintext

[[breaking-changes]]
= Breaking changes in 1.0
[partintro]
--
This section discusses the changes that you need to be aware of when migrating
your application to Elasticsearch 1.0.
--
== System and settings
* Elasticsearch now runs in the foreground by default. There is no more `-f`
flag on the command line. Instead, to run elasticsearch as a daemon, use
the `-d` flag:
[source,sh]
---------------
./bin/elasticsearch -d
---------------
* Command line settings can now be passed without the `-Des.` prefix, for
instance:
[source,sh]
---------------
./bin/elasticsearch --node.name=search_1 --cluster.name=production
---------------
* Elasticsearch on 64 bit Linux now uses <<mmapfs,`mmapfs`>> by default. Make
sure that you set <<setup-service,`MAX_MAP_COUNT`>> to a sufficiently high
number. The RPM and Debian packages default this value to `262144`.
* The RPM and Debian packages no longer start Elasticsearch by default.
* The `cluster.routing.allocation` settings (`disable_allocation`,
`disable_new_allocation` and `disable_replica_location`) have been
<<modules-cluster,replaced by the single setting>>:
[source,yaml]
---------------
cluster.routing.allocation.enable: all|primaries|new_primaries|none
---------------
== Stats and Info APIs
The <<cluster-state,`cluster_state`>>, <<cluster-nodes-info,`nodes_info`>>,
<<cluster-nodes-stats,`nodes_stats`>> and <<indices-stats,`indices_stats`>>
APIs have all been changed to make their format more RESTful and less clumsy.
For instance, if you just want the `nodes` section of the the `cluster_state`,
instead of:
[source,sh]
---------------
GET /_cluster/state?filter_metadata&filter_routing_table&filter_blocks
---------------
you now use:
[source,sh]
---------------
GET /_cluster/state/nodes
---------------
Simliarly for the `nodes_stats` API, if you want the `transport` and `http`
metrics only, instead of:
[source,sh]
---------------
GET /_nodes/stats?clear&transport&http
---------------
you now use:
[source,sh]
---------------
GET /_nodes/stats/transport,http
---------------
See the links above for full details.
== Indices APIs
The `mapping`, `alias`, `settings`, and `warmer` index APIs are all similar
but there are subtle differences in the order of the URL and the response
body. For instance, adding a mapping and a warmer look slightly different:
[source,sh]
---------------
PUT /{index}/{type}/_mapping
PUT /{index}/_warmer/{name}
---------------
These URLs have been unified as:
[source,sh]
---------------
PUT /{indices}/_mapping/{type}
PUT /{indices}/_alias/{name}
PUT /{indices}/_warmer/{name}
GET /{indices}/_mapping/{types}
GET /{indices}/_alias/{names}
GET /{indices}/_settings/{names}
GET /{indices}/_warmer/{names}
DELETE /{indices}/_mapping/{types}
DELETE /{indices}/_alias/{names}
DELETE /{indices}/_warmer/{names}
---------------
All of the `{indices}`, `{types}` and `{names}` parameters can be replaced by:
* `_all`, `*` or blank (ie left out altogether), all of which mean ``all''
* wildcards like `test*`
* comma-separated lists: `index_1,test_*`
The only exception is `DELETE` which doesn't accept blank (missing)
parameters. If you want to delete something, you should be specific.
Similarly, the return values for `GET` have been unified with the following
rules:
* Only return values that exist. If you try to `GET` a mapping which doesn't
exist, then the result will be an empty object: `{}`. We no longer throw a
`404` if the requested mapping/warmer/alias/setting doesn't exist.
* The response format always has the index name, then the section, then the
element name, for instance:
[source,json]
---------------
{
"my_index": {
"mappings": {
"my_type": {...}
}
}
}
---------------
+
This is a breaking change for the `get_mapping` API.
In the future we will also provide plural versions to allow putting multiple mappings etc in a single request.
See <<indices-put-mapping,`put-mapping`>>, <<indices-get-mapping,`get-
mapping`>>, <<indices-get-field-mapping,`get-field-mapping`>>,
<<indices-delete-mapping,`delete-mapping`>>,
<<indices-update-settings,`update-settings`>>, <<indices-get-settings,`get-settings`>>,
<<indices-warmers,`warmers`>>, and <<indices-aliases,`aliases`>> for more details.
== Index request
Previously a document could be indexed as itself, or wrapped in an outer
object which specified the `type` name:
[source,json]
---------------
PUT /my_index/my_type/1
{
"my_type": {
... doc fields ...
}
}
---------------
This led to some ambiguity when a document also included a field with the same
name as the `type`. We no longer accept the outer `type` wrapper, but this
behaviour can be reenabled on an index-by-index basis with the setting:
`index.mapping.allow_type_wrapper`.
== Search requests
While the `search` API takes a top-level `query` parameter, the
<<search-count,`count`>>, <<docs-delete-by-query,`delete-by-query`>> and
<<search-validate,`validate-query`>> requests expected the whole body to be a
query. These have been changed to all accept a top-level `query` parameter:
[source,json]
---------------
GET /_count
{
"query": {
"match": {
"title": "Interesting stuff"
}
}
}
---------------
Also, the top-level `filter` parameter in search has been renamed to
<<search-request-post-filter,`post_filter`>>, to indicate that it should not
be used as the primary way to filter search results (use a
<<query-dsl-filtered-query,`filtered` query>> instead), but only to filter
results AFTER facets/aggregations have been calculated.
This example counts the top colors in all matching docs, but only returns docs
with color `red`:
[source,json]
---------------
GET /_search
{
"query": {
"match_all": {}
},
"aggs": {
"colors": {
"terms": { "field": "color" }
}
},
"post_filter": {
"term": {
"color": "red"
}
}
}
---------------
== Multi-fields
Multi-fields are dead! Long live multi-fields! Well, the field type
`multi_field` has been removed. Instead, any of the core field types
(excluding `object` and `nested`) now accept a `fields` parameter. It's the
same thing, but nicer. Instead of:
[source,json]
---------------
"title": {
"type": "multi_field",
"fields": {
"title": { "type": "string" },
"raw": { "type": "string", "index": "not_analyzed" }
}
}
---------------
you can now write:
[source,json]
---------------
"title": {
"type": "string";
"fields": {
"raw": { "type": "string", "index": "not_analyzed" }
}
}
---------------
Existing multi-fields will be upgraded to the new format automatically.
== Stopwords
Previously, the <<analysis-standard-analyzer,`standard`>> and
<<analysis-pattern-analyzer,`pattern`>> analyzers used the list of English stopwords
by default, which caused some hard to debug indexing issues. Now they are set to
use the empty stopwords list (ie `_none_`) instead.
== Dates without years
When dates are specified without a year, for example: `Dec 15 10:00:00` they
are treated as dates in 2000 during indexing and range searches... except for
the upper included bound `lte` where they were treated as dates in 1970! Now,
all https://github.com/elasticsearch/elasticsearch/issues/4451[dates without years]
use `1970` as the default.
== Parameters
* Geo queries used to use `miles` as the default unit. And we
http://en.wikipedia.org/wiki/Mars_Climate_Orbiter[all know what
happened at NASA] because of that decision. The new default unit is
https://github.com/elasticsearch/elasticsearch/issues/4515[`meters`].
* For all queries that support _fuzziness_, the `min_similarity`, `fuzziness`
and `edit_distance` parameters have been unified as the single parameter
`fuzziness`. See <<fuzziness>> for details of accepted values.
* The `ignore_missing` parameter has been replaced by the `expand_wildcards`,
`ignore_unavailable` and `allow_no_indices` parameters, all of which have
sensible defaults. See <<multi-index,the multi-index docs>> for more.
* An index name (or pattern) is now required for destructive operations like
deleting indices:
[source,sh]
---------------
# v0.90 - delete all indices:
DELETE /
# v1.0 - delete all indices:
DELETE /_all
DELETE /*
---------------
+
Setting `action.destructive_requires_name` to `true` provides further safety
by disabling wildcard expansion on destructive actions.
== Return values
* The `ok` return value has been removed from all response bodies as it added
no useful information.
* The `found`, `not_found` and `exists` return values have been unified as
`found` on all relevant APIs.
* Field values, in response to the <<search-request-fields,`fields`>>
parameter, are now always returned as arrays. A field could have single or
multiple values, which meant that sometimes they were returned as scalars
and sometimes as arrays. By always returning arrays, this simplifies user
code. The only exception to this rule is when `fields` is used to retrieve
metadata like the `routing` value, which are always singular. Metadata
fields are always returned as scalars.
* Settings, like `index.analysis.analyzer.default` are now returned as proper
nested JSON objects, which makes them easier to work with programatically:
[source,json]
---------------
{
"index": {
"analysis": {
"analyzer": {
"default": xxx
}
}
}
}
---------------
+
You can choose to return them in flattened format by passing `?flat_settings`
in the query string.
* The <<indices-analyze,`analyze`>> API no longer supports the text response
format, but does support JSON and YAML.
== Deprecations
* The `text` query has been removed. Use the
<<query-dsl-match-query,`match`>> query instead.
* The `field` query has been removed. Use the
<<query-dsl-query-string-query,`query_string`>> query instead.
* Per-document boosting with the <<mapping-boost-field,`_boost`>> field has
been removed. You can use the
<<function-score-instead-of-boost,`function_score`>> instead.