A `_parent` field can only point to a type that doesn't exist yet.
The validation that checks this relied on have all mappings in the
MapperService. The issue is that this check is performed on the
elected master node and it may not have the IndexService at all
to perform this check. In that case it creates a temporary IndexService and
MapperService to perform mapping validation, but only the mappings
that are part of the put index call are created, not the already existing mappings.
Because of that the `_parent` field validation can't be performed.
By changing the validation to rely on the cluster state's IndexMetaData instead
we can get around the issue with the IndexService/MapperService on the elected
master node. The IndexMetaData always holds the MappingMetaData instances
for all types.
When the dfs phase runs a SearchContext is created on each node that has a shard
for this query. When the query phase (or query and fetch phase) failed that SearchContext
was released only if the query was actually executed on the node. If for example
the query was rejected because the thread pool queue was full then the search context
was not released.
This commit adds a dedicated call for releasing the SearchContext in this case.
In addition, we must set the docIdsToLoad to null in case the fetch phase failed, otherwise
search contexts might not be released in releaseIrrelevantSearchContexts.
closes#11400
Currently all settings that were specified during repository creation are displayed. This commit enables plugins such as cloud-aws to filter out sensitive settings from the repository.
Closes#11265
When using async fetch, we can end up with cluster updates and reroutes based on teh number of shards. While not disastrous we can optimize it, since a single reroute is enough to apply to all the async fetch results that arrived during that time.
* Cut the `has_child` and `has_parent` queries over to use Lucene's query time global ordinal join. The main benefit of this change is that parent/child queries can now efficiently execute if parent/child queries are wrapped in a bigger boolean query. If the rest of the query only hit a few documents both has_child and has_parent queries don't need to evaluate all parent or child documents any more.
* Cut the `_parent` field over to use doc values. This significantly reduces the on heap memory footprint of parent/child, because the parent id values are never loaded into memory.
Breaking changes:
* The `type` option on the `_parent` field can only point to a parent type that doesn't exist yet, so this means that an existing type/mapping can't become a parent type any longer.
* The `has_child` and `has_parent` queries can no longer be use in alias filters.
All these changes, improvements and breaks in compatibility only apply for indices created with ES version 2.0 or higher. For indices creates with ES <= 2.0 the older implementation is used.
It is highly recommended to re-index all your indices with parent and child documents to benefit from all the improvements that come with this refactoring. The easiest way to achieve this is by using the scan and bulk apis using a simple script.
Closes#6107Closes#8134
This is because commit 35a58d874e causes the following REST tests to fail and reverting the commit causes conflicts:
update/15_script/Script
script/10_basic/Indexed script
This change unifies the way scripts and templates are specified for all instances in the codebase. It builds on the Script class added previously and adds request building and parsing support as well as the ability to transfer script objects between nodes. It also adds a Template class which aims to provide the same functionality for template APIs
Closes#11091
Mappers are currently used at both index and query time for deciding
how to "use" a field. For #8871, we need the index wide view of
mappings to have a unified set of settings for each field of a given
name within the index.
This change moves all the current settings (and methods defining
query time behavior) into subclasses of FieldType. In a future
PR, this will allow storing the field type at the index level,
instead of mappers (which can still have settings that differ
per document type).
The change is quite large (I'm sorry). I could not see a way to
migrate to this in a more piecemeal way. I did leave out cutting
over callers of the query methods to using the field type, as
that can be done in a follow up.
LZF only stays for backward-compatibility reasons and can only read, not write.
DEFLATE is configured to use level=3, which is a nice trade-off between speed
and compression ratio and is the same as we use for Lucene's high compression
codec.
IndexRequestBuilder#setSource as well as CreateIndexRequestBuilder#setSettings and
CreateIndexRequestBuilder#setSouce() will not work with Map<String, String> argument
although the API looks like it should. This PR fixes the problem introducing correct
wildcard parameters and adds tests.
Closes#10825
We have a lot of module classes that don't contain any actual logic,
only declarative bind actions. These classes are unnecessary and can
be consolidated into the already existings IndexShardModule
We have a compression framework that we use internally, mainly to compress some
xcontent bytes. However it is quite lenient: for instance it relies on the
assumption that detection of the compression format can only be called on either
some compressed xcontent bytes or some raw xcontent bytes, but nothing checks
this. By the way, we are misusing it in BinaryFieldMapper so that if someone
indexes a binary field which happens to have the same header as a LZF stream,
then at read time, we will try to decompress it.
It also simplifies the API by removing block compression (only streaming) and
some code duplication caused by some methods accepting a byte[] and other
methods a BytesReference.
We recently run into two issues where mapping weren't serialized in a consistent manner (#10302 and #10318). We rely on this consistency to do a byte level checl that mappings we get from the master are indentical to the one we have locally. Mistakes here can cause endless refresh mapping loops.
This commit adds an assert that verifies this upon every update from the master.
Setting the number of replicas on a closed index can leave the index
in an unopenable state since we might not be able to recover a quorum.
This commit simply prevents updating this setting on a closed index.
Closes#9566
The ShardFilterCache relies on the fact that it's
closed once the last reader on the shard is closed.
This is only guaranteed once the Store and all its
references are closed. This commit moves the closing
into the internal callback mechanism we use for deleting
shard data etc. to close the cache once we have all
searchers released.
Martijn van Groningen
Britta Weber
Adrien Grand
Michael McCandless
Simon Willnauer
Robert Muir
Igor Motov
Shay Banon
Clinton Gormley
Colin Goodheart-Smithe
Areek Zillur
Colin Goodheart-Smithe
Ryan Ernst
Christoph Büscher
Spyros Kapnissis
Simon Willnauer
Boaz Leskes
Alexander Reelsen
Lee Hinman