On clusters with a large number of shards, the shards limits allocation
decider can exhibit poor performance leading to timeouts applying
cluster state updates. This occurs because for every shard, we do a loop
to count the number of shards on the node, and the number of shards for
the index of the shard. This is roughly quadratic in the number of
shards. This loop is not necessary, since we already have a O(1) method
to count the number of non-relocating shards on a node, and with this
commit we add some infrastructure to RoutingNode to make counting the
number of shards per index O(1).
* Adds per context settings:
`script.context.${CONTEXT}.cache_max_size` ~
`script.cache.max_size`
`script.context.${CONTEXT}.cache_expire` ~
`script.cache.expire`
`script.context.${CONTEXT}.max_compilations_rate` ~
`script.max_compilations_rate`
* Context cache is used if:
`script.max_compilations_rate=use-context`. This
value is dynamically updatable, so users can
switch back to the general cache if desired.
* Settings for context caches take the first value
that applies:
1) Context specific settings if set, eg
`script.context.ingest.cache_max_size`
2) Correlated general setting is set to the non-default
value, eg `script.cache.max_size`
3) Context default
The reason for 2's inclusion is to allow an easy
transition for users who've customized their general
cache settings.
Using the general cache settings for the context caches
results in higher effective settings, since they are
multiplied across the number of contexts. So a general
cache max size of 200 will become 200 * # of contexts.
However, this behavior it will avoid users snapping to a
value that is too low for them.
Backport of: #52855
Refs: #50152
This commit, built on top of #51708, allows to modify shard search requests based on informations collected on other shards. It is intended to speed up sorted queries on time-based indices. For queries that are only interested in the top documents.
This change will rewrite the shard queries to match none if the bottom sort value computed in prior shards is better than all values in the shard.
For queries that mix top documents and aggregations this change will reset the size of the top documents to 0 instead of rewriting to match none.
This means that we don't need to keep a search context open for this shard since we know in advance that it doesn't contain any competitive hit.
Sometimes we want to deprecate and remove a ParseField entirely, without replacement;
for example, the various places where we specify a _type field in 7x. Currently we can
tell users only that a particular field name should not be used, and that another name should
be used in its place. This commit adds the ability to say that a field should not be used at
all.
The introduction of the ExitableDirectoryReader showed increase of
latencies for range queries using pointvalues.
Check for cancellation every 1024 docs instead of every 15 to lower
the impact of the check in query's performance.
Follows: #52822Fixes: #53496
(cherry picked from commit 6b5fc35e4458e60a7ca5822584ec6a60562f2c01)
It is useful to be able to delay state recovery until enough data nodes have
joined the cluster, since this gives the shard allocator a decent opportunity
to re-use as much existing data as possible. However we also have the option to
delay state recovery until a certain number of master-eligible nodes have
joined, and this is unnecessary: we require a majority of master-eligible nodes
for state recovery, and there is no advantage in waiting for more.
This commit deprecates the unnecessary settings in preparation for their
removal.
Relates #51806
* Add REST API for ComponentTemplate CRUD
This adds the Put/Get/DeleteComponentTemplate APIs that allow inserting, retrieving, and removing
ComponentTemplateMetadata into the cluster state metadata.
These APIs are currently only available behind a feature flag system property -
`es.itv2_feature_flag_registered`.
Relates to #53101
Co-authored-by: Elastic Machine <elasticmachine@users.noreply.github.com>
Co-authored-by: Elastic Machine <elasticmachine@users.noreply.github.com>
Re-applies the change from #53523 along with test fixes.
closes#53626closes#53624closes#53622closes#53625
Co-authored-by: Nik Everett <nik9000@gmail.com>
Co-authored-by: Lee Hinman <dakrone@users.noreply.github.com>
Co-authored-by: Jake Landis <jake.landis@elastic.co>
Today it can happen that a transport message fails to send (for example,
because a transport interceptor rejects the request). In this case, the
response handler is never invoked, which can lead to necessary cleanups
not being performed. There are two ways to handle this. One is to expect
every callsite that sends a message to try/catch these exceptions and
handle them appropriately. The other is merely to invoke the response
handler to handle the exception, which is already equipped to handle
transport exceptions.
* Submit async search to work only with POST (#53368)
Currently the submit async search API can be called using both GET and POST at REST, but given that it submits a call and creates internal state, POST should be the only allowed method.
* Refine SearchProgressListener internal API (#53373)
The following cumulative improvements have been made:
- rename `onReduce` and `notifyReduce` to `onFinalReduce` and `notifyFinalReduce`
- add unit test for `SearchShard`
- on* methods in `SearchProgressListener` shouldn't need to be public as they should never be called directly, they only need to be overridden hence they can be made protected. They are actually called directly from a test which required some adapting, like making `AsyncSearchTask.Listener` class package private instead of private
- Instead of overriding `getProgressListener` in `AsyncSearchTask`, as it feels weird to override a getter method, added a specific method that allows to retrieve the Listener directly without needing to cast it. Made the getter and setter for the listener final in the base class.
- rename `SearchProgressListener#searchShards` methods to `buildSearchShards` and make it static given that it accesses no instance members
- make `SearchShard` and `SearchShardTask` classes final
* Move async search yaml tests to x-pack yaml test folder (#53537)
The yaml tests for async search currently sit in its qa folder. There is no reason though for them to live in a separate folder as they don't require particular setup. This commit moves them to the main folder together with the other x-pack yaml tests so that they will be run by the client test runners too.
* [DOCS] Add temporary redirect for async-search (#53454)
The following API spec files contain a link to a not-yet-created
async search docs page:
* [async_search.delete.json][0]
* [async_search.get.json][1]
* [async_search.submit.json][2]
The Elaticsearch-js client uses these spec files to create their docs.
This created a broken link in the Elaticsearch-js docs, which has broken
the docs build.
This PR adds a temporary redirect for the docs page. This redirect
should be removed when the actual API docs are added.
[0]: https://github.com/elastic/elasticsearch/blob/master/x-pack/plugin/src/test/resources/rest-api-spec/api/async_search.delete.json
[1]: https://github.com/elastic/elasticsearch/blob/master/x-pack/plugin/src/test/resources/rest-api-spec/api/async_search.get.json
[2]: https://github.com/elastic/elasticsearch/blob/master/x-pack/plugin/src/test/resources/rest-api-spec/api/async_search.submit.json
Co-authored-by: James Rodewig <james.rodewig@elastic.co>
I broke sorting aggregations by `doc_count` in #51271 by mixing up true
and false. This flips that comparison and adds a few tests to double
check that we don't so this again.
This begins to clean up how `PipelineAggregator`s and executed.
Previously, we would create the `PipelineAggregator`s on the data nodes
and embed them in the aggregation tree. When it came time to execute the
pipeline aggregation we'd use the `PipelineAggregator`s that were on the
first shard's results. This is inefficient because:
1. The data node needs to make the `PipelineAggregator` only to
serialize it and then throw it away.
2. The coordinating node needs to deserialize all of the
`PipelineAggregator`s even though it only needs one of them.
3. You end up with many `PipelineAggregator` instances when you only
really *need* one per pipeline.
4. `PipelineAggregator` needs to implement serialization.
This begins to undo these by building the `PipelineAggregator`s directly
on the coordinating node and using those instead of the
`PipelineAggregator`s in the aggregtion tree. In a follow up change
we'll stop serializing the `PipelineAggregator`s to node versions that
support this behavior. And, one day, we'll be able to remove
`PipelineAggregator` from the aggregation result tree entirely.
Importantly, this doesn't change how pipeline aggregations are declared
or parsed or requested. They are still part of the `AggregationBuilder`
tree because *that* makes sense.
This commit adjusts the _cat/indices and _cat/aliases APIs to allow
specifying indices options, so that these APIs can handle hidden
indices/aliases in the same way as other APIs.
Also adds the hidden option to the expand_wildcards parameter
in the YAML spec for every API that accepts it.
* New wildcard field optimised for wildcard queries (#49993)
Indexes values using size 3 ngrams and also stores the full original as a binary doc value.
Wildcard queries operate by using a cheap approximation query on the ngram field followed up by a more expensive verification query using an automaton on the binary doc values. Also supports aggregations and sorting.
Keyword field values with length more than ignore_above are not
indexed. But highlighters still were retrieving these values
from _source and were trying to highlight them. This sometimes lead to
errors if a field length exceeded max_analyzed_offset. But also this
is an overall wrong behaviour to attempt to highlight something that was
ignored during indexing.
This PR checks if a keyword value was ignored because of its length,
and if yes, skips highlighting it.
Backport: #53408Closes#43800
This change introduces a new API in x-pack basic that allows to track the progress of a search.
Users can submit an asynchronous search through a new endpoint called `_async_search` that
works exactly the same as the `_search` endpoint but instead of blocking and returning the final response when available, it returns a response after a provided `wait_for_completion` time.
````
GET my_index_pattern*/_async_search?wait_for_completion=100ms
{
"aggs": {
"date_histogram": {
"field": "@timestamp",
"fixed_interval": "1h"
}
}
}
````
If after 100ms the final response is not available, a `partial_response` is included in the body:
````
{
"id": "9N3J1m4BgyzUDzqgC15b",
"version": 1,
"is_running": true,
"is_partial": true,
"response": {
"_shards": {
"total": 100,
"successful": 5,
"failed": 0
},
"total_hits": {
"value": 1653433,
"relation": "eq"
},
"aggs": {
...
}
}
}
````
The partial response contains the total number of requested shards, the number of shards that successfully returned and the number of shards that failed.
It also contains the total hits as well as partial aggregations computed from the successful shards.
To continue to monitor the progress of the search users can call the get `_async_search` API like the following:
````
GET _async_search/9N3J1m4BgyzUDzqgC15b/?wait_for_completion=100ms
````
That returns a new response that can contain the same partial response than the previous call if the search didn't progress, in such case the returned `version`
should be the same. If new partial results are available, the version is incremented and the `partial_response` contains the updated progress.
Finally if the response is fully available while or after waiting for completion, the `partial_response` is replaced by a `response` section that contains the usual _search response:
````
{
"id": "9N3J1m4BgyzUDzqgC15b",
"version": 10,
"is_running": false,
"response": {
"is_partial": false,
...
}
}
````
Asynchronous search are stored in a restricted index called `.async-search` if they survive (still running) after the initial submit. Each request has a keep alive that defaults to 5 days but this value can be changed/updated any time:
`````
GET my_index_pattern*/_async_search?wait_for_completion=100ms&keep_alive=10d
`````
The default can be changed when submitting the search, the example above raises the default value for the search to `10d`.
`````
GET _async_search/9N3J1m4BgyzUDzqgC15b/?wait_for_completion=100ms&keep_alive=10d
`````
The time to live for a specific search can be extended when getting the progress/result. In the example above we extend the keep alive to 10 more days.
A background service that runs only on the node that holds the first primary shard of the `async-search` index is responsible for deleting the expired results. It runs every hour but the expiration is also checked by running queries (if they take longer than the keep_alive) and when getting a result.
Like a normal `_search`, if the http channel that is used to submit a request is closed before getting a response, the search is automatically cancelled. Note that this behavior is only for the submit API, subsequent GET requests will not cancel if they are closed.
Asynchronous search are not persistent, if the coordinator node crashes or is restarted during the search, the asynchronous search will stop. To know if the search is still running or not the response contains a field called `is_running` that indicates if the task is up or not. It is the responsibility of the user to resume an asynchronous search that didn't reach a final response by re-submitting the query. However final responses and failures are persisted in a system index that allows
to retrieve a response even if the task finishes.
````
DELETE _async_search/9N3J1m4BgyzUDzqgC15b
````
The response is also not stored if the initial submit action returns a final response. This allows to not add any overhead to queries that completes within the initial `wait_for_completion`.
The `.async-search` index is a restricted index (should be migrated to a system index in +8.0) that is accessible only through the async search APIs. These APIs also ensure that only the user that submitted the initial query can retrieve or delete the running search. Note that admins/superusers would still be able to cancel the search task through the task manager like any other tasks.
Relates #49091
Co-authored-by: Luca Cavanna <javanna@users.noreply.github.com>
Today the NodeConnectionsService emits a DEBUG-level log message each time it
calls TransportService#connectToNode, which happens for every node in the
cluster every ten seconds, and also at every cluster state update. That's a lot
of log messages. Most of these calls are no-ops and can be ignored, but if the
call was not a no-op then it may be worth investigating further. Since the logs
do not distinguish the interesting and uninteresting cases, they are not
useful.
This commit distinguishes the two cases and pushes the noisy logging for the
common no-op case down to TRACE level, leaving only useful and actionable
information in the DEBUG-level logs.
Previously, Term Vectors API was returning empty results for
artificial documents with keyword fields. Checking only for `string()`
on `IndexableField` is not enough, since for `KeywordFieldType`
`binaryValue()` must be used instead.
Fixes#53494
(cherry picked from commit 1fc3fe3d32f41eab2101c0536751b7c47e63cc48)
* Use snake case for nodes stats/info metric names (#53446)
The REST API uses "thread_pool" as the name of the thread pool metric.
If we use this name internally when we serialize nodes stats and info
requests, we won't need to do any fancy logic to check for and switch
out "threadPool", which was the previous internal name.
This commit fixes a bug on sorted queries with a primary sort field
that uses different types in the requested indices. In this scenario
the returned min/max values to sort the shards are not comparable so
we should avoid the sorting rather than throwing an obscure exception.
* Add ComponentTemplate to MetaData (#53290)
* Add ComponentTemplate to MetaData
This adds a `ComponentTemplate` datastructure that will be used as part of #53101 (Index Templates
v2) to the `MetaData` class. Currently there are no APIs for interacting with this class, so it will
always be an empty map (other than in tests). This infrastructure will be built upon to add APIs in
a subsequent commit.
A `ComponentTemplate` is made up of a `Template`, a version, and a MetaData.Custom class. The
`Template` contains similar information to an `IndexTemplateMetaData` object— settings, mappings,
and alias configuration.
* Update minimal supported version constant
Co-authored-by: Elastic Machine <elasticmachine@users.noreply.github.com>
This changes the `top_metrics` aggregation to return metrics in their
original type. Since it only supports numerics, that means that dates,
longs, and doubles will come back as stored, with their appropriate
formatter applied.
The spirit of StreamInput/StreamOutput is that common I/O patterns should
be handled by these classes so that the persistence methods in
application classes can be kept short, which facilitates easy visual
comparison between read and write methods, and reduces risks of having
serialization issues due to mismatched implementations.
To this end, this change adds readOptionalVLong and writeOptionalVLong
methods to these classes as we have started to build up cases where
that conditional/null logic has been implemented directly in the read &
write methods.
Co-authored-by: Tim Vernum <tim.vernum@elastic.co>
If an index was created in version 6 and contain a date field with a joda-style pattern it should still be allowed to search and insert document into it.
Those created in 6 but date pattern starts with 8, should be considered as java style.
It looks like `date_nanos` fields weren't likely to work properly in
composite aggs because composites iterate field values using points and
we weren't converting the points into milliseconds. Because the doc
values were coming back in milliseconds we ended up geting very confused
and just never collecting sub-aggregations.
This fixes that by adding a method to `DateFieldMapper.Resolution` to
`parsePointAsMillis` which is similarly in name and function to
`NumberFieldMapper.NumberType`'s `parsePoint` except that it normalizes
to milliseconds which is what aggs need at the moment.
Closes#53168
Using a Long alone is not strong enough for the id of search contexts
because we reset the id generator whenever a data node is restarted.
This can lead to two issues:
1. Fetch phase can fetch documents from another index
2. A scroll search can return documents from another index
This commit avoids these issues by adding a UUID to SearchContexId.
In rare circumstances it is possible for an isolated node to have a greater
term than the currently-elected leader. Today such a node will attempt to join
the cluster but will not offer a vote to the leader and will reject its cluster
state publications due to their stale term. This situation persists since there
is no mechanism for the joining node to inform the leader that its term is
stale and a new election is required.
This commit adds the current term of the joining node to the join request. Once
the join has been validated, the leader will perform another election to
increase its term far enough to allow the isolated node to join properly.
Fixes#53271
* Record Force Merges in live commit data
Prerequisite of #52182. Record force merges in the live commit data
so two shard states with the same sequence number that differ only in whether
or not they have been force merged can be distinguished when creating snapshots.
We can't always have the same segment stats and doc stats between
InternalEngine and ReadOnlyEngine if there are some fully deleted
segments. ReadOnlyEngine always filters out them. InternalEngine,
however, will keep them if peer recovery retention leases exist or the
number of the retaining operations is non-zero.
This change reverts the fix in #51331 and uses the wrapped reader to
calculate the segment stats and doc stats. For the test, we need to
disable the extra retaining soft-deletes operations.
Closes#51303
Prior to this commit, rollover did not propagate the `is_hidden` alias
property when rollover over an index. This commit ensures that an alias
that's rollover over will remain hidden.
This removes the `instanceof`s from `SiblingPipelineAggregator` by
adding a `rewriteBuckets` method to `InternalAggregation` that can be
called to, well, rewrite the buckets. The default implementation of
`rewriteBuckets` throws the same exception that was thrown when you
attempted to run a `SiblingPipelineAggregator` on an aggregation without
buckets. It is overridden by `InternalSingleBucketAggregation` and
`InternalMultiBucketAggregation` to correctly rewrite their buckets.
When an composite aggregation is run against an index with a sort that
*starts* with the "source" fields from the composite but has additional
fields it'd blow up in while trying to decide if it could use the sort.
This changes it to decide that it *can* use the sort.
Closes#52480
This change optimizes the merge of terms aggregations by removing
the priority queue that was used to collect all the buckets during
a non-final reduction. We don't need to keep the result sorted since
the merge of buckets in a subsequent reduce can modify the order.
I wrote a small micro-benchmark to test the change and the speed ups
are significative for small merge buffer sizes:
````
########## Master:
Benchmark (bufferSize) (cardinality) (numShards) (topNSize) Mode Cnt Score Error Units
TermsReduceBenchmark.reduceTopHits 5 10000 1000 1000 avgt 10 2459,690 ± 198,682 ms/op
TermsReduceBenchmark.reduceTopHits 16 10000 1000 1000 avgt 10 1030,620 ± 91,544 ms/op
TermsReduceBenchmark.reduceTopHits 32 10000 1000 1000 avgt 10 558,608 ± 44,915 ms/op
TermsReduceBenchmark.reduceTopHits 128 10000 1000 1000 avgt 10 287,333 ± 8,342 ms/op
TermsReduceBenchmark.reduceTopHits 512 10000 1000 1000 avgt 10 257,325 ± 54,515 ms/op
########## Patch:
Benchmark (bufferSize) (cardinality) (numShards) (topNSize) Mode Cnt Score Error Units
TermsReduceBenchmark.reduceTopHits 5 10000 1000 1000 avgt 10 805,611 ± 14,630 ms/op
TermsReduceBenchmark.reduceTopHits 16 10000 1000 1000 avgt 10 378,851 ± 17,929 ms/op
TermsReduceBenchmark.reduceTopHits 32 10000 1000 1000 avgt 10 261,094 ± 10,176 ms/op
TermsReduceBenchmark.reduceTopHits 128 10000 1000 1000 avgt 10 241,051 ± 19,558 ms/op
TermsReduceBenchmark.reduceTopHits 512 10000 1000 1000 avgt 10 231,643 ± 6,170 ms/op
````
The code for the benchmark can be found [here](). It seems to be up to 3x faster for terms aggregations
that return 10,000 unique terms (1000 terms per shard). For a cardinality of 100,000 terms, this patch is up to 5x faster:
````
########## Patch:
Benchmark (bufferSize) (cardinality) (numShards) (topNSize) Mode Cnt Score Error Units
TermsReduceBenchmark.reduceTopHits 5 100000 1000 1000 avgt 10 12791,083 ± 397,128 ms/op
TermsReduceBenchmark.reduceTopHits 16 100000 1000 1000 avgt 10 3974,939 ± 324,617 ms/op
TermsReduceBenchmark.reduceTopHits 32 100000 1000 1000 avgt 10 2186,285 ± 267,124 ms/op
TermsReduceBenchmark.reduceTopHits 128 100000 1000 1000 avgt 10 914,657 ± 160,784 ms/op
TermsReduceBenchmark.reduceTopHits 512 100000 1000 1000 avgt 10 604,198 ± 145,457 ms/op
########## Master:
Benchmark (bufferSize) (cardinality) (numShards) (topNSize) Mode Cnt Score Error Units
TermsReduceBenchmark.reduceTopHits 5 100000 1000 1000 avgt 10 60696,107 ± 929,944 ms/op
TermsReduceBenchmark.reduceTopHits 16 100000 1000 1000 avgt 10 16292,894 ± 783,398 ms/op
TermsReduceBenchmark.reduceTopHits 32 100000 1000 1000 avgt 10 7705,444 ± 77,588 ms/op
TermsReduceBenchmark.reduceTopHits 128 100000 1000 1000 avgt 10 2156,685 ± 88,795 ms/op
TermsReduceBenchmark.reduceTopHits 512 100000 1000 1000 avgt 10 760,273 ± 53,738 ms/op
````
The merge of buckets can also be optimized. Currently we use an hash map to merge buckets coming from different shards so this can be costly if the number of unique terms is high. Instead, we could always sort the shard terms result by key and perform a merge sort to reduce the results. This would save memory and make the merge more linear in terms
of complexity in the coordinating node at the expense of an additional sort in the shards. I plan to test this possible optimization in a follow up.
Relates #51857
It doesn't make a whole lot of sense for `BitArray#clear` to grow the
underlying storage array just to clear the bit. We *already* treat
indices outside of the storage array as unset. This turns such
operations into a noop.
Lucene 8.5.0 release candidates are imminent. This commit upgrades master to use
the latest snapshot to check that there are no last-minute bugs or regressions.
This commit changes the Get Aliases API to include hidden indices by
default - this is slightly different from other APIs, but is necessary
to make this API work intuitively.
* Add unit tests before refactoring
* Convert boolean fields to set of strings
In order to make nodes stats plugins pluggable, we need to make the
NodesStatsRequest class capable of carrying a flexible list of metrics
rather than a fixed list of boolean flags. This commit changes the
internal storage of the class without changing its serialization.
* Change serialization of NodesStatsRequest
* Set up BWC before merging
* Singularize enum name
Under certain circumstances SpanMultiTermQueryWrapper uses
SpanBooleanQueryRewriteWithMaxClause as its rewrite method, which in turn tries
to get a TermsEnum from the wrapped MultiTermQuery currently using a `null`
AttributeSource. While queries TermsQuery or subclasses of AutomatonQuery ignore
this argument, FuzzyQuery uses it to create a FuzzyTermsEnum which triggers an
NPE when the AttributeSource is not provided. This PR fixes this by supplying an
empty AttributeSource instead of a `null` value.
Closes#52894
Today when notifying a global checkpoint listener, we use the listener
thread pool. This commit turns this inside out so that the global
checkpoint listener must provide an executor on which to notify the
listener.
This commit drops the dispatching listenable action future that forks to
the listener thread pool. This was previously used in the transport
client but is no longer used.
It can be that a failure is repeated to a grouped action listener. For
example, if the same exception such as a connect transport exception, is
the cause of repeated failures. Previously we were unconditionally
self-suppressing the exception into the first exception, but
self-supressing is not allowed. Thus, we would throw an exception and
the grouped action listener would never complete. This commit addresses
this by guarding against self-suppression.
Today we notify refresh listeners by forking to the listener thread pool
and then serially notifying listeners on a thread there. Refreshes are
expensive though, so the expectation is that we are executing refreshes
on threads that can afford an expensive operation (e.g., not a network
thread) and as such, executing listeners that we expect to be cheap aon
the calling thread is okay. This commit removes the forking of notifying
refresh listeners to run directly on the calling thread that executed a
refresh.
This commit introduces hidden aliases. These are similar to hidden
indices, in that they are not visible by default, unless explicitly
specified by name or by indicating that hidden indices/aliases are
desired.
The new alias property, `is_hidden` is implemented similarly to
`is_write_index`, except that it must be consistent across all indices
with a given alias - that is, all indices with a given alias must
specify the alias as either hidden, or all specify it as non-hidden,
either explicitly or by omitting the `is_hidden` property.
Our lovely `BitArray` compactly stores "flags", lazilly growing its
underlying storage. It is super useful when you need to store one bit of
data for a zillion buckets or a documents or something. Usefully, it
defaults to `false`. But there is a wrinkle! If you ask it whether or
not a bit is set but it hasn't grown its underlying storage array
"around" that index then it'll throw an `ArrayIndexOutOfBoundsException`.
The per-document use cases tend to show up in order and don't tend to
mind this too much. But the use case in aggregations, the per-bucket use
case, does. Because buckets are collected out of order all the time.
This changes `BitArray` so it'll return `false` if the index is too big
for the underlying storage. After all, that index *can't* have been set
or else we would have grown the underlying array. Logically, I believe
this makes sense. And it makes my life easy. At the cost of three lines.
*but* this adds an extra test to every call to `get`. I think this is
likely ok because it is "very close" to an array index lookup that
already runs the same test. So I *think* it'll end up merged with the
array bounds check.
This commit updates the template used for watch history indices with
the hidden index setting so that new indices will be created as hidden.
Relates #50251
Backport of #52962
Currently the AbstractBulkByScrollRequest accepts slice values of 0 via its
`setSlices` method, denoting the "auto" slicing behaviour that is usable by
settting the "slices=auto" parameter on rest requests. When using the High Level
Rest Client, however, we send the 0 value as an integer, which is then rejected
as invalid by `AbstractBulkByScrollRequest#parseSlices`. Instead of making
parsing of the rest request more lenient, this PR opts for changing the
RequestConverter logic in the client to translate 0 values to "auto" on the rest
requests.
Closes#53044
For Node Info to be pluggable, NodesInfoRequest must be able to carry
arbitrary strings. This commit reworks the internals of that class to
use a set rather than hard-coded boolean fields.
NodesInfoRequest defaults to specifying all values. We test for
this behavior as we refactor and use random testing for the
various combinations of metrics.
Add backwards compatibility for transport requests.
With ExitableDirectoryReader in place, check for query cancellation
during QueryPhase#preProcess where the query rewriting takes place.
Follows: #52822
(cherry picked from commit 0d38626d8e6e9e2620a7a446b617a2ac42852461)
The bool query builder in elasticsearch accepts both must_not and mustNot
fields. Given that leniency is abhorrent and must be eschewed, we should deprecate
the latter as it doesn't fit with the style of parameters elsewhere in the DSL.
Use assertBusy when doing reroute after bridged disruption,
since it can return non-acked if a node is marked faulty
by follower check after disruption ended.
Closes#53064
7.5 and 7.6 had a regression that allowed for
script_score queries to have negative scores.
We have corrected this regression in #52478.
This is an addition to #52478 that adds
a test and release notes.
Implement an Exitable DirectoryReader that wraps the original
DirectoryReader so that when a search task is cancelled the
DirectoryReaders also stop their work fast. This is usuful for
expensive operations like wilcard/prefix queries where the
DirectoryReaders can spend lots of time and consume resources,
as previously their work wouldn't stop even though the original
search task was cancelled (e.g. because of timeout or dropped client
connection).
(cherry picked from commit 67acaf61f33bc5f54e26541514d07e375c202e03)
With #50871 aggrgations should now be parsed directly by an
`ObjectParser` or `ConstructingObjectParser` without the need for the
ceremonial `parse` method. This removes 9 of those `parse` methods and
parses the aggregation directly from their `ObjectParser`.
Currently the remote connection manager will delegate the size() call to
the underlying cluster connection manager. This introduces the
possibility that call will return 1 before the nodeConnection method has
been triggered to add the connection to the remote connection list. This
can cause issues, as the ensureConnected method checks the connection
managers size and executes synchronously if the size is > 0. This leads
to a potential cluster not connected exception while we are still
waiting for the connection opened callback to be triggered.
This commit fixes this issue by using the remote connection manager's
size to report the connection manager's size.
Fixes#52029.
This commit removes the hand-rolled x-content parsing logic from BoolQueryBuilder
and instead uses an ObjectParser to handle parsing. It also removes the long-deprecated
(since version 6) disable_coord parameter.
Converts the deprecations to `deprecatedAndMaybeLog` to reduce the
number of times we log deprecations, since some of these could be called
at a high frequency (due to unconverted queries, aggs, etc)
This commit introduces a module for Kibana that exposes REST APIs that
will be used by Kibana for access to its system indices. These APIs are wrapped
versions of the existing REST endpoints. A new setting is also introduced since
the Kibana system indices' names are allowed to be changed by a user in case
multiple instances of Kibana use the same instance of Elasticsearch.
Additionally, the ThreadContext has been extended to indicate that the use of
system indices may be allowed in a request. This will be built upon in the future
for the protection of system indices.
Backport of #52385
With #50871 aggrgations should now be parsed directly by an
`ObjectParser` or `ConstructingObjectParser` without the need for the
ceremonial `parse` method. This removes 10 of those `parse` methods and
parses the aggregation directly from their `ObjectParser`.
`MinAndMax` encapsulates min and max values for a shard. It uses generics to make sure that the values are of the same type and are also comparable. Though there are warnings whenever this class is currently used, which are addressed with this commit.
Relates to #49092
This field is a specialization of the `keyword` field for the case when all
documents have the same value. It typically performs more efficiently than
keywords at query time by figuring out whether all or none of the documents
match at rewrite time, like `term` queries on `_index`.
The name is up for discussion. I liked including `keyword` in it, so that we
still have room for a `singleton_numeric` in the future. However I'm unsure
whether to call it `singleton`, `constant` or something else, any opinions?
For this field there is a choice between
1. accepting values in `_source` when they are equal to the value configured
in mappings, but rejecting mapping updates
2. rejecting values in `_source` but then allowing updates to the value that
is configured in the mapping
This commit implements option 1, so that it is possible to reindex from/to an
index that has the field mapped as a keyword with no changes to the source.
Backport of #49713
When notifying global checkpoint listeners, we have an opportunity to
early return if there are not any registered listeners. This is
important since it saves some allocations, and also saves forking some
empty work to another thread. This commit adds an early return from
notifying listeners if there are not any registered.
Allow AffixSetting as validator dependencies. If a validator
specifies AffixSettings as a dependency, then `validate(T, Map)`
will have the concrete setting in a map.
Backport of: #52973, 1e0ba70
Fixes: #52933
This commit removes a TODO in the IndexNameExpressionResolver that
indicated the API should use a Set instead of a List. However, this
TODO was not completely correct since the ordering of arguments matters
due to negations when evaluating wildcards and since we also allow
a list of patterns like `*,-foo,*`, which would have a different
meaning even when using a Set with insertion ordering.
Relates #52788
Backport of #52963
Since version 8.4, `MMapDirectory` has an optimization to read long[]
arrays directly in little endian order, which postings leverage. So it'd
be more efficient to open postings with `MMapDirectory`.
I refactored a bit the existing logic to better explain why every listed
file extension is open with `mmap`.
Backport of #51233 to the seven dot x branch.
Tries to load a `Mapper` instance for the mapping snippet of a dynamic template.
This should catch things like using an analyzer that is undefined or mapping attributes that are unused.
This is best effort:
* If `{{name}}` placeholder is used in the mapping snippet then validation is skipped.
* If `match_mapping_type` is not specified then validation is performed for all mapping types.
If parsing succeeds with a single mapping type then this the dynamic mapping is considered valid.
If is detected that a dynamic template mapping snippet is invalid at mapping update time then the mapping update is failed for indices created on 8.0.0-alpha1 and later. For indices created on prior version a deprecation warning is omitted instead. In 7.x clusters the mapping update will never fail in case of an invalid dynamic template mapping snippet and a deprecation warning will always be omitted.
Closes#17411Closes#24419
Co-authored-by: Adrien Grand <jpountz@gmail.com>
The `terms` aggregation can be sortd by the results of its
sub-aggregations. Because it uses that sorting for filtering to the
top-n it tries not to construct all of the buckets for the child
aggregations. This has its own interesting problem around reduction, but
they aren't super relevant to this change. This change moves that
optimization from the `TermsAggregator` and into the aggregators being
sorted on. This should make it more clear what is going on and it
unifies this optimization with validating the sort.
Finally, this should enable some minor optimizations to save a few
comparisons when sorting multi-valued buckets. I'll get those in a
follow up because they are now *fairly* obvious. They probably won't be
a huge performance improvement, but it'll be nice anyway.
Add index name(s) into the source for the cluster state update done when putting mapping.
This ensures that the pending tasks API includes information on source indices.
Computing the stats for completion fields may involve a significant amount of
work since it walks every field of every segment looking for completion fields.
Innocuous-looking APIs like `GET _stats` or `GET _cluster/stats` do this for
every shard in the cluster. This repeated work is unnecessary since these stats
do not change between refreshes; in many indices they remain constant for a
long time.
This commit introduces a cache for these stats which is invalidated on a
refresh, allowing most stats calls to bypass the work needed to compute them on
most shards.
Closes#51915
Backport of #51991
We aren't able to reproduce or figure out the reason that failed this test.
This commit adds more assertions so we can narrow the scope.
Relates #52223
Since #51905, we use the local checkpoint of the safe commit to
calculate the number of uncommitted operations of a translog stats. If a
periodic flush triggered by afterWriteOperation completes before we sync
translog, then the last commit is not safe. We also need to sync
translog from Engine instead of the translog so that we can advance the
safe commit.
Relates #51905Closes#52223
Since #51905, we skip translog recovery if the local checkpoint of the
safe commit equals to the global checkpoint. This change adjusts the
test not to create a new snapshot in that case.
Closes#52221
Relates #51905
Separates the translog from the index deletion conditions (allowing the translog to be cleaned
up more eagerly), and avoids taking the write lock on the translog if no clean-up is actually
necessary.
Today we use the translog_generation of the safe commit as the minimum
required translog generation for recovery. This approach has a
limitation, where we won't be able to clean up translog unless we flush.
Reopening an already recovered engine will create a new empty translog,
and we leave it there until we force flush.
This commit removes the translog_generation commit tag and uses the
local checkpoint of the safe commit to calculate the minimum required
translog generation for recovery instead.
Closes#49970
This commit changes the `index.hidden` setting from being final to a
dynamic setting. While the setting being final allows for easier
reasoning about an index, making this setting update-able has more
benefits in that we can upgrade existing indices to be hidden and it
will enable future features that would dynamically make indices hidden.
Backport of #52772
We've pretty well settled on `ContextParser` for a generic interface to
`ObjectParser`-like-things. This switches the interface used for
building parsing pipeline aggregations to `ContextParser` which saves a
couple of little wrappers around `ObjectParser`.
Currently 3 remote cluster settings (ping interval, skip unavailable,
and compression) have a dependency on the seeds setting being
comfigured. With proxy mode, it is now possible that these settings the
seeds setting has not been configured. This commit removes this
dependency and adds new validation for these settings.
Generalize how queries on `_index` are handled at rewrite time (#52486)
Since this change refactors rewrites, I also took it as an opportunity to adrress #49254: instead of returning the same queries you would get on a keyword field when a field is unmapped, queries get rewritten to a MatchNoDocsQueryBuilder.
This change exposed a couple bugs, like the fact that the percolator doesn't rewrite queries at query time, or that the significant_terms aggregation doesn't rewrite its inner filter, which I fixed.
Closes#49254
Currently we have two ways to create a GroupShardsIterator: one that will resort the iterators based on their natural ordering, and another one that will leave them in their original order. This is currently done through two constructors, one that accepts a single argument which does the sorting, and another which accepts a second boolean argument to control whether sorting should happen or not. This second constructor is only called externally to disable the sorting.
By introducing a specific method to create a sorted shard iterator we clarify and make it easier to track when we do sort and when we do not as the iterators are externally sorted.
This change fixes the incomplete backport of #46731 in 7.x (as of 7.5).
We now check if `max_children` is set on the top level nested sort and fails with an
exception if it's not the case.
Relates #46731Closes#52202
* Remove TODO in MaxAgeCondition serialization
This removes the TODO with a message for any future readers regarding the code in question.
Resolves#52505
Currently the shard bulk request can be rejected by the write threadpool
after a mapping update. This introduces a scenario where the mapping
listener thread will attempt to finish the request and fsync. This
thread can potentially be a transport thread. This commit fixes this
issue by forcing the finish action to happen on the write threadpool.
Fixes#51904.
Lucene's RAMDirectory has been deprecated. This commit replaces all uses of
RAMDirectory in elasticsearch with the newer ByteBuffersDirectory. Most uses
are in tests, but the percolator and painless executor may get some small speedups.
Currently, date ranges queries using NOW-based date math are rewritten to
MatchAllDocs queries when being preprocessed for the percolator. However,
since we added the verification step, this can result in incorrect matches when
percolator queries are run without scores. This commit changes things to instead
wrap date queries that use NOW with a new DateRangeIncludingNowQuery.
This is a simple wrapper query that returns its delegate at rewrite time, but it can
be detected by the percolator QueryAnalyzer and be dealt with accordingly.
This also allows us to remove a method on QueryRewriteContext, and push all
logic relating to NOW-based ranges into the DateFieldMapper.
Fixes#52617
When the Node class is being constructed, an initial environment is
passed in with the initial settings for the node. Once the plugin
servicie is initialized, the final Environment+Settings are created, at
which point the initial environment should no longer be used. This
commit renames the constructor arg to avoid naming clashes with the
final environment variable.
Before boost in script_score query was wrongly applied only to the subquery.
This commit makes sure that the boost is applied to the whole score
that comes out of script.
Closes#48465
In #42838 we moved the terms index of all fields off-heap except the
`_id` field because we were worried it might make indexing slower. In
general, the indexing rate is only affected if explicit IDs are used, as
otherwise Elasticsearch almost never performs lookups in the terms
dictionary for the purpose of indexing. So it's quite wasteful to
require the terms index of `_id` to be loaded on-heap for users who have
append-only workloads. Furthermore I've been conducting benchmarks when
indexing with explicit ids on the http_logs dataset that suggest that
the slowdown is low enough that it's probably not worth forcing the terms
index to be kept on-heap. Here are some numbers for the median indexing
rate in docs/s:
| Run | Master | Patch |
| --- | ------- | ------- |
| 1 | 45851.2 | 46401.4 |
| 2 | 45192.6 | 44561.0 |
| 3 | 45635.2 | 44137.0 |
| 4 | 46435.0 | 44692.8 |
| 5 | 45829.0 | 44949.0 |
And now heap usage in MB for segments:
| Run | Master | Patch |
| --- | ------- | -------- |
| 1 | 41.1720 | 0.352083 |
| 2 | 45.1545 | 0.382534 |
| 3 | 41.7746 | 0.381285 |
| 4 | 45.3673 | 0.412737 |
| 5 | 45.4616 | 0.375063 |
Indexing rate decreased by 1.8% on average, while memory usage decreased
by more than 100x.
The `http_logs` dataset contains small documents and has a simple
indexing chain. More complex indexing chains, e.g. with more fields,
ingest pipelines, etc. would see an even lower decrease of indexing rate.
This drops more of the `instanceof`s from `AggregationPath`. There are
still a couple in `AggregationPath`. And I ended up moving two into
`BucketsAggregator`, but I think this is still an improvement!
We consider index level read_only_allow_delete blocks temporary since
the DiskThresholdMonitor can automatically release those when an index
is no longer allocated on nodes above high threshold.
The rest status has therefore been changed to 429 when encountering this
index block to signal retryability to clients.
Related to #49393
This commit renames ElasticsearchAssertions#assertThrows to
assertRequestBuilderThrows and assertFutureThrows to avoid a
naming clash with JUnit 4.13+ and static imports of these methods.
Additionally, these methods have been updated to make use of
expectThrows internally to avoid duplicating the logic there.
Relates #51787
Backport of #52582
Phase 1 of adding compilation limits per context.
* Refactor rate limiting and caching into separate class,
`ScriptCache`, which will be used per context.
* Disable compilation limit for certain tests.
Backport of 0866031
Refs: #50152
This commit modifies the codebase so that our production code uses a
single instance of the IndexNameExpressionResolver class. This change
is being made in preparation for allowing name expression resolution
to be augmented by a plugin.
In order to remove some instances of IndexNameExpressionResolver, the
single instance is added as a parameter of Plugin#createComponents and
PersistentTaskPlugin#getPersistentTasksExecutor.
Backport of #52596
Cache latest `RepositoryData` on heap when it's absolutely safe to do so (i.e. when the repository is in strictly consistent mode).
`RepositoryData` can safely be assumed to not grow to a size that would cause trouble because we often have at least two copies of it loaded at the same time when doing repository operations. Also, concurrent snapshot API status requests currently load it independently of each other and so on, making it safe to cache on heap and assume as "small" IMO.
The benefits of this move are:
* Much faster repository status API calls
* listing all snapshot names becomes instant
* Other operations are sped up massively too because they mostly operate in two steps: load repository data then load multiple other blobs to get the additional data
* Additional cloud cost savings
* Better resiliency, saving another spot where an IO issue could break the snapshot
* We can simplify a number of spots in the current code that currently pass around the repository data in tricky ways to avoid loading it multiple times in follow ups.
* Refactor Inflexible Snapshot Repository BwC (#52365)
Transport the version to use for a snapshot instead of whether to use shard generations in the snapshots in progress entry. This allows making upcoming repository metadata changes in a flexible manner in an analogous way to how we handle serialization BwC elsewhere.
Also, exposing the version at the repository API level will make it easier to do BwC relevant changes in derived repositories like source only or encrypted.
AllocationDeciders would collect Yes decisions when not asking for debug
info. Changed to only include Yes decisions when debug is requested
(explain).
Added ability to specify comma separated list of source indices without
array. Also fixed so that empty string results in validation error
rather than index does not exist.
Closes#51949
Issue #52000 looks like a case of cluster state updates being slower than
expected, but it seems that these slowdowns are relatively rare: most
invocations of `testDelayWithALargeAmountOfShards` take well under a minute in
CI, but there are occasional failures that take 6+ minutes instead. When it
fails like this, cluster state persistence seems generally slow: most are
slower than expected, with some small updates even taking over 2 seconds to
complete.
The failures all have in common that they use `WindowsFS` to emulate Windows'
behaviour of refusing to delete files that are still open, by tracking all
files (really, inodes) and validating that deleted files are really closed
first. There is a suggestion that this is a little slow in the Lucene test
framework [1]. To see if we can attribute the slowdown to that common factor,
this commit suppresses the use of `WindowsFS` for this test suite.
[1] 4a513fa99f/lucene/test-framework/src/java/org/apache/lucene/util/TestRuleTemporaryFilesCleanup.java (L166)
Currently we lock when generating time based uuids. The lock is
implemented to prevent concurrent writes to the last timestamp. The uuid
generation is an area of contention when indexing. This commit modifies
the code to use atomic compare and set operations to update the last
timestamp.
Every time a setting#exist call is made we lock on the keyset to ensure
that it has been initialized. This a heavyweight operation that only
should be done once. This commit moves to a volatile read instead to
prevent unnecessary locking.
Currently we have three different implementations representing a
`ConnectionManager`. There is the basic `ConnectionManager` which
holds all connections for a cluster. And a remote connection manager
which support proxy behavior. And a stubbable connection manager for
tests. The remote and stubbable instances use the delegate pattern,
so this commit extracts an interface for them all to implement.
It looks like #52000 is caused by a slowdown in cluster state application
(maybe due to #50907) but I would like to understand the details to ensure that
there's nothing else going on here too before simply increasing the timeout.
This commit enables some relevant `DEBUG` loggers and also captures stack
traces from all threads rather than just the three hottest ones.
When `FilterStreamInput` wraps a Netty `ByteBuf` based stream it
did not forward the bulk primitive reads to the delegate.
These are optimized on the delegate but if they're not forwarded
then the delegate will be called e.g. 4 times to read an `int`.
This happens for essentially all network reads prior to this
change because they all run from a `NamedWritableAwareStreamInput`.
This also required optimising `BufferedChecksumStreamInput` individually to use bulk reads from the buffer because it implicitly assumed that the filter stream input wouldn't override any of the bulk operations.
The `top_metrics` agg is kind of like `top_hits` but it only works on
doc values so it *should* be faster.
At this point it is fairly limited in that it only supports a single,
numeric sort and a single, numeric metric. And it only fetches the "very
topest" document worth of metric. We plan to support returning a
configurable number of top metrics, requesting more than one metric and
more than one sort. And, eventually, non-numeric sorts and metrics. The
trick is doing those things fairly efficiently.
Co-Authored by: Zachary Tong <zach@elastic.co>
Fixes the the no-query optimization for `min` and `max` aggregations
for `date_nanos` fields by delegating decoding dates "through" their
`resolution` member.
Closes#52220
This commit makes the names of fetch subphases more consistent:
* Now the names end in just 'Phase', whereas before some ended in
'FetchSubPhase'. This matches the query subphases like AggregationPhase.
* Some names include 'fetch' like FetchScorePhase to avoid ambiguity about what
they do.
This adds a builder and parsed results for the `string_stats`
aggregation directly to the high level rest client. Without this the
HLRC can't access the `string_stats` API without the elastic licensed
`analytics` module.
While I'm in there this adds a few of our usual unit tests and
modernizes the parsing.
When `date_histogram` attempts to optimize itself it for a particular
time zone it checks to see if the entire shard is within the same
"transition". Most time zone transition once every size months or
thereabouts so the optimization can usually kicks in.
*But* it crashes when you attempt feed it a time zone who's last DST
transition was before epoch. The reason for this is a little twisted:
before this patch it'd find the next and previous transitions in
milliseconds since epoch. Then it'd cast them to `Long`s and pass them
into the `DateFieldType` to check if the shard's contents were within
the range. The trouble is they are then converted to `String`s which are
*then* parsed back to `Instant`s which are then convertd to `long`s. And
the parser doesn't like most negative numbers. And everything before
epoch is negative.
This change removes the
`long` -> `Long` -> `String` -> `Instant` -> `long` chain in favor of
passing the `long` -> `Instant` -> `long` which avoids the fairly complex
parsing code and handles a bunch of interesting edge cases around
epoch. And other edge cases around `date_nanos`.
Closes#50265
We need to reduce the translog sync interval for indices with translog
async setting so that we can have the safe commit in the assertBusy
interval. This is needed since #51905, where we use the local checkpoint
of the safe commit to calculate the number of uncommitted operations of
a translog stats.
Closes#52251
Relates #51905
This commit removes the need for DeprecatedRoute and ReplacedRoute to
have an instance of a DeprecationLogger. Instead the RestController now
has a DeprecationLogger that will be used for all deprecated and
replaced route messages.
Relates #51950
Backport of #52278
Add a new cluster setting `search.allow_expensive_queries` which by
default is `true`. If set to `false`, certain queries that have
usually slow performance cannot be executed and an error message
is returned.
- Queries that need to do linear scans to identify matches:
- Script queries
- Queries that have a high up-front cost:
- Fuzzy queries
- Regexp queries
- Prefix queries (without index_prefixes enabled
- Wildcard queries
- Range queries on text and keyword fields
- Joining queries
- HasParent queries
- HasChild queries
- ParentId queries
- Nested queries
- Queries on deprecated 6.x geo shapes (using PrefixTree implementation)
- Queries that may have a high per-document cost:
- Script score queries
- Percolate queries
Closes: #29050
(cherry picked from commit a8b39ed842c7770bd9275958c9f747502fd9a3ea)
The buffer in LoggingOutputStream skips flushing when only a newline
appears. However, if a windows newline appeared, the buffer length was
not reset. This commit resets the length so the \r does not appear in
the next logging message.
closes#51838
MockRandomMergePolicy randomly determines if a segment should use a
compound format. This can cause a force merge performing two merges: (1)
merging to a single segment, (2) rewriting the new segment using the
compound format. If the second merge completes after we have flushed,
then it can flip the flag shouldPeriodicallyFlushAfterBigMerge to true.
Closes#52205
Modifies SLM's and ILM's history indices to be hidden indices for added
protection against accidental querying and deletion, and improves
IndexTemplateRegistry to handle upgrading index templates.
Also modifies the REST test cleanup to delete hidden indices.
This removes a bunch of `instanceof`s in favor of two new methods on
`InernalAggregation`. The default implementations of these methods just
throw exceptions explaining that you can't sort on this aggregation.
They are overridden by all of the classes that used to have `instanceof`
checks against them.
I doubt this is really any faster in practice. The real benefit here is
that it is a little more obvious *that* you can sort by the results of
an aggregation and it should be *much* more obvious where to look at
*how* aggregations sort themselves.
There are still a bunch more `instanceof`s in left in `AggregationPath`
but those will wait for a followup change.
disallow to specify percentile out of range [0,100]. This also fixes a problem in transform by failing
validation if an invalid percentile configuration is used.
Changes the misleading error message when attempting to open
a job while the "cluster.persistent_tasks.allocation.enable"
setting is set to "none" to a clearer message that names the
setting.
Closes#51956
Today we use `cluster.join.timeout` to prevent nodes from waiting indefinitely
if joining a faulty master that is too slow to respond, and
`cluster.publish.timeout` to allow a faulty master to detect that it is unable
to publish its cluster state updates in a timely fashion. If these timeouts
occur then the node restarts the discovery process in an attempt to find a
healthier master.
In the special case of `discovery.type: single-node` there is no point in
looking for another healthier master since the single node in the cluster is
all we've got. This commit suppresses these timeouts and instead lets the node
wait for joins and publications to succeed no matter how long this might take.
Previously, the dot-index rules (namely, that indices with dot-prefixed
names should be either hidden indices or system indices) was done
before* template application, and so only checked for the `index.hidden`
setting in the request, ignoring if that setting was set via a template.
This commit moves that check to a different method, which is applied
after templates have been resolved and applied to the index settings.
This commit fixes another edge case in handling windows newlines in our
capture of stdout/stderr to log4j. The case is that the \r appears at
the beginning of the buffer when flushing, which would unintentionally
be emitted as an empty string. This commit skips the flush if only a \r
was found.
closes#51838
Currently, the logic for looking up `flattened` field types lives in the
top-level `FieldTypeLookup`. This PR moves it into a dedicated class
`DynamicKeyFieldTypeLookup`.
Segment(s) info blobs are already stored with their full content
in the "hash" field in the shard snapshot metadata as long as they are
smaller than 1MB. We can make use of this fact and never upload them
physically to the repo.
This saves a non-trivial number of uploads and downloads when restoring
and might also lower the latency of searchable snapshots since they can save
phyiscally loading this information as well.
- Enable SunJGSS provider for Kerberos tests
- Handle the fact that in the decrypt method in KeyStoreWrapper might
not throw immediately when the GCM cipher is from BouncyCastle FIPS
and we end up with a DataInputStream that has reached it's end.
- Disable tests, jarHell, testingConventions for ingest attachment
plugin. We don't support this plugin (and document this) in FIPS
mode.
- Don't attempt to install ingest-attachment in smoke-test-plugins
This commit changes how RestHandlers are registered with the
RestController so that a RestHandler no longer needs to register itself
with the RestController. Instead the RestHandler interface has new
methods which when called provide information about the routes
(method and path combinations) that are handled by the handler
including any deprecated and/or replaced combinations.
This change also makes the publication of RestHandlers safe since they
no longer publish a reference to themselves within their constructors.
Closes#51622
Co-authored-by: Jason Tedor <jason@tedor.me>
Backport of #51950
We might leak a searcher if the target shard is removed (i.e., its index
is deleted) or relocated while we are creating a SearchContext from a
SearchRewriteContext.
Relates #51708Closes#52021
I labelled this non-issue for an unreleased bug introduced in #51708.
We can just put the `IndexId` instead of just the index name into the recovery soruce and
save one load of `RepositoryData` on each shard restore that way.
We need to either exclude null responses from the scroll search response
or always create a search context for every target shards, although that
scroll query can be written to match_no_docs. Otherwise, we won't find
search_context for subsequent scroll requests.
This commit implements the latter option as it's less error-prone.
Relates #51708
This change ensures that the rewrite of the shard request is executed in the network thread or in the refresh listener when waiting for an active shard. This allows queries that rewrite to match_no_docs to bypass the search thread pool entirely even if the can_match phase was skipped (pre_filter_shard_size > number of shards). Coordinating nodes don't have the ability to create empty responses so this change also ensures that at least one shard creates a full empty response while the other can return null ones. This is needed since creating true empty responses on shards require to create concrete aggregators which would be too costly to build on a network thread. We should move this functionality to aggregation builders in a follow up but that would be a much bigger change.
This change is also important for #49601 since we want to add the ability to use the result of other shards to rewrite the request of subsequent ones. For instance if the first M shards have their top N computed, the top worst document in the global queue can be pass to subsequent shards that can then rewrite to match_no_docs if they can guarantee that they don't have any document better than the provided one.
QueryBuilders that throw exceptions on shards when building the Lucene query
returns the full serialization of the query builder in the exception message.
For large queries that fails to execute due to the max boolean clause, this means
that we keep a reference of these big messages for every shard that participate
in the request. In order to limit the memory needed to hold these query shard
exceptions in the coordinating node, this change removes the query builder
serialization from the shard exception. The query is known by the user so
there should be no need to repeat it on every shard exception. We could also
omit the entire stack trace for known bad request exception but it would deserve
a separate issue/pr.
Closes#51843Closes#48910
When the `rare_terms` aggregation contained another aggregation it'd
break them. Most of the time. This happened because the process that it
uses to remove buckets that turn out not to be rare was incorrectly
merging results from multiple leaves. This'd cause array index out of
bounds issues. We didn't catch it in the test because the issue doesn't
happen on the very first bucket. And the tests generated data in such a
way that the first bucket always contained the rare terms. Randomizing
the order of the generated data fixed the test so it caught the issue.
Closes#51020
Currently, the same class `FieldCapabilities` is used both to represent the
capabilities for one index, and also the merged capabilities across indices. To
help clarify the logic, this PR proposes to create a separate class
`IndexFieldCapabilities` for the capabilities in one index. The refactor will
also help when adding `source_path` information in #49264, since the merged
source path field will have a different structure from the field for a single index.
Individual changes:
* Add a new class IndexFieldCapabilities.
* Remove extra constructor from FieldCapabilities.
* Combine the add and merge methods in FieldCapabilities.Builder.
Currently, a mappings update request, where dynamic_mappings is an object
instead of an array, results in a http response with a 500 code. This PR checks
for this condition and throws a MapperParsingException like we do for other
malformed mapping cases.
Closes#51486
ActionListener.completeWith would catch exceptions from
listener.onResponse and deliver them to lister.onFailure, essentially
double notifying the listener. Instead we now assert that listeners do
not throw when using ActionListener.completeWith.
Relates #50886
when a timezone is not provided Ingest logic should consider a time to be in a timezone provided as a parameter.
When a timezone is provided Ingest should recalculate a time to the timezone provided as a parameter
closes#51108
backport(#51215)
With the new mechanism for storing cluster state in lucene, we store
index metadata in multiple data paths too. This causes cluster state
publish to timeout too frequently with a 1s timeout, so increasing it to
5s. Also increasing follower check timeout to 5s since it also sometimes
has fsync in its timeout path and leader check for symmetry.
Closes#51329
While we use `== false` as a more visible form of boolean negation
(instead of `!`), the true case is implied and the true value does not
need to explicitly checked. This commit converts cases that have slipped
into the code checking for `== true`.
LoggingOutputStream reads a stream and breaks on newlines. This commit
fixes the behavior to account for windows newlines also containing `\r`.
closes#51532
This commit inlines the `weightShardAdded` and `weightShardRemoved` methods
from the `BalancedShardsAllocator#WeightFunction` that respectively add and
subtract 1 (±ε) from the result of `weight`. It then follows up with a number
of simplifications that this inlining enables.
As a side-effect it also somewhat reduces the number of calls to canRebalance
and canAllocate during rebalancing when there are multiple shards of the same
index on a node that is heavier than average.
When running Elasticsearch on a flaky network, we may see nodes leaving the
cluster with reason `disconnected`. It may be useful to the cluster
administrator to see the full exception that caused the disconnection, but this
is only available with `TRACE` level logging which commingles the details of
the problem with other messages that are not useful to end users.
This commit promotes logging of exceptions in `TcpTransport` from `TRACE` to
`DEBUG` to separate them from the truly `TRACE`-level messages.
This commit switches the strategy for managing dot-prefixed indices that
should be hidden indices from using "fake" system indices to an explicit
exclusions list that must be updated when those indices are converted to
hidden indices.
* Fix InternalEngineTests.testSeqNoAndCheckpoints
If we force flush while possibly triggering a merge the local checkpoint may change
from the expectation from the loop that just increments on every operation.
Closes#51604
There is no reason not to allow deletes in parallel to restores
if they're dealing with different snapshots.
A delete will not remove any files related to the snapshot that
is being restored if it is different from the deleted snapshot
because those files will still be referenced by the restoring
snapshot.
Loading RepositoryData concurrently to modifying it is concurrency
safe nowadays as well since the repo generation is tracked in the
cluster state.
Closes#41463
ActionListener.map would call listener.onFailure for exceptions from
listener.onResponse, but this means we could double trigger some
listeners which is generally unexpected. Instead, we should assume that
a listener's onResponse (and onFailure) implementation is responsible
for its own exception handling.
In #50259 we redirected stdout and stderr to log4j, to capture jdk
and external library messages. However, a typo in the method name used
to redirect the stream in java means stdout is currently being
duplicated twice, and stderr not captured. This commit corrects that
mistake. Unfortunately this is at a level that cannot really be tested,
thus we are still missing tests for this behavior.
We need to warm up the engine (i.e., perform an external refresh) before
accessing the external refresh. Note that we refresh externally before
allowing reading from a shard.
Relates #48605Closes#51548
When checking if a device is up, today we can run into virtual ethernet
devices that disappear while we are in the middle of checking. This
leads to "no such device". This commit addresses such devices by
treating them as not being up, if they are virtual ethernet devices that
disappeared while we were checking.
This commit moves the logic that cancels search requests when the rest channel is closed
to a generic client that can be used by other APIs. This will be useful for any rest action
that wants to cancel the execution of a task if the underlying rest channel is closed by the
client before completion.
Relates #49931
Relates #50990
Relates #50990
* Allow Repository Plugins to Filter Metadata on Create
Add a hook that allows repository plugins to filter the repository metadata
before it gets written to the cluster state.
This commit deprecates the creation of dot-prefixed index names (e.g.
.watches) unless they are either 1) a hidden index, or 2) registered by
a plugin that extends SystemIndexPlugin. This is the first step
towards more thorough protections for system indices.
This commit also modifies several plugins which use dot-prefixed indices
to register indices they own as system indices, and adds a plugin to
register .tasks as a system index.
* Reload secure settings with password (#43197)
If a password is not set, we assume an empty string to be
compatible with previous behavior.
Only allow the reload to be broadcast to other nodes if TLS is
enabled for the transport layer.
* Add passphrase support to elasticsearch-keystore (#38498)
This change adds support for keystore passphrases to all subcommands
of the elasticsearch-keystore cli tool and adds a subcommand for
changing the passphrase of an existing keystore.
The work to read the passphrase in Elasticsearch when
loading, which will be addressed in a different PR.
Subcommands of elasticsearch-keystore can handle (open and create)
passphrase protected keystores
When reading a keystore, a user is only prompted for a passphrase
only if the keystore is passphrase protected.
When creating a keystore, a user is allowed (default behavior) to create one with an
empty passphrase
Passphrase can be set to be empty when changing/setting it for an
existing keystore
Relates to: #32691
Supersedes: #37472
* Restore behavior for force parameter (#44847)
Turns out that the behavior of `-f` for the add and add-file sub
commands where it would also forcibly create the keystore if it
didn't exist, was by design - although undocumented.
This change restores that behavior auto-creating a keystore that
is not password protected if the force flag is used. The force
OptionSpec is moved to the BaseKeyStoreCommand as we will presumably
want to maintain the same behavior in any other command that takes
a force option.
* Handle pwd protected keystores in all CLI tools (#45289)
This change ensures that `elasticsearch-setup-passwords` and
`elasticsearch-saml-metadata` can handle a password protected
elasticsearch.keystore.
For setup passwords the user would be prompted to add the
elasticsearch keystore password upon running the tool. There is no
option to pass the password as a parameter as we assume the user is
present in order to enter the desired passwords for the built-in
users.
For saml-metadata, we prompt for the keystore password at all times
even though we'd only need to read something from the keystore when
there is a signing or encryption configuration.
* Modify docs for setup passwords and saml metadata cli (#45797)
Adds a sentence in the documentation of `elasticsearch-setup-passwords`
and `elasticsearch-saml-metadata` to describe that users would be
prompted for the keystore's password when running these CLI tools,
when the keystore is password protected.
Co-Authored-By: Lisa Cawley <lcawley@elastic.co>
* Elasticsearch keystore passphrase for startup scripts (#44775)
This commit allows a user to provide a keystore password on Elasticsearch
startup, but only prompts when the keystore exists and is encrypted.
The entrypoint in Java code is standard input. When the Bootstrap class is
checking for secure keystore settings, it checks whether or not the keystore
is encrypted. If so, we read one line from standard input and use this as the
password. For simplicity's sake, we allow a maximum passphrase length of 128
characters. (This is an arbitrary limit and could be increased or eliminated.
It is also enforced in the keystore tools, so that a user can't create a
password that's too long to enter at startup.)
In order to provide a password on standard input, we have to account for four
different ways of starting Elasticsearch: the bash startup script, the Windows
batch startup script, systemd startup, and docker startup. We use wrapper
scripts to reduce systemd and docker to the bash case: in both cases, a
wrapper script can read a passphrase from the filesystem and pass it to the
bash script.
In order to simplify testing the need for a passphrase, I have added a
has-passwd command to the keystore tool. This command can run silently, and
exit with status 0 when the keystore has a password. It exits with status 1 if
the keystore doesn't exist or exists and is unencrypted.
A good deal of the code-change in this commit has to do with refactoring
packaging tests to cleanly use the same tests for both the "archive" and the
"package" cases. This required not only moving tests around, but also adding
some convenience methods for an abstraction layer over distribution-specific
commands.
* Adjust docs for password protected keystore (#45054)
This commit adds relevant parts in the elasticsearch-keystore
sub-commands reference docs and in the reload secure settings API
doc.
* Fix failing Keystore Passphrase test for feature branch (#50154)
One problem with the passphrase-from-file tests, as written, is that
they would leave a SystemD environment variable set when they failed,
and this setting would cause elasticsearch startup to fail for other
tests as well. By using a try-finally, I hope that these tests will fail
more gracefully.
It appears that our Fedora and Ubuntu environments may be configured to
store journald information under /var rather than under /run, so that it
will persist between boots. Our destructive tests that read from the
journal need to account for this in order to avoid trying to limit the
output we check in tests.
* Run keystore management tests on docker distros (#50610)
* Add Docker handling to PackagingTestCase
Keystore tests need to be able to run in the Docker case. We can do this
by using a DockerShell instead of a plain Shell when Docker is running.
* Improve ES startup check for docker
Previously we were checking truncated output for the packaged JDK as
an indication that Elasticsearch had started. With new preliminary
password checks, we might get a false positive from ES keystore
commands, so we have to check specifically that the Elasticsearch
class from the Bootstrap package is what's running.
* Test password-protected keystore with Docker (#50803)
This commit adds two tests for the case where we mount a
password-protected keystore into a Docker container and provide a
password via a Docker environment variable.
We also fix a logging bug where we were logging the identifier for an
array of strings rather than the contents of that array.
* Add documentation for keystore startup prompting (#50821)
When a keystore is password-protected, Elasticsearch will prompt at
startup. This commit adds documentation for this prompt for the archive,
systemd, and Docker cases.
Co-authored-by: Lisa Cawley <lcawley@elastic.co>
* Warn when unable to upgrade keystore on debian (#51011)
For Red Hat RPM upgrades, we warn if we can't upgrade the keystore. This
commit brings the same logic to the code for Debian packages. See the
posttrans file for gets executed for RPMs.
* Restore handling of string input
Adds tests that were mistakenly removed. One of these tests proved
we were not handling the the stdin (-x) option correctly when no
input was added. This commit restores the original approach of
reading stdin one char at a time until there is no more (-1, \r, \n)
instead of using readline() that might return null
* Apply spotless reformatting
* Use '--since' flag to get recent journal messages
When we get Elasticsearch logs from journald, we want to fetch only log
messages from the last run. There are two reasons for this. First, if
there are many logs, we might get a string that's too large for our
utility methods. Second, when we're looking for a specific message or
error, we almost certainly want to look only at messages from the last
execution.
Previously, we've been trying to do this by clearing out the physical
files under the journald process. But there seems to be some contention
over these directories: if journald writes a log file in between when
our deletion command deletes the file and when it deletes the log
directory, the deletion will fail.
It seems to me that we might be able to use journald's "--since" flag to
retrieve only log messages from the last run, and that this might be
less likely to fail due to race conditions in file deletion.
Unfortunately, it looks as if the "--since" flag has a granularity of
one-second. I've added a two-second sleep to make sure that there's a
sufficient gap between the test that will read from journald and the
test before it.
* Use new journald wrapper pattern
* Update version added in secure settings request
Co-authored-by: Lisa Cawley <lcawley@elastic.co>
Co-authored-by: Ioannis Kakavas <ikakavas@protonmail.com>
This commit modifies the bounding box for geogrid unit tests
to only consider bounding boxes that have significant longitudinal
width and whose coordinates are normalized to quantized space
Closes#51103.
We added a new rounding in #50609 that handles offsets to the start and
end of the rounding so that we could support `offset` in the `composite`
aggregation. This starts moving `date_histogram` to that new offset.
This is a redo of #50873 with more integration tests.
This reverts commit d114c9db3e1d1a766f9f48f846eed0466125ce83.
We've been parsing the `time_zone` parameter on `date_hitogram` for a
while but it hasn't *done* anything. This wires it up.
Closes#45199
Inspired by #45200
We shouldn't be using potentially changing versions of the cluster state
when answering a snapshot status API call by calling `SnapshotService#currentSnapshots` multiple times (each time using `ClusterService#state` under the hood) but instead pass down the state from the transport action.
Having these API behave more in a more deterministic way will make it easier to use them once parallel repository operations
are introduced.
This commit overrides the stdout and stderr print streams to be
redirected to the main elasticsearch.log file. While the Elasticsearch
project ensures stdout and stderr are not written to, the jdk or 3rd
party libs may do this, which can be unexepected for users used to
looking the elasticsearch log.
closes#50156
Wait for the cluster to have settled down and have the same accepted version on all nodes before
executing and cancelling request so that a slow CS accept on one node doesn't make it fall behind
and then get sent the full CS because of the diff-version mismatch, breaking the mechanics of this test.
Closes#51308
Added node closed exception to the retryable remote exceptions as it's possible to run into this exception instead of a connect exception when the master node is just shutting down but still responding to requests.
* Fix Inconsistent Shard Failure Count in Failed Snapshots
This fix was necessary to allow for the below test enhancement:
We were not adding shard failure entries to a failed snapshot for those
snapshot entries that were never attempted because the snapshot failed during
the init stage and wasn't partial. This caused the never attempted snapshots
to be counted towards the successful shard count which seems wrong and
broke repository consistency tests.
Also, this change adjusts snapshot resiliency tests to run another snapshot
at the end of each test run to guarantee a correct `index.latest` blob exists
after each run.
Closes#47550
The order indices are returned in in the metadata is not guaranteed.
This commit accounts for any possible ordering in assertions about
hidden indices.
closes#51340
It is permitted for nodes to accept transport connections at addresses other
than their publish address, which allows a good deal of flexibility when
configuring discovery. However, it is not unusual for users to misconfigure
nodes to pick a publish address which is inaccessible to other nodes. We see
this happen a lot if the nodes are on different networks separated by a proxy,
or if the nodes are running in Docker with the wrong kind of network config.
In this case we offer no useful feedback to the user unless they enable
TRACE-level logs. It's particularly tricky to diagnose because if we test
connectivity between the nodes (using their discovery addresses) then all will
appear well.
This commit adds a WARN-level log if this kind of misconfiguration is detected:
the probe connection has succeeded (to indicate that we are really talking to a
healthy Elasticsearch node) but the followup connection attempt fails.
It also tidies up some loose ends in `HandshakingTransportAddressConnector`,
removing some TODOs that need not be completed, and registering its
accidentally-unregistered timeout settings.
IndexWriter might not filter out fully deleted segments if retention
leases exist or the number of the retaining operations is non-zero.
SoftDeletesDirectoryReaderWrapper, however, always filters out fully
deleted segments.
This change uses the original directory reader when calculating segment
stats instead.
Relates #51192Closes#51303
We were loading `RepositoryData` twice during snapshot initialization,
redundantly checking if a snapshot existed already.
The first snapshot existence check is somewhat redundant because a snapshot could be
created between loading `RepositoryData` and updating the cluster state with the `INIT`
state snapshot entry.
Also, it is much safer to do the subsequent checks for index existence in the repo and
and the presence of old version snapshots once the `INIT` state entry prevents further
snapshots from being created concurrently.
While the current state of things will never lead to corruption on a concurrent snapshot
creation, it could result in a situation (though unlikely) where all the snapshot's work
is done on the data nodes, only to find out that the repository generation was off during
snapshot finalization, failing there and leaving a bunch of dead data in the repository
that won't be used in a subsequent snapshot (because the shard generation was never referenced
due to the failed snapshot finalization).
Note: This is a step on the way to parallel repository operations by making snapshot related CS
and repo related CS more tightly correlated.
This reverts commit c7fd24ca1569a809b499caf34077599e463bb8d6.
Now that JDK-8236582 is fixed in JDK 14 EA, we can revert the workaround.
Relates #50523 and #50512
LuceneChangesSnapshot can be slow if nested documents are heavily used.
Also, it estimates the number of operations to be recovered in peer
recoveries inaccurately. With this change, we prefer excluding the
nested non-root documents in a Lucene query instead.
The method parameter is not used in the percentile aggs, instead
the method is determined by the presence of `hdr` or `tdigest`
objects.
Relates to #8324
After we rollover the index we wait for the configured number of shards for the
rolled index to become active (based on the index.write.wait_for_active_shards
setting which might be present in a template, or otherwise in the default case,
for the primaries to become active).
This wait might be long due to disk watermarks being tripped, replicas not
being able to spring to life due to cluster nodes reconfiguration and others
and, the RolloverStep might not complete successfully due to this inherent
transient situation, albeit the rolled index having been created.
(cherry picked from commit 457a92fb4c68c55976cc3c3e2f00a053dd2eac70)
Signed-off-by: Andrei Dan <andrei.dan@elastic.co>
On master failover we have to resent all the shard failed messages,
but the transport requests remain the same in the eyes of `equals`.
If the master failover is registered and the requests to the new master
are sent before all the callbacks have executed and the request to the
old master removed from the deduplicator then the requuests to the new
master will incorrectly fail and the snapshot get stuck.
Closes#51253
Add the character position of a scripting error to error responses.
The contents of the `position` field are experimental and subject to
change. Currently, `offset` refers to the character location where the
error was encountered, `start` and `end` define a range of characters
that contain the error.
eg.
```
{
"error": {
"root_cause": [
{
"type": "script_exception",
"reason": "runtime error",
"script_stack": [
"y = x;",
" ^---- HERE"
],
"script": "def x = new ArrayList(); Map y = x;",
"lang": "painless",
"position": {
"offset": 33,
"start": 29,
"end": 35
}
}
```
Refs: #50993
This replaces the message we return for unknown queries with the standard
one that we use for unknown fields from `ObjectParser`. This is nice
because it includes "did you mean". One day we might convert parsing
queries to using object parser, but that looks complex. This change is
much smaller and seems useful.
Adding back accidentally removed jvm option that is required to enforce
start of the week = Monday in IsoCalendarDataProvider.
Adding a `feature` to yml test in order to skip running it in JDK8
commit that removed it 398c802
commit that backports SystemJvmOptions c4fbda3
relates 7.x backport of code that enforces CalendarDataProvider use #48349
The tests, when creating broken serialized blobs could randomly create
a sequence of bytes that is partially readable by the deserializer and then
not throw `IOException` but instead `ElasticsearchParseException`.
We should just handle these unexpected exceptions downstream properly and pass them
wrapped as `RepositoryException` to the listener to fix the test and keep the API consistent.
Jason Tedor
Stuart Tettemer
Jim Ferenczi
Nhat Nguyen
Alan Woodward
Marios Trivyzas
Tanguy Leroux
Tianlun Li
Lee Hinman
Ryan Ernst
Luca Cavanna
Nik Everett
Gordon Brown
markharwood
Mayya Sharipova
David Turner
Mark Vieira
Dan Hermann
William Brafford
Jay Modi
Przemyslaw Gomulka
Armin Braun
Christoph Büscher
Henning Andersen
James Baiera
Ignacio Vera
Tim Brooks
Yannick Welsch
Jake Landis
Zachary Tong
Adrien Grand
Martijn van Groningen
hezhen Zhang
Sachin Frayne
bellengao
Przemysław Witek
Benedict Jin
Julie Tibshirani
Hendrik Muhs
David Roberts
Zachary Tong
Igor Motov
David Kyle
Ignacio Vera
Ioannis Kakavas
Maria Ralli
feifeiiiiiiiiii
Noor
Dominic Page
Tal Levy
Ryan Ernst
Nirmal Chidambaram
Andrei Dan
Paul Sanwald
zacharymorn