When a node tries to join a cluster, it goes through a validation step to make sure the node is compatible with the cluster. Currently we validation that the node can read the cluster state and that it is compatible with the indexes of the cluster. This PR adds validation that the joining node's version is compatible with the versions of existing nodes. Concretely we check that:
1) The node's min compatible version is higher or equal to any node in the cluster (this prevents a too-new node from joining)
2) The node's version is higher or equal to the min compat version of all cluster nodes (this prevents a too old join where, for example, the master is on 5.6, there's another 6.0 node in the cluster and a 5.4 node tries to join).
3) The node's major version is at least as higher as the lowest node in the cluster. This is important as we use the minimum version in the cluster to stop executing bwc code for operations that require multiple nodes. If the nodes are already operating in "new cluster mode", we should prevent nodes from the previous major to join (even if they are wire level compatible). This does mean that if you have a very unlucky partition during the upgrade which partitions all old nodes which are also a minority / data nodes only, the may not be able to re-join the cluster. We feel this edge case risk is well worth the simplification it brings to BWC layers only going one way.
Also, the node join validation can now selectively fail specific nodes (previously the entire batch was failed). This is an important preparation for a follow up PR where we plan to have a rejected joining node die with dignity.
Today we provide a lot of functionality on the `QueryRewriteContext` that
we potentially don't have ie. if we rewrite on a coordinating node or when
we percolating. This change moves most of the unnecessary shard level or
index level services and dependencies to `QueryShardContext` instead.
If a request contains an invalid error trace parameter, we send a error
on the channel. This should immediately abort any additional processing
of the request but instead we march on, dispatch the request and
subsequently send another message on the channel. The problem here is
this means two writes on the channel which leads to the request being
released twice ultimately raising in illegal reference count
exception. This commit addresses this by performing an early return in
the case that the request contained an invalid error trace parameter.
Relates #25785
This commit removes a timed latch await in a transport client listeners
test. The problem with a timed wait here is that on an overloaded
machine, the test can fail because the waiting thread was not unlatched
quickly enough. This makes the test unnecessarily flaky. Instead, we
should wait indefinitely and simply let the test fail by the test
timeout if the latch is not counted down for some reason.
Closes#25760
This commit expands on the migration note regarding the removal of
default.path.data and default.path.logs to include a note that users
that were relying on the defaults (the common case for path.logs), and
they carry over their previous elasticsearch.yml configruation file,
then they must add explicit values for path.data and path.logs.
Currently we ignore unknown field names when parsing RangeAggregator.Range and
GeoDistanceAggregationBuilder.Range from `range`, `date_range` or `geo_distance`
aggregations. This can hide subtle errors in the query. This change makes parsing `ranges`
stricter.
This is an appealing assertion, but there scenarios where it can happen under normal operations. For example, when an index is created it may run into an exception when the lucene files have already been created. The master will try to assign the shard to another node (it's empty, so no need to look for data) but if there is no other node, it will reassign it to the same node. At that point the deletion will get a list of existing commits (which it will typically delete).
With #23997 and #25268 we have changed put alias, delete alias, update aliases and delete index to not accept aliases. Instead concrete indices should be provided as their index parameter.
This commit improves the error message in case aliases are provided, from an IndexNotFoundException (404 status code) with "no such index" message, to an IllegalArgumentException (400 status code) with "The provided expression [alias] matches an alias, specify the corresponding concrete indices instead." message.
Note that there is no specific error message for the case where wildcard expressions match one or more aliases. In fact, aliases are simply ignored when expanding wildcards for such APIs. An error is thrown only when the expression ends up matching no indices at all, and allow_no_indices is set to false. In that case the error is still the generic "404 - no such index".
403 can be confused with security. If an API doesn't support working against closed indices and closed indices are referred to in a request, that is a bad request, hence 400 is more appropriate.
The test checks if a file based or ops based recovery happened, but if the replica shard never finished recovering expectations are not met.
Fixes#25761
Currently the `to` and `from` parameter in the `date_range` aggregation is not
parsed with the correct date field format from the mappings or the aggregation
if the argument is numeric, but always treated as a long value specifying
`epoch_millis`. This leads to problems e.g. when the format is `epoch_second`,
but the `to` and `from` are currently treated as millis.
With this change, we interpret these parameters according to the `format` of the target field.
If the `format` in the mappings is not compatible with numeric input values,
a compatible `format` (e.g. `epoch_millis`, `epoch_second`) must be specified in
the `date_range` aggregation itself, otherwise an error is thrown.
#Closes #17920
The problem here is simple: when using direct buffers as in NIO, the JDK
relies on explict GC invocataions to trigger cleaning up direct buffers;
if such GCs do not occur and the direct buffer limit is reached, the JVM
will throw an out of memory exception. With explicit GCs disabled, the
JVM is neutered from explicitly cleaning up direct buffers in the act of
reserving a new direct buffer and instead relies on a GC occurring for
another reason. If such a GC never occurs, the JVM will OOM. This commit
removes disabling of explicit GCs. Note that these explicit GCs only
occur as a last ditch effort before going OOM when the JVM is trying to
reserve more direct memory. This is a known issue, see for example:
JDK-8142537.
Relates #25759
This commit changes how the offset source is picked for each field using the es mapping rather than the underlying Lucene field infos.
It's mandatory for large mappings where field infos retrieval can be costly (the global field infos is merged for each highlighted field in every hit by the Lucene impl).
Fixes#25699
* Register data node stats from info carried back in search responses
This is part of #24915, where we now calculate the EWMA of service time for
tasks in the search threadpool, and send that as well as the current queue size
back to the coordinating node. The coordinating node now tracks this information
for each node in the cluster.
This information will be used in the future the determining the best replica a
search request should be routed to. This change has no user-visible difference.
* Move response time timing into ResponseListenerWrapper
* Move ResponseListenerWrapper to ActionListener instead of SearchActionListener
Also removes the logger
* Move `requestIndex` back to private
* De-guice-ify ResponseCollectorService \o/
* Undo all changes to SearchQueryThenFetchAsyncAction
* Remove unneeded response collector from TransportSearchAction
* Undo all changes to SearchDfsQueryThenFetchAsyncAction
* Completely rewrite the inside of ResponseCollectorService's record keeping
* Documentation and cleanups for ResponseCollectorService
* Add unit test for collection of queue size and service time
* Fix Guice construction error
* Add basic unit tests for ResponseCollectorService
* Fix version constant for the master merge
* Fix test compilation after master merge
* Add a test for node removal on cluster changed event
* Remove integration test as there are now unit tests
* Rename ResponseListenerWrapper -> SearchExecutionStatsCollector
* Fix line-length
* Make classes private and final where appropriate
* Pass nodeId into SearchExecutionStatsCollector and use only ActionListener
* Get nodeId from connection so searchShardTarget can be private
* Remove threadpool from SearchContext, get it from IndexShard instead
* Add missing import
* Use BiFunction for responseWrapper rather than passing in collector service
This change removes the leniency of having a `null` index to fetch
terms from in 6.0 onwards. This feature will be deprecated in the 5.x series
and 6.0 nodes will require the index to be set.
Closes#25750
We used to compare agaisnt the min compatible version which is misleading since
it might move over time and since we backported the `can_match` API entirely
it's better to compare against a version constant.
We can't do it in the general case because of prefix queries, but I believe this
is mostly used in query strings and not in explicit `terms` queries.
Closes#25667
This commit reverts some changes to the shrink API ignore template
mapping REST test in favor of simply skipping the test for BWC
purposes. The complexity here is due to deprecations and lacking the
infrastructure to gracefully handle a situation like this.
This change skips the rest test in `rest-api-spec/test/indices.shrink/20_source_mapping.yml` as it currently fails because if we don’t expect the deprecation warning the normal rest tests fail because they get a warning they don’t expect but if we do expect the deprecation warning the mixed cluster tests fail because they don’t get a warning which they expected.
This commit fixes an issue with the REST test that the shrink API
ignores templates. The problem is that we have to use a BWC version of
the API (for the BWC tests) but this raises deprecation warnings. This
commit adds an expectation for these deprecation warnings.
When sending replica requests for replication operations, we skip
sending the request to pre-6.0 nodes for operations that such nodes
would not be aware of (e.g., the background global checkpoint sync, or
the primary/replica resync) since they would not know what to do with
these requests. Yet, we simulate that we received responses from these
nodes. Today, this is done by simulating that they sent us that their
local checkpoint is unassigned sequence number. However, for pre-6.0
nodes we have introduced a special local checkpoint used in the global
checkpoint tracker for such nodes and that is what we should use here
too. This commit fixes this issue.
Relates #25744
The following token filters were moved: arabic_normalization, german_normalization, hindi_normalization, indic_normalization, persian_normalization, scandinavian_normalization, serbian_normalization, sorani_normalization, cjk_width and cjk_width
Relates to #23658
When simulating an ingest pipeline against an existing pipeline, the
_source field is required to wrap each doc. This commit fixes another
example in the docs that is missing this.
Relates #25743, relates e3a0c11239
When simulating an ingest pipeline against an existing pipeline, the
_source field is required to wrap each doc. This commit fixes an example
in the docs that is missing this.
Relates #25742
This commit adjusts the skip version for a shrink index test that
ensures that a shrunken index ignores templates; the version can be
adjusted after the fix was backported targeting 5.6.0 and later.
Relates #25380
#25521 changed channel closing to be handled async on anything but transport stop. This means it may take a while before
calling `connection.close()` and the node being removed from the `connectedNodes` list (but the connection is immediately unusuable).
Fixes#25686
The alpha2 docs is built out of master which may make users think that the high level client was already released as part of alpha2 which it was not. This note should clarify that the client will be released with 6.0.0-beta1
This commit changes the default heap size to 1 GB. Experimenting with
elasticsearch is often done on laptops, and 1 GB is much friendlier to
laptop memory. It does put more pressure on the gc, but the tradeoff is
a smaller default footprint. Users running in production can (and
should) adjust the heap size as necessary for their usecase.
Currently replication and recovery are both coordinated through the latest cluster state available on the ClusterService as well as through the GlobalCheckpointTracker (to have consistent local/global checkpoint information), making it difficult to understand the relation between recovery and replication, and requiring some tricky checks in the recovery code to coordinate between the two. This commit makes the primary the single owner of its replication group, which simplifies the replication model and allows to clean up corner cases we have in our recovery code. It also reduces the dependencies in the code, so that neither RecoverySourceXXX nor ReplicationOperation need access to the latest state on ClusterService anymore. Finally, it gives us the property that in-sync shard copies won't receive global checkpoint updates which are above their local checkpoint (relates #25485).
This adds a section about how to add aggregations to the SearchSourceBuilder and how
to retrieve them from a SearchRepsonse to the documentation for the high level rest client.
The Filtered Query has been deprecated in favour of the Bool Query with a filter context. However, this deleted page for the Filtered Query is often ranked highly in search results when searching for documentation on "filtered queries". Often people just copy the first code snippet they see, which in this case is the INCORRECT syntax (the correct syntax follows). I think reordering the examples would help avoid a lot of confusion (I have seen people make this same mistake 3 times now)
Adding a comment to indicate that the first example shouldn't be used
Boaz Leskes
Simon Willnauer
Jason Tedor
Deb Adair
Jim Ferenczi
Christoph Büscher
Luca Cavanna
Clinton Gormley
Lee Hinman
Adrien Grand
Colin Goodheart-Smithe
Martijn van Groningen
Glen Smith
Ryan Ernst
javanna
Yannick Welsch
Bodecker DellaMaria