Adds a tighter threshold for logging a warning about slowness in the
`MasterService` instead of relying on the cluster service's 30-second warning
threshold. This new threshold applies to the computation of the cluster state
update in isolation, so we get a warning if computing a new cluster state
update takes longer than 10 seconds even if it is subsequently applied quickly.
It also applies independently to the length of time it takes to notify the
cluster state tasks on completion of publication, in case any of these
notifications holds up the master thread for too long.
Relates #45007
Backport of #45086
This commit adds a deprecation warning in 7.x for the Force Merge API
when both only_expunge_deletes and max_num_segments are set in a request.
Relates #44761
This commit applies a normalization process to environment paths, both
in how they are stored internally, also their settings values. This
normalization is done via two means:
- we make the paths absolute
- we remove redundant name elements from the path (what Java calls
"normalization")
This change ensures that when we compare and refer to these paths within
the system, we are using a common ground. For example, prior to the
change if the data path was relative, we would not compare it correctly
to paths from disk usage. This is because the paths in disk usage were
being made absolute.
Uses JDK 11's per-socket configuration of TCP keepalive (supported on Linux and Mac), see
https://bugs.openjdk.java.net/browse/JDK-8194298, and exposes these as transport settings.
By default, these options are disabled for now (i.e. fall-back to OS behavior), but we would like
to explore whether we can enable them by default, in particular to force keepalive configurations
that are better tuned for running ES.
This adjusts the `buckets_path` parser so that pipeline aggs can
select specific buckets (via their bucket keys) instead of fetching
the entire set of buckets. This is useful for bucket_script in
particular, which might want specific buckets for calculations.
It's possible to workaround this with `filter` aggs, but the workaround
is hacky and probably less performant.
- Adjusts documentation
- Adds a barebones AggregatorTestCase for bucket_script
- Tweaks AggTestCase to use getMockScriptService() for reductions and
pipelines. Previously pipelines could just pass in a script service
for testing, but this didnt work for regular aggs. The new
getMockScriptService() method fixes that issue, but needs to be used
for pipelines too. This had a knock-on effect of touching MovFn,
AvgBucket and ScriptedMetric
Introduce shift field to MovingFunction aggregation.
By default, shift = 0. Behavior, in this case, is the same as before.
Increasing shift by 1 moves starting window position by 1 to the right.
To simply include current bucket to the window, use shift = 1
For center alignment (n/2 values before and after the current bucket), use shift = window / 2
For right alignment (n values after the current bucket), use shift = window.
* Resolve TODO in `readString` by moving to reading chunks of `byte[]` instead of going byte by byte
* Motivated by `readString` showing up as a significant user of CPU time on the IO thread in Rally PMC benchmark
* Benchmarking this:
* Could not reproduce a slowdown in the potential worst case (one or two non-ascii chars) since in this case the cost of creating the string itself exceeds the read times anyway
* Speedup for 50%+ for reading 200 char ascii strings from `ByteBuf` or pages bytes backed streams
* Longer strings obviously get bigger speedups
* More ascii chars -> more speedup
This commit switches to using the full hash to build into the JAR
manifest, which is used in node startup and the REST main action to
display the build hash.
Currently in the transport-nio work we connect and bind channels on the
a thread before the channel is registered with a selector. Additionally,
it is at this point that we set all the socket options. This commit
moves these operations onto the event-loop after the channel has been
registered with a selector. It attempts to set the socket options for a
non-server channel at registration time. If that fails, it will attempt
to set the options after the channel is connected. This should fix
#41071.
Introduce shift field to MovingFunction aggregation.
By default, shift = 0. Behavior, in this case, is the same as before.
Increasing shift by 1 moves starting window position by 1 to the right.
To simply include current bucket to the window, use shift = 1
For center alignment (n/2 values before and after the current bucket), use shift = window / 2
For right alignment (n values after the current bucket), use shift = window.
Today we recover a replica by copying operations from the primary's translog.
However we also retain some historical operations in the index itself, as long
as soft-deletes are enabled. This commit adjusts peer recovery to use the
operations in the index for recovery rather than those in the translog, and
ensures that the replication group retains enough history for use in peer
recovery by means of retention leases.
Reverts #38904 and #42211
Relates #41536
Backport of #45136 to 7.x.
* Stop Passing Around REST Request in Multiple Spots
* Motivated by #44564
* We are currently passing the REST request object around to a large number of places. This works fine since we simply copy the full request content before we handle the rest itself which is needlessly hard on GC and heap.
* This PR removes a number of spots where the request is passed around needlessly. There are many more spots to optimize in follow-ups to this, but this one would already enable bypassing the request copying for some error paths in a follow up.
Sparse role queries are executed differently than other queries in order
to account for the fact that most of the documents are filtered from search.
However this special execution does not set the scorer for the query so any
collector that needs to access the score of a document fails with an NPE.
This change fixed this bug by setting the scorer before collecting any hits
when intersecting the main query and the sparse role.
The Settings#processSetting method is intended to take a setting map and add a
setting to it, adjusting the keys as it goes in case of "conflicts" where the
new setting implies an object where there is currently a string, or vice
versa. processSetting was failing in two cases: adding a setting two levels
under a string, and adding a setting two levels under a string and four levels
under a map. This commit fixes the bug and adds test coverage for the
previously faulty edge cases.
* fix issue #43791 about settings
* add unit test in testProcessSetting()
We keep adding the current primary term to operations for which we do not assign a sequence
number. This does not make sense anymore as all operations which we care about have
sequence numbers now. The goal of this commit is to clean things up in InternalEngine and
reduce the complexity.
* There's no need to have the trie iterator hold another reference to the request object (which could be huge, see #44564)
* Also removed unused boolean field from trie node
Previously, we use ThreadPoolStats to ensure that the scheduledRefresh
triggered by the internal refresh setting update is executed before we
index a new document. With that change (#40387), this test did not fail for
the last 3 months. However, using ThreadPoolStats is not entirely watertight
as both "active" and "queue" count can be 0 in a very small interval
when ThreadPoolExecutor pulls a task from the queue but before marking
the corresponding worker as active (i.e., lock it).
Closes#39565
Adds a `waitForEvents(Priority.LANGUID)` to the cluster health request in
`ESIntegTestCase#waitForRelocation()` to deal with the case that this health
request returns successfully despite the fact that there is a pending reroute task which
will relocate another shard.
Relates #44433Fixes#45003
Today the lag detector may remove nodes from the cluster if they fail to apply
a cluster state within a reasonable timeframe, but it is rather unclear from
the default logging that this has occurred and there is very little extra
information beyond the fact that the removed node was lagging. Moreover the
only forewarning that the lag detector might be invoked is a message indicating
that cluster state publication took unreasonably long, which does not contain
enough information to investigate the problem further.
This commit adds a good deal more detail to make the issues of slow nodes more
prominent:
- after 10 seconds (by default) we log an INFO message indicating that a
publication is still waiting for responses from some nodes, including the
identities of the problematic nodes.
- when the publication times out after 30 seconds (by default) we log a WARN
message identifying the nodes that are still pending.
- the lag detector logs a more detailed warning when a fatally-lagging node is
detected.
- if applying a cluster state takes too long then the cluster applier service
logs a breakdown of all the tasks it ran as part of that process.
* Dry up code for creating simple `ActionRunnable` a little
* Shorten some other code around `ActionListener` usage, in particular
when wrapping it in a `TransportResponseListener`
* As a result of #44096 this test shouldn't fail anymore on `master` and `7.4`+ so we should reenable it there
* For older versions we won't backport that change so the tests should stay disabled there
* Closes#44671
Reading up on #33673 it looks like parts of these tests have been reworked and
there is no intention to fix the remains on 7.x, so I think we can remove the
entire test.
The task that TaskManager#register returns cannot be null. The method
enforces that it is not null after calling request#createTask. It is
then needless to check for null in the listener later. Also, added the
call to the delegate listener in a finally block, just to make sure.
* We shouldn't be recreating wrapped REST handlers over and over for every request. We only use this hook in x-pack and the wrapper there does not have any per request state.
This is inefficient and could lead to some very unexpected memory behavior
=> I made the logic create the wrapper on handler registration and adjusted the x-pack wrapper implementation to correctly forward the circuit breaker and content stream flags
When finding nodes in a connected cluster for cross cluster
search the requests to get cluster state on the connected
cluster should be made in the system context because
logically they are equivalent to checking a single detail
in the local cluster state and should not require that the
user who made the request that is using this method in its
implementation is authorized to view the entire cluster
state.
Fixes#43974
Today closing a `ClusterNode` in an `AbstractCoordinatorTestCase` uses
`onNode()` so has no effect if the node is not in the current list of nodes.
It also discards the `Runnable` it creates without having run it, so has no
effect anyway.
This commit makes these tests much stricter about properly closing the nodes
started during `Coordinator` tests, by tracking the persisted states that are
opened, and adds an assertion to catch the trappy requirement that the closing
node still belongs to the cluster.
This commit fixes a bug when a deferred aggregator tries to early terminate the collection. In such case the CollectionTerminatedException is not caught and
the search fails on the shard. This change makes sure that we catch the exception in order to continue the deferred collection on the next leaf.
Fixes#44909
Replaying operations from the local translog must never fail as those
operations were processed successfully on the primary before and the
mapping is up to update already. This change removes leniency during
resetting engine from translog in IndexShard and InternalEngine.
This is a temporary fix during the Joda to Java datetime transition. This will
implicitly cast a JodaCompatibleZonedDateTime to a ZonedDateTime for
both def and static types. This is necessary to insulate users from needing
to know about JodaCompatibleZonedDateTime explicitly.
TaskListener accepts today Throwable in its onFailure method. Though
looking at where it is called (TransportAction), it can never be
notified of a Throwable.
This commit changes the signature of TaskListener#onFailure so that it
accepts an `Exception` rather than a `Throwable` as second argument.
We currently block the transport thread on startup, which has caused test failures. I think this is
some kind of deadlock situation. I don't think we should even block a transport thread, and
there's also no need to do so. We can just reject requests as long we're not fully set up. Note
that the HTTP layer is only started much later (after we've completed full start up of the
transport layer), so that one should be completely unaffected by this.
Closes#41745
Fixes an issue where a call to openConnection was not properly guarded, allowing an exception
to bubble up to the uncaught exception handler, causing test failures.
Closes#44912
SimpleClusterStateIT testIndicesOptions failed in #44817 because it tries to close
an index at the beginning of the test. With random index settings, it is possible that
the index has a high number of shards (10) and replicas (1), which means that on
CI this index can take time to be fully allocated.
The close index request can fail in the case where replicas are still recovering operations.
Thiscommit adds a simple ensureGreen() at the beginning of the test to be sure that all
replicas are started before trying to close the index.
closes#44817
The test ShrinkIndexIT.testShrinkThenSplitWithFailedNode sometimes fails
because the resize operation is not acknowledged (see #44736). This resize
operation creates a new index "splitagain" and it results in a cluster state
update (TransportResizeAction uses MetaDataCreateIndexService.createIndex()
to create the resized index). This cluster state update is expected to be
acknowledged by all nodes (see IndexCreationTask.onAllNodesAcked()) but
this is not always true: the data node that was just stopped in the test before
executing the resize operation might still be considered as a "faulty" node
(and not yet removed from the cluster nodes) by the FollowersChecker. The
cluster state is then acked on all nodes but one, and it results in a non
acknowledged resize operation.
This commit adds an ensureStableCluster() check after stopping the node in
the test. The goal is to ensure that the data node has been correctly removed
from the cluster and that all nodes are fully connected to each before moving
forward with the resize operation.
Closes#44736
Today the processors setting is permitted to be set to more than the
number of processors available to the JVM. The processors setting
directly sizes the number of threads in the various thread pools, with
most of these sizes being a linear function in the number of
processors. It doesn't make any sense to set processors very high as the
overhead from context switching amongst all the threads will overwhelm,
and changing the setting does not control how many physical CPU
resources there are on which to schedule the additional threads. We have
to draw a line somewhere and this commit deprecates setting processors
to more than the number of available processors. This is the right place
to draw the line given the linear growth as a function of processors in
most of the thread pools, and that some are capped at the number of
available processors already.
With this change, we will return primary_term and seq_no of the current
document if an update is detected as a noop. We already return the
version; hence we should also return seq_no and primary_term.
Relates #42497
Adds an API to clone an index. This is similar to the index split and shrink APIs, just with the
difference that the number of primary shards is kept the same. In case where the filesystem
provides hard-linking capabilities, this is a very cheap operation.
Indexing cloning can be done by running `POST my_source_index/_clone/my_target_index` and it
supports the same options as the split and shrink APIs.
Closes#44128
While joda no longer exists in the apis for 7.x, the compatibility layer
still exists with helper methods mimicking the behavior of joda for
ZonedDateTime objects returned for date fields in scripts. This layer
was originally intended to be removed in 7.0, but is now likely to exist
for the lifetime of 7.x.
This commit adds missing methods from ChronoZonedDateTime to the compat
class. These methods were not part of joda, but are needed to act like a
real ZonedDateTime.
relates #44411
This PR makes two changes to FetchSourceSubPhase when _source is disabled and
we're in a nested context:
* If no source filters are provided, return early to avoid an NPE.
* If there are source filters, make sure to throw an exception.
The behavior was chosen to match what currently happens in a non-nested context.
Refactors GeoShapeQueryBuilder to derive from a new AbstractGeometryQueryBuilder that provides common parsing and build logic for spatial geometries. This will allow development of custom geometry queries by extending AbstractGeometryQueryBuilder preventing duplication of common spatial query logic.
The problem is that RemoteClusterConnection closes the connection manager asynchronously, which races with the threadpool being shutdown at the end of the test.
Closes#44339Closes#44610
Removes unnecessary now timeline decompositions from shape builders
and deprecates ShapeBuilders in QueryBuilder in favor of libs/geo
shapes.
Relates to #40908
This test can fail (super-rarely) if it generates a list of length 11
containing a duplicate, because the `.distinct()` reduces the list length to 10
and then it is not abbreviated any more. This change generalises the test to
cover lists of any random length.
* In both fake connection validators we were potentially executing the listener twice. This lead to the situation that the locking via `connectionLock` that ensures that each listener is only executed once ever
would fail and the lister would run twice (in which case the listeners for that node are already `null` and we get an NPE)
* The fact that two different tests fail is due to the fact that we weren't safely shutting down the threadpool which meant the the task that trips the assertion (on the generic pool) would leak into the next test and fail it
* Closes#44758
We have some old permissions lying around, granted to untrusted code
from the days of yore when we supported Groovy and Javascript
scripting. This commit removes these stale permissions.
Today our systemd service defaults to a service type of simple. This
means that systemd assumes Elasticsearch is ready as soon as the
ExecStart (bin/elasticsearch) process is forked off. This means that the
service appears ready long before it actually is, so before it is ready
to receive requests. It also means that services that want to depend on
Elasticsearch being ready to start can not as there is not a reliable
mechanism to determine this. This commit changes the service type to
notify. This requires that Elasticsearch sends a notification message
via libsystemd sd_notify method. This commit does that by using JNA to
invoke this native method. Additionally, we use this integration to also
notify systemd when we are stopping.
* Fix this test randomly failing when running into async translog persistence edge case and failing to successfully close index
* Also, slightly improve debug logging on close failure
* Closes#44681
Switches to more robust way of generating random test geometries by
reusing lucene's GeoTestUtil. Removes duplicate random geometry
generators by moving them to the test framework.
Closes#37278
The `elasticsearch-node override-version` command fails if it cannot read the
existing node metadata file. However, it reads this file strictly and fails if
there are any unknown fields, which means it will not be useful if we add
another field in future.
This commit adds leniency to this command, allowing it to ignore any unknown
fields and proceed with the downgrade. A downgrade is already unsafe, and the
user is already copiously warned about this, so being lenient in this case does
not make things much worse.
Today when creating an index and checking cluster shard limits, we check
the number of shards before applying index templates. At this point, we
do not know the actual number of shards that will be used to create the
index. In a case when the defaults are used and a template would
override, we could be grossly underestimating the number of shards that
would be created, and thus incorrectly applying the limits. This commit
addresses this by checking the shard limits after applying index
templates.
We often start testing with early access versions of new Java
versions and this have caused minor issues in our tests
(i.e. #43141) because the version string that the JVM reports
cannot be parsed as it ends with the string -ea.
This commit changes how we parse and compare Java versions to
allow correct parsing and comparison of the output of java.version
system property that might include an additional alphanumeric
part after the version numbers
(see [JEP 223[(https://openjdk.java.net/jeps/223)). In short it
handles a version number part, like before, but additionally a
PRE part that matches ([a-zA-Z0-9]+).
It also changes a number of tests that would attempt to parse
java.specification.version in order to get the full version
of Java. java.specification.version only contains the major
version and is thus inappropriate when trying to compare against
a version that might contain a minor, patch or an early access
part. We know parse java.version that can be consistently
parsed.
Resolves#43141
This commit removes the method AllocationService.reroute(ClusterState, String, boolean)
in favor of AllocationService.reroute(ClusterState, String).
Motivations are:
there are already 3 other reroute methods in this class
this method is always called with the debug parameter set to false
almost all tests use the method reroute(ClusterState, String)
* As a result of #44665 the collections returned by the deserialization methods on `StreamInput` may be either mutable or immutable now,
this PR adds documentation for that fact
In 7.x we cannot start a new master-eligible node before the cluster has formed
since we first try and update minimum_master_nodes and this is blocked. This
commit changes the test to start a data-only node so that no such adjustment is
necessary.
Relates #44685
Add more logging to indexRandom
Seems that asynchronous indexing from indexRandom sometimes indexes
the same document twice, which will mess up the expected score calculations.
For example, indexing:
{ "index" : {"_id" : "1" } }
{"important" :"phrase match", "less_important": "nothing important"}
{ "index" : {"_id" : "2" } }
{"important" :"nothing important", "less_important" :"phrase match"}
Produces the expected scores: 13.8 for doc1, and 1.38 for doc2
indexing:
{ "index" : {"_id" : "1" } }
{"important" :"phrase match", "less_important": "nothing important"}
{ "index" : {"_id" : "2" } }
{"important" :"nothing important", "less_important" :"phrase match"}
{ "index" : {"_id" : "3" } }
{"important" :"phrase match", "less_important": "nothing important"}
Produces scores: 9.4 for doc1, and 1.96 for doc2 which are found in the
error logs.
Relates to #43144
Fields in JSON logs should be an escaped JSON fields. It is a broken json value at the moment
"stats": "["group1", "group2"]", -> "stats": "[\"group1\", \"group2\"]",
This should later be refactored into a JSON array of strings (the same as types in 7.x)
Today we block access to the pending tasks API before the cluster has recovered
its state. There's no real need to do so, and the master does meaningful work
even before performing state recovery so it might sometimes be useful to allow
access to this API. This commit changes this API to ignore all cluster blocks.
Fixes#44652
The field has to be defined in log4j2.properties and should be an
escaped JSON for now (it is a broken JSON at the moment). This should later be refactored into a JSON array
of strings.
Deprecation logger was filtering log entries by key, that means that if two log messages with the same key are logged from different users, then the second log messages will be filtered.
This change allows to log deprecation message with the same key by different users.
relates #41354
backport #44587
* It seems redundant to synchronize here and check that the map hasn't checked via the `isRunning` under the mutex
* The map won't change if under the mutex that locks on all the updates to it
* Without the mutex it's very unlikely to change inside the method call relative to the likelihood of changing until the generic pool where we check for `isRunning` again anyway
-> just remove the synchronization (it's on the IO loop) and check since we do check the running state on the generic pool under the mutex anyway when we actually fail it
We have to copy the field names otherwise we either have a handle of a
list that a caller might mutate or we might mutate when they aren't
expecting it, or worse, a handle of a list that is not mutable (and we
end up mutating the list).
Relates #44665
* Mute failing test
tracked in #44552
* mute EvilSecurityTests
tracking in #44558
* Fix line endings in ESJsonLayoutTests
* Mute failing ForecastIT test on windows
Tracking in #44609
* mute BasicRenormalizationIT.testDefaultRenormalization
tracked in #44613
* fix mute testDefaultRenormalization
* Increase busyWait timeout windows is slow
* Mute failure unconfigured node name
* mute x-pack internal cluster test windows
tracking #44610
* Mute JvmErgonomicsTests on windows
Tracking #44669
* mute SharedClusterSnapshotRestoreIT testParallelRestoreOperationsFromSingleSnapshot
Tracking #44671
* Mute NodeTests on Windows
Tracking #44256
* We only had the `size == 0` optimization in some but not all spots of deserializing collections in this class, fixed the remaining spots.
* Also fixed the a similar spot when deserializing `ThreadContextStruct` that could now be simplified (it was apparently doing it's own version of this optimization for the first map it deserialized before ... but not for the second map -> made it not instantiate anything if both maps are empty since it's always the same object here anyway)
* Optimize some StreamOutput Operations
* Writing numbers byte by byte adds a lot of unnecessary bounds checks to serialization
* Serializing to a threadlocal `byte[]` instead and bulk writing gives about a 50% speedup on `long` and `vlong` (for large numbers) writes and 30% for `int`, `vint` on Linux on an i9
* Using a threadlocal of the maximum string buffer size we used to allocate before also removes allocations when writing strings in general since we now never have to allocate a `byte[]` for that
* And don't have to GC one either resolving the TODO removed here
David Turner
Tanguy Leroux
Jason Tedor
Yannick Welsch
Igor Motov
Zachary Tong
Nhat Nguyen
Armin Braun
Tim Brooks
Nikita Glashenko
Jim Ferenczi
William Brafford
Hendrik Muhs
Christoph Büscher
Luca Cavanna
Andrey Ershov
David Roberts
Mayya Sharipova
Jake Landis
David Kyle
Jack Conradson
Michał Perlak
Ignacio Vera
Ryan Ernst
Julie Tibshirani
Nicholas Knize
Dimitris Athanasiou
Ioannis Kakavas
Evgenia Badiyanova
Przemyslaw Gomulka
Alpar Torok
Tal Levy