We need to limit the search request aggregations to whole multiples
of the configured interval for both histogram and date_histogram.
Otherwise, agg buckets won't overlap with the rolled up buckets
and the results will be incorrect.
For histogram, the validation is very simple: request must be >= the config,
and modulo evenly.
Dates are more tricky.
- If both request and config are fixed dates, we can convert to millis
and treat them just like the histo
- If both are calendar, we make sure the request is >= the config with
a static lookup map that ranks the calendar values relatively. All
calendar units are "singles", so they are evenly divisible already
- We disallow any other combination (one fixed, one calendar, etc)
This extracts a super class out of the rollup indexer called the AsyncTwoPhaseIterator.
The implementor of it can define the query, transformation of the response,
indexing and the object to persist the position/state of the indexer.
The stats object used by the indexer to record progress is also now abstract, allowing
the implementation provide custom stats beyond what the indexer provides. It also
allows the implementation to decide how the stats are presented (leaves toXContent()
up to the implementation).
This should allow new projects to reuse the search-then-index persistent task that Rollup
uses, but without the restrictions/baggage of how Rollup has to work internally to
satisfy time-based rollups.
We don't allow the user to configure a rollup index against an
existing index, but the exceptions that we return are not clear about
that. They indicate issues with metadata, instead of stating
the real reason (not allowed to use a non-rollup index to store
rollup data).
This makes the exception better, and adds a bit more testing
This committ removes the getMetadata() methods from the DateHistoGroupConfig
and HistoGroupConfig objects. This way the configuration objects do not rely on RollupField.formatMetaField() anymore and do not expose a getMetadata()
method that is tighlty coupled to the rollup indexer.
If a search request doesn't contain aggs (or an empty agg object),
we should just retun an empty response. This is how the normal search
API works if you specify zero hits and empty aggs.
The existing behavior throws an exception because it tries to send
an empty msearch.
Closes#32256
This reworks how we configure the `shadow` plugin in the build. The major
change is that we no longer bundle dependencies in the `compile` configuration,
instead we bundle dependencies in the new `bundle` configuration. This feels
more right because it is a little more "opt in" rather than "opt out" and the
name of the `bundle` configuration is a little more obvious.
As an neat side effect of this, the `runtimeElements` configuration used when
one project depends on another now contains exactly the dependencies needed
to run the project so you no longer need to reference projects that use the
shadow plugin like this:
```
testCompile project(path: ':client:rest-high-level', configuration: 'shadow')
```
You can instead use the much more normal:
```
testCompile "org.elasticsearch.client:elasticsearch-rest-high-level-client:${version}"
```
Subclasses of `EsIntegTestCase` run multiple Elasticsearch nodes in the
same JVM and when we log we look at the name of the thread to figure out
the node name. This makes sure that all calls to `daemonThreadFactory`
include the node name.
Closes#32574
I'd like to follow this up with more drastic changes that make it
impossible to do this incorrectly but that change is much larger than
this and I'd like to get these log lines fixed up sooner rather than
later.
This removes custom Response classes that extend `AcknowledgedResponse` and do nothing, these classes are not needed and we can directly use the non-abstract super-class instead.
While this appears to be a large PR, no code has actually changed, only class names have been changed and entire classes removed.
Previously, we were using a simple CRC32 for the IDs of rollup documents.
This is a very poor choice however, since 32bit IDs leads to collisions
between documents very quickly.
This commit moves Rollups over to a 128bit ID. The ID is a concatenation
of all the keys in the document (similar to the rolling CRC before),
hashed with 128bit Murmur3, then base64 encoded. Finally, the job
ID and a delimiter (`$`) are prepended to the ID.
This gurantees that there are 128bits per-job. 128bits should
essentially remove all chances of collisions, and the prepended
job ID means that _if_ there is a collision, it stays "within"
the job.
BWC notes:
We can only upgrade the ID scheme after we know there has been a good
checkpoint during indexing. We don't rely on a STARTED/STOPPED
status since we can't guarantee that resulted from a real checkpoint,
or other state. So we only upgrade the ID after we have reached
a checkpoint state during an active index run, and only after the
checkpoint has been confirmed.
Once a job has been upgraded and checkpointed, the version increments
and the new ID is used in the future. All new jobs use the
new ID from the start
Same motivation as #32507 but for the DateHistogramGroupConfig
configuration object. This pull request also changes the format of the
time zone from a Joda's DateTimeZone to a simple String.
It should help to port the API to the high level rest client and allows
clients to not be forced to use the Joda Time library. Serialization is
impacted but does not need a backward compatibility layer as
DateTimeZone are serialized as String anyway. XContent also expects
a String for timezone, so I found it easier to move everything to String.
Related to #29827
While working on adding the Create Rollup Job API to the
high level REST client (#29827), I noticed that the configuration
objects like TermsGroupConfig rely on the Builder pattern in
order to create or parse instances. These builders are doing
some validation but the same validation could be done within
the constructor itself or on the server side when appropriate.
This commit removes the builder for TermsGroupConfig,
removes some other methods that I consider not really usefull
once the TermsGroupConfig object will be exposed in the
high level REST client. It also simplifies the parsing logic.
Related to #29827
Java 11 uses more verbose exceptions messages, causing this assertion
to fail. Changed the test to be less restrictive and only look
for the classes we care about.
This bundles the x-pack:protocol project into the x-pack:plugin:core
project because we'd like folks to consider it an implementation detail
of our build rather than a separate artifact to be managed and depended
on. It is now bundled into both x-pack:plugin:core and
client:rest-high-level. To make this work I had to fix a few things.
Firstly, I had to make PluginBuildPlugin work with the shadow plugin.
In that case we have to bundle only the `shadow` dependencies and the
shadow jar.
Secondly, every reference to x-pack:plugin:core has to use the `shadow`
configuration. Without that the reference is missing all of the
un-shadowed dependencies. I tried to make it so that applying the shadow
plugin automatically redefines the `default` configuration to mirror the
`shadow` configuration which would allow us to use bare project references
to the x-pack:plugin:core project but I couldn't make it work. It'd *look*
like it works but then fail for transitive dependencies anyway. I think
it is still a good thing to do but I don't have the willpower to do it
now.
Finally, I had to fix an issue where Eclipse and IntelliJ didn't properly
reference shadowed transitive dependencies. Neither IDE supports shadowing
natively so they have to reference the shadowed projects. We fix this by
detecting `shadow` dependencies when in "Intellij mode" or "Eclipse mode"
and adding `runtime` dependencies to the same target. This convinces
IntelliJ and Eclipse to play nice.
Adds a new single-value metrics aggregation that computes the weighted
average of numeric values that are extracted from the aggregated
documents. These values can be extracted from specific numeric
fields in the documents.
When calculating a regular average, each datapoint has an equal "weight"; it
contributes equally to the final value. In contrast, weighted averages
scale each datapoint differently. The amount that each datapoint contributes
to the final value is extracted from the document, or provided by a script.
As a formula, a weighted average is the `∑(value * weight) / ∑(weight)`
A regular average can be thought of as a weighted average where every value has
an implicit weight of `1`.
Closes#15731
The rollup indexer uses a range query to select the next page
of results based on the last time bucket of the previous round
and the `delay` configured on the rollup job. This query uses
the `epoch_millis` format implicitly but doesn't set the `format`.
This result in errors during the rollup job if the field
definition doesn't allow this format. It can also miss documents
if the format is not accepted but another format in the field
definition is able to parse the query (e.g.: `epoch_second`).
This change ensures that we use `epoch_millis` as the only format
to parse the rollup range query.
This introduces a new GetRollupIndexCaps API which allows the user to retrieve rollup capabilities of a specific rollup index (or index pattern). This is distinct from the existing RollupCaps endpoint.
- Multiple jobs can be stored in multiple indices and point to a single target data index pattern (logstash-*). The existing API finds capabilities/config of all jobs matching that data index pattern.
- One rollup index can hold data from multiple jobs, targeting multiple data index patterns. This new API finds the capabilities based on the concrete rollup indices.
The old RollupIT was a node IT, an flaky for a number of reasons.
This new version is an ESRestTestCase and should be a little more robust.
This was added to the multi-node QA tests as that seemed like the most
appropriate location. It didn't seem necessary to create a whole new
QA module.
Note: The only test that was ported was the "Big" test for validating
a larger dataset. The rest of the tests are represented in existing
yaml tests.
Closes#31258Closes#30232
Related to #30290
We can leverage the composite agg's new `missing_bucket` feature on
terms groupings. This means the aggregation criteria used in the indexer
will now return null buckets for missing keys.
Because all buckets are now returned (even if a key is null),
we can guarantee correct doc counts with
"combined" jobs (where a job rolls up multiple schemas). This was
previously impossible since composite would ignore documents that
didn't have _all_ the keys, meaning non-overlapping schemas would
cause composite to return no buckets.
Note: date_histo does not use `missing_bucket`, since a timestamp is
always required.
The docs have been adjusted to recommend a single, combined job. It
also makes reference to the previous issue to help users that are upgrading
(rather than just deleting the sections).
* Remove deprecation warnings to prepare for Gradle 5
Gradle replaced `project.sourceSets.main.output.classesDir` of type
`File` with `project.sourceSets.main.output.classesDirs` of type
`FileCollection`
(see [SourceSetOutput](https://github.com/gradle/gradle/blob/master/subprojects/plugins/src/main/java/org/gradle/api/tasks/SourceSetOutput.java))
Build output is now stored on a per language folder.
There are a few places where we use that, here's these and how it's
fixed:
- Randomized Test execution
- look in all test folders ( pass the multi dir configuration to the
ant runner )
- DRY the task configuration by introducing `basedOn` for
`RandomizedTestingTask` DSL
- Extend the naming convention test to support passing in multiple
directories
- Fix the standalon test plugin, the dires were not passed trough,
checked with a debuger and the statement had no affect due to a
missing `=`.
Closes#30354
* Only check Java tests, PR feedback
- Name checker was ran for Groovy tests that don't adhere to the same
convections causing the check to fail
- implement PR feedback
* Replace `add` with `addAll`
This worked because the list is passed to `project.files` that does the
right thing.
* Revert "Only check Java tests, PR feedback"
This reverts commit 9bd9389875d8b88aadb50df57a45cd0d2b073241.
* Remove `basedOn` helper
* Bring some changes back
Previus revert accidentally reverted too much
* Fix negation
* add back public
* revert name check changes
* Revert "revert name check changes"
This reverts commit a2800c0b363168339ea65e2a79ec8256e5883e6d.
* Pass all dirs to name check
Only run on Java for build-tools, this is safe because it's a self test.
It needs more work before we could pass in the Groovy classes as well as
these inherit from `GroovyTestCase`
* remove self tests from name check
The self complicates the task setup and disable real checks on
build-tools.
With this change there are no more self tests, and the build-tools tests
adhere to the conventions.
The self test will be replaced by gradle test kit, thus the addition of
the Gradle plugin builder plugin.
* First test to run a Gradle build
* Add tests that replace the name check self test
* Clean up integ test base class
* Always run tests
* Align with test naming conventions
* Make integ. test case inherit from unit test case
The check requires this
* Remove `import static org.junit.Assert.*`
* Move to Gradle 4.8 RC1
* Use latest version of plugin
The current does not work with Gradle 4.8 RC1
* Switch to Gradle GA
* Add and configure build compare plugin
* add work-around for https://github.com/gradle/gradle/issues/5692
* work around https://github.com/gradle/gradle/issues/5696
* Make use of Gradle build compare with reference project
* Make the manifest more compare friendly
* Clear the manifest in compare friendly mode
* Remove animalsniffer from buildscript classpath
* Fix javadoc errors
* Fix doc issues
* reference Gradle issues in comments
* Conditionally configure build compare
* Fix some more doclint issues
* fix typo in build script
* Add sanity check to make sure the test task was replaced
Relates to #31324. It seems like Gradle has an inconsistent behavior and
the taks is not always replaced.
* Include number of non conforming tasks in the exception.
* No longer replace test task, create implicit instead
Closes#31324. The issue has full context in comments.
With this change the `test` task becomes nothing more than an alias for `utest`.
Some of the stand alone tests that had a `test` task now have `integTest`, and a
few of them that used to have `integTest` to run multiple tests now only
have `check`.
This will also help separarate unit/micro tests from integration tests.
* Revert "No longer replace test task, create implicit instead"
This reverts commit f1ebaf7d93e4a0a19e751109bf620477dc35023c.
* Fix replacement of the test task
Based on information from gradle/gradle#5730 replace the task taking
into account the task providres.
Closes#31324.
* Only apply build comapare plugin if needed
* Make sure test runs before integTest
* Fix doclint aftter merge
* PR review comments
* Switch to Gradle 4.8.1 and remove workaround
* PR review comments
* Consolidate task ordering
This pull request adds a full cluster restart test for a Rollup job.
The test creates and starts a Rollup job on the cluster and checks
that the job already exists and is correctly started on the upgraded
cluster.
This test allows to test that the persistent task state is correctly
parsed from the cluster state after the upgrade, as the status field
has been renamed to state in #31031.
The test undercovers a ClassCastException that can be thrown in
the RollupIndexer when the timestamp as a very low value that fits
into an integer. When it's the case, the value is parsed back as an
Integer instead of Long object and (long) position.get(rollupFieldName)
fails.
TransportAction currently contains 2 doExecute methods, one which takes
a the task, and one that does not. The latter is what some subclasses
implement, while the first one just calls the latter, dropping the given
task. This commit combines these methods, in favor of just always
assuming a task is present.
TransportRequestHandler currently contains 2 messageReceived methods,
one which takes a Task, and one that does not. The first just delegates
to the second. This commit changes all existing implementors of
TransportRequestHandler to implement the version which takes Task, thus
allowing the class to be a functional interface, and eliminating the
need to throw exceptions when a task needs to be ensured.
Most transport actions don't need the node ThreadPool. This commit
removes the ThreadPool as a super constructor parameter for
TransportAction. The actions that do need the thread pool then have a
member added to keep it from their own constructor.
Most transport actions don't need to resolve index names. This commit
removes the index name resolver as a super constructor parameter for
TransportAction. The actions that do need the resolver then have a
member added to keep the resolver from their own constructor.
This pull request removes the relationship between the state
of persistent task (as stored in the cluster state) and the status
of the task (as reported by the Task APIs and used in various
places) that have been confusing for some time (#29608).
In order to do that, a new PersistentTaskState interface is added.
This interface represents the persisted state of a persistent task.
The methods used to update the state of persistent tasks are
renamed: updatePersistentStatus() becomes updatePersistentTaskState()
and now takes a PersistentTaskState as a parameter. The
Task.Status type as been changed to PersistentTaskState in all
places were it make sense (in persistent task customs in cluster
state and all other methods that deal with the state of an allocated
persistent task).
We should not allow the user to configure index patterns that also match
the index which stores the rollup index.
For example, it is quite natural for a user to specify `metricbeat-*`
as the index pattern, and then store the rollups in `metricbeat-rolled`.
This will start throwing errors as soon as the rollup index is created
because the indexer will try to search it.
Note: this does not prevent the user from matching against existing
rollup indices. That should be prevented by the field-level validation
during job creation.
Otherwise we could end up with persistent tasks metadata in the cluster that some of the nodes
might not understand in case where the cluster is during rolling upgrade from the default 6.2 to the
default 6.3 distribution.
Follow-up to #30743
This commit renames methods in the PersistentTasksService, to
make obvious that the methods send requests in order to change
the state of persistent tasks.
Relates to #29608.
* Refactors ClientHelper to combine header logic
This change removes all the `*ClientHelper` classes which were
repeating logic between plugins and instead adds
`ClientHelper.executeWithHeaders()` and
`ClientHelper.executeWithHeadersAsync()` methods to centralise the
logic for executing requests with stored security headers.
* Removes Watcher headers constant
When validating the search request, we make sure any date_histogram
aggregations have timezones that match the jobs. But we didn't
do any such validation on range queries.
While it wouldn't produce incorrect results, it would be confusing
to the user as no documents would match the aggregation (because we
add a filter clause on the timezone for the agg).
Now the user gets an exception up front, and some helpful text about
why the range query didnt match, and which timezones are acceptable
Zachary Tong
Hendrik Muhs
Tanguy Leroux
Nik Everett
Lee Hinman
Jim Ferenczi
Vladimir Dolzhenko
Alpar Torok
Ryan Ernst
Yannick Welsch
Colin Goodheart-Smithe