* Parallel indexing single dim partitions
Implements single dimension range partitioning for native parallel batch
indexing as described in #8769. This initial version requires the
druid-datasketches extension to be loaded.
The algorithm has 5 phases that are orchestrated by the supervisor in
`ParallelIndexSupervisorTask#runRangePartitionMultiPhaseParallel()`.
These phases and the main classes involved are described below:
1) In parallel, determine the distribution of dimension values for each
input source split.
`PartialDimensionDistributionTask` uses `StringSketch` to generate
the approximate distribution of dimension values for each input
source split. If the rows are ungrouped,
`PartialDimensionDistributionTask.UngroupedRowDimensionValueFilter`
uses a Bloom filter to skip rows that would be grouped. The final
distribution is sent back to the supervisor via
`DimensionDistributionReport`.
2) The range partitions are determined.
In `ParallelIndexSupervisorTask#determineAllRangePartitions()`, the
supervisor uses `StringSketchMerger` to merge the individual
`StringSketch`es created in the preceding phase. The merged sketch is
then used to create the range partitions.
3) In parallel, generate partial range-partitioned segments.
`PartialRangeSegmentGenerateTask` uses the range partitions
determined in the preceding phase and
`RangePartitionCachingLocalSegmentAllocator` to generate
`SingleDimensionShardSpec`s. The partition information is sent back
to the supervisor via `GeneratedGenericPartitionsReport`.
4) The partial range segments are grouped.
In `ParallelIndexSupervisorTask#groupGenericPartitionLocationsPerPartition()`,
the supervisor creates the `PartialGenericSegmentMergeIOConfig`s
necessary for the next phase.
5) In parallel, merge partial range-partitioned segments.
`PartialGenericSegmentMergeTask` uses `GenericPartitionLocation` to
retrieve the partial range-partitioned segments generated earlier and
then merges and publishes them.
* Fix dependencies & forbidden apis
* Fixes for integration test
* Address review comments
* Fix docs, strict compile, sketch check, rollup check
* Fix first shard spec, partition serde, single subtask
* Fix first partition check in test
* Misc rewording/refactoring to address code review
* Fix doc link
* Split batch index integration test
* Do not run parallel-batch-index twice
* Adjust last partition
* Split ITParallelIndexTest to reduce runtime
* Rename test class
* Allow null values in range partitions
* Indicate which phase failed
* Improve asserts in tests
* Address security vulnerabilities CVSS >= 7
Update dependencies to address security vulnerabilities with CVSS scores
of 7 or higher. A new Travis CI job is added to prevent new
high/critical security vulnerabilities from being added.
Updated dependencies:
- api-util 1.0.0 -> 1.0.3
- jackson 2.9.10 -> 2.10.1
- kafka 2.1.0 -> 2.1.1
- libthrift 0.10.0 -> 0.13.0
- protobuf 3.2.0 -> 3.11.0
The following high/critical security vulnerabilities are currently
suppressed (so that the new Travis CI job can be added now) and are left
as future work to fix:
- hibernate-validator:5.2.5
- jackson-mapper-asl:1.9.13
- libthrift:0.6.1
- netty:3.10.6
- nimbus-jose-jwt:4.41.1
* Rename EDL1 license file
* Fix inspection errors
* Fix dependency analyze warnings
Update the maven dependency plugin to the latest version and fix all
warnings for unused declared and used undeclared dependencies in the
compile scope. Added new travis job to add the check to CI. Also fixed
some source code files to use the correct packages for their imports and
updated druid-forbidden-apis to prevent regressions.
* Address review comments
* Adjust scope for org.glassfish.jaxb:jaxb-runtime
* Fix dependencies for hdfs-storage
* Consolidate netty4 versions
* set encoding for license and notice scripts, split generate-license.py into generate-binary-license.py and check-licenses.py, check-licenses when -Papache-release is used
* missing docs
* doc fix
* more doc fix
* remove comments
* good catch travis +1
* fix lgtm alerts
* enable unit tests with JDK11
This enables unit tests with openjdk11, splitting up the build into
stages to have it fail faster
The integration test docker image still uses openjdk8, so there is
little reason to run those tests with JDK11 yet
* remove stages
* Use Codecov
Upload coverage reports to Codecov. For now, having Codecov comment on
PRs or enforcing a minimum coverage threshold are both disabled until
the Codecov coverage reports look reliable:
https://codecov.io/gh/apache/incubator-druid
* Split bash and curl into separate lines
* Speed up javascript Travis CI jobs
Skip mvn install for javascript CI jobs since it is not needed.
* Specify base filepath for source files
* Remove coveralls
The coveralls code coverage reports inaccurate coverage for our parallel
builds. Disable it until it can be fixed or a better alternative can be
found.
* migrate binary notice entries to live in licenses.yaml, use licenses.yaml and NOTICE to generate NOTICE.BINARY at distribution time
* +x
* move release scripts to distribution/bin, fixup notice script, trim dependencies for avro and kerberos in licenses.yaml
* add missing hdfs-storage dependencies
* revert to old syntax, fixes
* formatting
* update notices for recently updated dependencies
* Enable code coverage
Code coverage was disabled via
https://github.com/apache/incubator-druid/pull/3122 due to an issue with
cobertura in Travis CI. Switch code coverage tool from cobertura to
jacoco to avoid issue and re-enable coveralls for Travis CI.
* Exclude non-production code
* Exclude benchmark generated code
* Exclude DruidTestRunnerFactory
The package check job sometimes hits the 50 minute Travis CI job time
limit. Move license checking tasks in "package check" job to "license
check" job to rebalance the job runtime (the "license check" job
currently takes about 1 minute). Moving the logic from build.sh to
.travis.yml also gives more visibility into how long each step takes
(i.e., generate-license-dependency-reports.py vs generate-license.py).
After enabling parallel builds for "mvn install", the sigar dependency
would sometimes resolve to the incorrect artifact repo for some of the
maven modules. This issue seems to be fixed by moving the definition of
the sigar dependency's artifact repo to the root POM.
Also, depending on network speeds, "mvn -q install" may take longer than
the default 10 minute timeout to print any output. Use travis_wait to
extend the timeout to 15 minutes.
Reorganize Travis CI jobs into smaller faster (and more) jobs. Add
various maven options to skip unnecessary work and refactored Travis CI
job definitions to follow DRY.
Detailed changes:
.travis.yml
- Refactor build logic to get rid of copy-and-paste logic
- Skip static checks and enable parallelism for maven install
- Split static analysis into different jobs to ease triage
- Use "name" attribute instead of NAME environment variable
- Split "indexing" and "web console" out of "other modules test"
- Split 2 integration test jobs into multiple smaller jobs
build.sh
- Enable parallelism
- Disable more static checks
travis_script_integration.sh
travis_script_integration_part2.sh
integration-tests/README.md
- Use TestNG groups instead of shell scripts and move definition of jobs
into Travis CI yaml
integration-tests/pom.xml
- Show elapsed time of individual tests to aid in future rebalancing of
Travis CI integration test jobs run time
TestNGGroup.java
- Use TestNG groups to make it easy to have multiple Travis CI
integration test jobs. TestNG groups also make it easier to have an
"other" integration test group and make it less likely a test will
accidentally not be included in a CI job.
IT*Test.java
AbstractITBatchIndexTest.java
AbstractKafkaIndexerTest.java
- Add TestNG group
- Fix various IntelliJ inspection warnings
- Reduce scope of helper methods since the TestNG group annotation on
the class makes TestNG consider all public methods as test methods
pom.xml
- Allow enforce plugin to be run from command-line
- Bump resources plugin version so that "[debug] execute contextualize"
output is correctly suppressed by "mvn -q"
- Bump exec plugin version so that skip property is renamed from "skip"
to "exec.skip"
web-console/pom.xml
- Add property to allow disabling javascript-related work. This property
is overridden in Travis CI to speed up the jobs.
* Fix dependency analyze warnings
Update the maven dependency plugin to the latest version and fix all
warnings for unused declared and used undeclared dependencies in the
compile scope. Added new travis job to add the check to CI. Also fixed
some source code files to use the correct packages for their imports.
* Fix licenses and dependencies
* Fix licenses and dependencies again
* Fix integration test dependency
* Address review comments
* Fix unit test dependencies
* Fix integration test dependency
* Fix integration test dependency again
* Fix integration test dependency third time
* Fix integration test dependency fourth time
* Fix compile error
* Fix assert package
* Benchmarks: New SqlBenchmark, add caching & vectorization to some others.
- Introduce a new SqlBenchmark geared towards benchmarking a wide
variety of SQL queries. Rename the old SqlBenchmark to
SqlVsNativeBenchmark.
- Add (optional) caching to SegmentGenerator to enable easier
benchmarking of larger segments.
- Add vectorization to FilteredAggregatorBenchmark and GroupByBenchmark.
* Query vectorization.
This patch includes vectorized timeseries and groupBy engines, as well
as some analogs of your favorite Druid classes:
- VectorCursor is like Cursor. (It comes from StorageAdapter.makeVectorCursor.)
- VectorColumnSelectorFactory is like ColumnSelectorFactory, and it has
methods to create analogs of the column selectors you know and love.
- VectorOffset and ReadableVectorOffset are like Offset and ReadableOffset.
- VectorAggregator is like BufferAggregator.
- VectorValueMatcher is like ValueMatcher.
There are some noticeable differences between vectorized and regular
execution:
- Unlike regular cursors, vector cursors do not understand time
granularity. They expect query engines to handle this on their own,
which a new VectorCursorGranularizer class helps with. This is to
avoid too much batch-splitting and to respect the fact that vector
selectors are somewhat more heavyweight than regular selectors.
- Unlike FilteredOffset, FilteredVectorOffset does not leverage indexes
for filters that might partially support them (like an OR of one
filter that supports indexing and another that doesn't). I'm not sure
that this behavior is desirable anyway (it is potentially too eager)
but, at any rate, it'd be better to harmonize it between the two
classes. Potentially they should both do some different thing that
is smarter than what either of them is doing right now.
- When vector cursors are created by QueryableIndexCursorSequenceBuilder,
they use a morphing binary-then-linear search to find their start and
end rows, rather than linear search.
Limitations in this patch are:
- Only timeseries and groupBy have vectorized engines.
- GroupBy doesn't handle multi-value dimensions yet.
- Vector cursors cannot handle virtual columns or descending order.
- Only some filters have vectorized matchers: "selector", "bound", "in",
"like", "regex", "search", "and", "or", and "not".
- Only some aggregators have vectorized implementations: "count",
"doubleSum", "floatSum", "longSum", "hyperUnique", and "filtered".
- Dimension specs other than "default" don't work yet (no extraction
functions or filtered dimension specs).
Currently, the testing strategy includes adding vectorization-enabled
tests to TimeseriesQueryRunnerTest, GroupByQueryRunnerTest,
GroupByTimeseriesQueryRunnerTest, CalciteQueryTest, and all of the
filtering tests that extend BaseFilterTest. In all of those classes,
there are some test cases that don't support vectorization. They are
marked by special function calls like "cannotVectorize" or "skipVectorize"
that tell the test harness to either expect an exception or to skip the
test case.
Testing should be expanded in the future -- a project in and of itself.
Related to #3011.
* WIP
* Adjustments for unused things.
* Adjust javadocs.
* DimensionDictionarySelector adjustments.
* Add "clone" to BatchIteratorAdapter.
* ValueMatcher javadocs.
* Fix benchmark.
* Fixups post-merge.
* Expect exception on testGroupByWithStringVirtualColumn for IncrementalIndex.
* BloomDimFilterSqlTest: Tag two non-vectorizable tests.
* Minor adjustments.
* Update surefire, bump up Xmx in Travis.
* Some more adjustments.
* Javadoc adjustments
* AggregatorAdapters adjustments.
* Additional comments.
* Remove switching search.
* Only missiles.
* Fix license check in travis and make it optional
* debug
* fix build
* too loud maven
* move MAVEN_OPTS to top and add comments
* adjust script
* remove mvn option from python script
* Add Spotbugs
Exclude all the issues for now, so we can add them one by one.
(cherry picked from commit ceda4754dc8c703d1e0de85b48cd5f5409cfd5b7)
* Add additional rules to the list
* More rules
* More rules
* Add comments to the xml
* Move the spotbugs-exclude.xml to codestyle/
This change only enables compilation to ensure code compiles against
recent Java versions going forward. Tests are still disabled in this
profile until test failures are addressed.
* Adding new web console.
* fixed css
* fix form height
* fix typo
* do import custom react-table css
* added repo field so npm does not complain
* ask travis for node 10
* move indexing-service/src/main/resources/indexer_static into web-console
* fix resource names and paths
* add licenses
* fix exclude file
* add licenses to misc files and tidy up
* remove rebase marker
* fix link
* updated env variable name
* tidy up licenses and surface errors
* cleanup
* remove unused code, fix missing await
* TeamCity does not like the name aux
* add more links to tasks view
* rm pages
* update gitignore
* update readme to be accurate
* make clean script
* removed old console dependancy
* update Jetty routes
* add a comment for welcome files for coordinator
* do not show inital notifaction for now
* renamed overlord console back to console.html
* fix coordinator console
* rename coordinator-console.html to index.html
* Rename io.druid to org.apache.druid.
* Fix META-INF files and remove some benchmark results.
* MonitorsConfig update for metrics package migration.
* Reorder some dimensions in inner queries for some reason.
* Fix protobuf tests.
* Revert "Fix a bug in GroupByQueryEngine (#6062)"
This reverts commit f3595c93d9.
* Revert "Add definition of 'NONE' to queryGranularity in ingestion.index doc (#6073)"
This reverts commit 7f89c72932.
* Revert "skip travis on doc only changes (#6061)"
This reverts commit 66af403f7d.
* Use the official aws-sdk instead of jet3t
* fix compile and serde tests
* address comments and fix test
* add http version string
* remove redundant dependencies, fix potential NPE, and fix test
* resolve TODOs
* fix build
* downgrade jackson version to 2.6.7
* fix test
* resolve the last TODO
* support proxy and endpoint configurations
* fix build
* remove debugging log
* downgrade hadoop version to 2.8.3
* fix tests
* remove unused log
* fix it test
* revert KerberosAuthenticator change
* change hadoop-aws scope to provided in hdfs-storage
* address comments
* address comments
* Use trusty for travis jobs.
The distro was set to "precise" in #4572 due to memory issues on trusty,
but we've been seeing performance issues on "precise" recently so let's
see how trusty is working these days.
* Less quiet.
* Adjust memory settings.
* Add back -q option.
* Tweak memory again.
* Adjustments.
* Try squeezing memory a bit more.
Chi Cao Minh
Clint Wylie
Xavier Léauté
Gian Merlino
Jihoon Son
Khwunchai Jaengsawang
Fokko Driesprong
Clint Wylie
Vadim Ogievetsky
David Lim
Jonathan Wei
Roman Leventov
Nishant Bangarwa
Benedict Jin
Egor Riashin