Refcounting prevents releasing the merge buffer, or closing the concurrent
grouper, before the processing threads have all finished. The better
error handling prevents an avalanche of per-runner exceptions when grouping
resources are exhausted, by grouping those all up into a single merged
exception.
This reverts commit a931debf79.
Fixes#3283
The core issue here is that realtime nodes announce their size as 0, so a coordinator which interns the realtime version of the data segment will not be able to see the new sized announcement when handoff occurs.
This is caused by the `eauals` method on a `DataSegment` only evaluating the identifier. the `eauals` method *should* be correct for object equivalence, and things which need to check equivalence of some sub-portion of the object should do so explicitly.
* InputRowParser to decode OrcStruct from OrcNewInputFormat
* add unit test for orc hadoop indexing
* update docs and fix test code bug
* doc updated
* resove maven dependency conflict
* remove unused imports
* fix returning array type from Object[] to correct primitive array type
* fix to support getDimension() of MapBasedRow : changing return type of orc list from array to list
* rebase and updated based on comments
* updated based on comments
* on reflecting review comments
* fix bug in typeStringFromParseSpec() and add unit test
* add license header
* Support filtering on __time column
* Rename DruidPredicate
* Add docs for ValueMatcherFactory, add comment on getColumnCapabilities
* Combine ValueMatcherFactory predicate methods to accept DruidCompositePredicate
* Address PR comments (support filter on all long columns)
* Use predicate factory instead of composite predicate
* Address PR comments
* Lazily initialize long handling in selector/in filter
* Move long value parsing from InFilter to InDimFilter, make long value parsing thread-safe
* Add multithreaded selector/in filter test
* Fix non-final lock object in SelectorDimFilter
Fixes inconsistent metric handling between the two implementations. Formerly,
RealtimePlumber only emitted query/segmentAndCache/time and query/wait and
Appenderator only emitted query/partial/time and query/wait (all per sink).
Now they both do the same thing:
- query/segmentAndCache/time, query/segment/time are the time spent per sink.
- query/cpu/time is the CPU time spent per query.
- query/wait/time is the executor waiting time per sink.
These generally match historical metrics, except segmentAndCache & segment
mean the same thing here, because one Sink may be partially cached and
partially uncached and we aren't splitting that out.
1) Modify CliHadoopIndexer to share constant from `TaskConfig.DEFAULT_DEFAULT_HADOOP_COORDINATES`
2) add comment to pom.xml as discussed in
https://github.com/druid-io/druid/pull/3044
fix name
- Attempt to make things clearer in general
- Point out that HDFS deep storage and MR jobs don't use the same loading mechanism
- Recommend using mapreduce.job.classloader = true when possible
constantly timing out on one of slow build machines, increasing the
timeout fixed it.
Running io.druid.granularity.QueryGranularityTest
Tests run: 33, Failures: 0, Errors: 0, Skipped: 0, Time elapsed: 11.776
sec - in io.druid.granularity.QueryGranularityTest
All query metrics now start with toolChest.makeMetricBuilder, and all of
*those* now start with DruidMetrics.makePartialQueryTimeMetric. Also, "id"
moved to common code, since all query metrics added it anyway.
In particular this will add query-type specific dimensions like "threshold"
and "numDimensions" to servlet-originated metrics like query/time.
Because timestamps at the end instant are not actually part of the interval. This
affected benchmark numbers, since it meant some data points would not be queried
(the interval for the query was based on getDataInterval) and also the
TimestampCheckingOffsets could not use the allWithinThreshold optimization.
* Migrate IndexerSQLMetadataStorageCoordinator.getUnusedSegmentsForInterval to streaming
* Missed query from #2859
* Make inReadOnlyTransaction part of SQLMetadataConnector
* Initial commit of caffeine cache
* Address code comments
* Move and fixup README.md a bit
* Improve caffeine readme information
* Cleanup caffeine pom
* Address review comments
* Bump caffeine to 2.3.1
* Bump druid version to 0.9.2-SNAPSHOT
* Make test not fail randomly.
See https://github.com/ben-manes/caffeine/pull/93#issuecomment-227617998 for an explanation
* Fix distribution and documentation
* Add caffeine to extensions.md
* Fix links in extensions.md
* Lexicographic
This is actually reasonable for a groupBy or lexicographic topNs that is
being used to do a "COUNT DISTINCT" kind of query. No aggregators are
needed for that query, and including a dummy aggregator wastes 8 bytes
per row.
It's kind of silly for timeseries, but why not.
* support alphanumeric sort in search query
* address a comment about handling equals() and hashCode()
* address comments
* add Ut for string comparators
* address a comment about space indentations.
This patch introduces a GroupByStrategy concept and two strategies: "v1"
is the current groupBy strategy and "v2" is a new one. It also introduces
a merge buffers concept in DruidProcessingModule, to try to better
manage memory used for merging.
Both of these are described in more detail in #2987.
There are two goals of this patch:
1. Make it possible for historical/realtime nodes to return larger groupBy
result sets, faster, with better memory management.
2. Make it possible for brokers to merge streams when there are no order-by
columns, avoiding materialization.
This patch does not do anything to help with memory management on the broker
when there are order-by columns or when there are nested queries. That could
potentially be done in a future patch.
Gian Merlino
Charles Allen
Keuntae Park
Erik Dubbelboer
kaijianding
David Lim
Jonathan Wei
Navis Ryu
Parag Jain
Nishant
Emanuele Cesena
Himanshu
Hyukjin Kwon
Fangjin Yang
Bingkun Guo
Xavier Léauté
jaehong choi
du00cs