Makes kinesis ingestion resilient to `LimitExceededException` caused by resharding.
Replace `describeStream` with `listShards` (recommended) to get shard related info.
`describeStream` has a limit (100) to the number of shards returned per call and a low default TPS limit of 10.
`listShards` returns the info for at most 1000 shards and has a higher TPS limit of 100 as well.
Key changed/added classes in this PR
* `KinesisRecordSupplier`
* `KinesisAdminClient`
This fixes a bug that causes TaskClient in overlord to continuously retry to pause tasks. This can happen when a task is not responding to the pause command. Ideally, in such a case when the task is unresponsive, the overlord would have given up after a few retries and would have killed the task. However, due to this bug, retries go on forever.
* Ingestion will fail for HLLSketchBuild instead of creating with incorrect values
* Addressing review comments for HLL< updated error message introduced test case
* Add jsonPath functions support
* Add jsonPath function test for Avro
* Add jsonPath function length() to Orc
* Add jsonPath function length() to Parquet
* Add more tests to ORC format
* update doc
* Fix exception during ingestion
* Add IT test case
* Revert "Fix exception during ingestion"
This reverts commit 5a5484b9ea.
* update IT test case
* Add 'keys()'
* Commit IT test case
* Fix UT
This PR fixes an issue in which if a lookup is configured incorreclty; does not serialize properly when being pulled by peon node, it causes the task to fail. The failure occurs because the peon and other leaf nodes (broker, historical), have retry logic that continues to retry the lookup loading for 3 minutes by default. The http listener thread on the peon task is not started until lookup loading completes, by default, the overlord waits 1 minute by default, to communicate with the peon task to get the task status, after which is orders the task to shut down, causing the ingestion task to fail.
To fix the issue, we catch the exception serialization error, and do not retry. Also fixed an issue in which a bad lookup config interferes with any other good lookup configs from being loaded.
* Enhancements to IndexTaskClient.
1) Ability to use handlers other than StringFullResponseHandler. This
functionality is not used in production code yet, but is useful
because it will allow tasks to communicate with each other in
non-string-based formats and in streaming fashion. In the future,
we'll be able to use this to make task-to-task communication
more efficient.
2) Truncate server errors at 1KB, so long errors do not pollute logs.
3) Change error log level for retryable errors from WARN to INFO. (The
final error is still WARN.)
4) Harmonize log and exception messages to have a more consistent format.
* Additional tests and improvements.
This PR fixes a problem where the com.sun.jndi.ldap.Connection tries to build BasicSecuritySSLSocketFactory when calling LDAPCredentialsValidator.validateCredentials since BasicSecuritySSLSocketFactory is in extension class loader and not visible to system classloader.
changes:
* adds new config, druid.expressions.useStrictBooleans which make longs the official boolean type of all expressions
* vectorize logical operators and boolean functions, some only if useStrictBooleans is true
* Code cleanup from query profile project
* Fix spelling errors
* Fix Javadoc formatting
* Abstract out repeated test code
* Reuse constants in place of some string literals
* Fix up some parameterized types
* Reduce warnings reported by Eclipse
* Reverted change due to lack of tests
Add a "guessAggregatorHeapFootprint" method to AggregatorFactory that
mitigates #6743 by enabling heap footprint estimates based on a specific
number of rows. The idea is that at ingestion time, the number of rows
that go into an aggregator will be 1 (if rollup is off) or will likely
be a small number (if rollup is on).
It's a heuristic, because of course nothing guarantees that the rollup
ratio is a small number. But it's a common case, and I expect this logic
to go wrong much less often than the current logic. Also, when it does
go wrong, users can fix it by lowering maxRowsInMemory or
maxBytesInMemory. The current situation is unintuitive: when the
estimation goes wrong, users get an OOME, but actually they need to
*raise* these limits to fix it.
* Add support for custom reset condition & support for other args to have defaults to make the method api consistent
* Add support for custom reset condition to InputEntity
* Fix test names
* Clarifying comments to why we need to read the message's content to identify S3's resettable exception
* Add unit test to verify custom resettable condition for S3Entity
* Provide a way to customize retries since they are expensive to test
* add back and deprecate aggregator factory methods so i can say i told you so when i delete these later
* rename to make less ambiguous, fix fill method
* adjust
* add missing json type for ListFilteredVirtualColumn, and tests to try to avoid this happening again
* fixes
* ugly, but maybe this
* oops
* too many mappers
* complex typed expressions
* add built-in hll collector expressions to get coverage on druid-processing, more types, more better
* rampage!!!
* more javadoc
* adjustments
* oops
* lol
* remove unused dependency
* contradiction?
* more test
Enhanced the ExtractionNamespace interface in lookups-cached-global core extension with the ability to set a maxHeapPercentage for the cache of the respective namespace. The reason for adding this functionality, is make it easier to detect when a lookup table grows to a size that the underlying service cannot handle, because it does not have enough memory. The default value of maxHeap for the interface is -1, which indicates that no maxHeapPercentage has been set. For the JdbcExtractionNamespace and UriExtractionNamespace implementations, the default value is null, which will cause the respective service that the lookup is loaded in, to warn when its cache is beyond mxHeapPercentage of the service's configured max heap size. If a positive non-null value is set for the namespace's maxHeapPercentage config, this value will be honored for all services that the respective lookup is loaded onto, and consequently log warning messages when the cache of the respective lookup grows beyond this respective percentage of the services configured max heap size. Warnings are logged every time that either Uri based or Jdbc based lookups are regenerated, if the maxHeapPercentage constraint is violated. No other implementations will log warnings at this time. No error is thrown when the size exceeds the maxHeapPercentage at this time, as doing so could break functionality for existing users. Previously the JdbcCacheGenerator generated its cache by materializing all rows of the underling table in memory at once; this made it difficult to log warning messages in the case that the results from the jdbc query were very large and caused the service to run out of memory. To help with this, this pr makes it so that the jdbc query results are instead streamed through an iterator.
Add support for hadoop 3 profiles . Most of the details are captured in #11791 .
We use a combination of maven profiles and resource filtering to achieve this. Hadoop2 is supported by default and a new maven profile with the name hadoop3 is created. This will allow the user to choose the profile which is best suited for the use case.
* Remove OffheapIncrementalIndex and clarify aggregator thread-safety needs.
This patch does the following:
- Removes OffheapIncrementalIndex.
- Clarifies that Aggregators are required to be thread safe.
- Clarifies that BufferAggregators and VectorAggregators are not
required to be thread safe.
- Removes thread safety code from some DataSketches aggregators that
had it. (Not all of them did, and that's OK, because it wasn't necessary
anyway.)
- Makes enabling "useOffheap" with groupBy v1 an error.
Rationale for removing the offheap incremental index:
- It is only used in one rare scenario: groupBy v1 (which is non-default)
in "useOffheap" mode (also non-default). So you have to go pretty deep
into the wilderness to get this code to activate in production. It is
never used during ingestion.
- Its existence complicates developer efforts to reason about how
aggregators get used, because the way it uses buffer aggregators is so
different from how every other query engine uses them.
- It doesn't have meaningful testing.
By the way, I do believe that the given way the offheap incremental index
works, it actually didn't require buffer aggregators to be thread-safe.
It synchronizes on "aggregate" and doesn't call "get" until it has
stopped calling "aggregate". Nevertheless, this is a bother to think about,
and for the above reasons I think it makes sense to remove the code anyway.
* Remove things that are now unused.
* Revert removal of getFloat, getLong, getDouble from BufferAggregator.
* OAK-related warnings, suppressions.
* Unused item suppressions.
* Add druid.sql.approxCountDistinct.function property.
The new property allows admins to configure the implementation for
APPROX_COUNT_DISTINCT and COUNT(DISTINCT expr) in approximate mode.
The motivation for adding this setting is to enable site admins to
switch the default HLL implementation to DataSketches.
For example, an admin can set:
druid.sql.approxCountDistinct.function = APPROX_COUNT_DISTINCT_DS_HLL
* Fixes
* Fix tests.
* Remove erroneous cannotVectorize.
* Remove unused import.
* Remove unused test imports.
* SQL: Allow Scans to be used as outer queries.
This has been possible in the native query system for a while, but the capability
hasn't yet propagated into the SQL layer. One example of where this is useful is
a query like:
SELECT * FROM (... LIMIT X) WHERE <filter>
Because this expands the kinds of subquery structures the SQL layer will consider,
it was also necessary to improve the cost calculations. These changes appear in
PartialDruidQuery and DruidOuterQueryRel. The ideas are:
- Attach per-column penalties to the output signature of each query, instead of to
the initial projection that starts a query. This encourages moving projections
into subqueries instead of leaving them on outer queries.
- Only attach penalties to projections if there are actually expressions happening.
So, now, projections that simply reorder or remove fields are free.
- Attach a constant penalty to every outer query. This discourages creating them
when they are not needed.
The changes are generally beneficial to the test cases we have in CalciteQueryTest.
Most plans are unchanged, or are changed in purely cosmetic ways. Two have changed
for the better:
- testUsingSubqueryWithLimit now returns a constant from the subquery, instead of
returning every column.
- testJoinOuterGroupByAndSubqueryHasLimit returns a minimal set of columns from
the innermost subquery; two unnecessary columns are no longer there.
* Fix various DS operator conversions.
These were all implemented as direct conversions, which isn't appropriate
because they do not actually map onto native functions. These are only
usable as post-aggregations.
* Test case adjustment.
* Remove CloseQuietly and migrate its usages to other methods.
These other methods include:
1) New method CloseableUtils.closeAndWrapExceptions, which wraps IOExceptions
in RuntimeExceptions for callers that just want to avoid dealing with
checked exceptions. Most usages were migrated to this method, because it
looks like they were mainly attempts to avoid declaring a throws clause,
and perhaps were unintentionally suppressing IOExceptions.
2) New method CloseableUtils.closeInCatch, designed to properly close something
in a catch block without losing exceptions. Some usages from catch blocks
were migrated here, when it seemed that they were intended to avoid checked
exception handling, and did not really intend to also suppress IOExceptions.
3) New method CloseableUtils.closeAndSuppressExceptions, which sends all
exceptions to a "chomper" that consumes them. Nothing is thrown or returned.
The behavior is slightly different: with this method, _all_ exceptions are
suppressed, not just IOExceptions. Calls that seemed like they had good
reason to suppress exceptions were migrated here.
4) Some calls were migrated to try-with-resources, in cases where it appeared
that CloseQuietly was being used to avoid throwing an exception in a finally
block.
🎵 You don't have to go home, but you can't stay here... 🎵
* Remove unused import.
* Fix up various issues.
* Adjustments to tests.
* Fix null handling.
* Additional test.
* Adjustments from review.
* Fixup style stuff.
* Fix NPE caused by holder starting out null.
* Fix spelling.
* Chomp Throwables too.
* Null handling fixes for DS HLL and Theta sketches.
For HLL, this fixes an NPE when processing a null in a multi-value dimension.
For both, empty strings are now properly treated as nulls (and ignored) in
replace-with-default mode. Behavior in SQL-compatible mode is unchanged.
* Fix expectation.
* add ColumnInspector argument to PostAggregator.getType to allow post-aggs to compute their output type based on input types
* add test for test for coverage
* simplify
* Remove unused imports.
Co-authored-by: Gian Merlino <gian@imply.io>
* latest datasketches-java and datasketches-memory
* updated versions of datasketches-java and datasketches-memory
Co-authored-by: AlexanderSaydakov <AlexanderSaydakov@users.noreply.github.com>
* better type system
* needle in a haystack
* ColumnCapabilities is a TypeSignature instead of having one, INFORMATION_SCHEMA support
* fixup merge
* more test
* fixup
* intern
* fix
* oops
* oops again
* ...
* more test coverage
* fix error message
* adjust interning, more javadocs
* oops
* more docs more better
### Description
Today we ingest a number of high cardinality metrics into Druid across dimensions. These metrics are rolled up on a per minute basis, and are very useful when looking at metrics on a partition or client basis. Events is another class of data that provides useful information about a particular incident/scenario inside a Kafka cluster. Events themselves are carried inside kafka payload, but nonetheless there are some very useful metadata that is carried in kafka headers that can serve as useful dimension for aggregation and in turn bringing better insights.
PR(https://github.com/apache/druid/pull/10730) introduced support of Kafka headers in InputFormats.
We still need an input format to parse out the headers and translate those into relevant columns in Druid. Until that’s implemented, none of the information available in the Kafka message headers would be exposed. So first there is a need to write an input format that can parse headers in any given format(provided we support the format) like we parse payloads today. Apart from headers there is also some useful information present in the key portion of the kafka record. We also need a way to expose the data present in the key as druid columns. We need a generic way to express at configuration time what attributes from headers, key and payload need to be ingested into druid. We need to keep the design generic enough so that users can specify different parsers for headers, key and payload.
This PR is designed to solve the above by providing wrapper around any existing input formats and merging the data into a single unified Druid row.
Lets look at a sample input format from the above discussion
"inputFormat":
{
"type": "kafka", // New input format type
"headerLabelPrefix": "kafka.header.", // Label prefix for header columns, this will avoid collusions while merging columns
"recordTimestampLabelPrefix": "kafka.", // Kafka record's timestamp is made available in case payload does not carry timestamp
"headerFormat": // Header parser specifying that values are of type string
{
"type": "string"
},
"valueFormat": // Value parser from json parsing
{
"type": "json",
"flattenSpec": {
"useFieldDiscovery": true,
"fields": [...]
}
},
"keyFormat": // Key parser also from json parsing
{
"type": "json"
}
}
Since we have independent sections for header, key and payload, it will enable parsing each section with its own parser, eg., headers coming in as string and payload as json.
KafkaInputFormat will be the uber class extending inputFormat interface and will be responsible for creating individual parsers for header, key and payload, blend the data resolving conflicts in columns and generating a single unified InputRow for Druid ingestion.
"headerFormat" will allow users to plug parser type for the header values and will add default header prefix as "kafka.header."(can be overridden) for attributes to avoid collision while merging attributes with payload.
Kafka payload parser will be responsible for parsing the Value portion of the Kafka record. This is where most of the data will come from and we should be able to plugin existing parser. One thing to note here is that if batching is performed, then the code is augmenting header and key values to every record in the batch.
Kafka key parser will handle parsing Key portion of the Kafka record and will ingest the Key with dimension name as "kafka.key".
## KafkaInputFormat Class:
This is the class that orchestrates sending the consumerRecord to each parser, retrieve rows, merge the columns into one final row for Druid consumption. KafkaInputformat should make sure to release the resources that gets allocated as a part of reader in CloseableIterator<InputRow> during normal and exception cases.
During conflicts in dimension/metrics names, the code will prefer dimension names from payload and ignore the dimension either from headers/key. This is done so that existing input formats can be easily migrated to this new format without worrying about losing information.
* refactor sql authorization to get resource type from schema, refactor resource type from enum to string
* information schema auth filtering adjustments
* refactor
* minor stuff
* Update SqlResourceCollectorShuttle.java
When CommonCachedNotifier is being stopped while the thread is waiting on updateQueue.take(),
an InterruptedException is thrown. The stack trace from this exception gives the wrong idea that something went wrong with the shutdown.
* Make persists concurrent with ingestion
* Remove semaphore but keep concurrent persists (with add) and add push in the backround as well
* Go back to documented default persists (zero)
* Move to debug
* Remove unnecessary Atomics
* Comments on synchronization (or not) for sinks & sinkMetadata
* Some cleanup for unit tests but they still need further work
* Shutdown & wait for persists and push on close
* Provide support for three existing batch appenderators using batchProcessingMode flag
* Fix reference to wrong appenderator
* Fix doc typos
* Add BatchAppenderators class test coverage
* Add log message to batchProcessingMode final value, fix typo in enum name
* Another typo and minor fix to log message
* LEGACY->OPEN_SEGMENTS, Edit docs
* Minor update legacy->open segments log message
* More code comments, mostly small adjustments to naming etc
* fix spelling
* Exclude BtachAppenderators from Jacoco since it is fully tested but Jacoco still refuses to ack coverage
* Coverage for Appenderators & BatchAppenderators, name change of a method that was still using "legacy" rather than "openSegments"
Co-authored-by: Clint Wylie <cjwylie@gmail.com>
* Configurable maxStreamLength for doubles sketches
* fix equals/hashcode and it test failure
* fix test
* fix it test
* benchmark
* doc
* grouping key
* fix comment
* dependency check
* Update docs/development/extensions-core/datasketches-quantiles.md
Co-authored-by: Charles Smith <techdocsmith@gmail.com>
* Update docs/querying/sql.md
Co-authored-by: Charles Smith <techdocsmith@gmail.com>
* Update docs/querying/sql.md
Co-authored-by: Charles Smith <techdocsmith@gmail.com>
* Update docs/querying/sql.md
Co-authored-by: Charles Smith <techdocsmith@gmail.com>
* Update docs/querying/sql.md
Co-authored-by: Charles Smith <techdocsmith@gmail.com>
* Update docs/querying/sql.md
Co-authored-by: Charles Smith <techdocsmith@gmail.com>
* Update docs/querying/sql.md
Co-authored-by: Charles Smith <techdocsmith@gmail.com>
* Update docs/querying/sql.md
Co-authored-by: Charles Smith <techdocsmith@gmail.com>
Co-authored-by: Charles Smith <techdocsmith@gmail.com>
Fixes#11297.
Description
Description and design in the proposal #11297
Key changed/added classes in this PR
*DataSegmentPusher
*ShuffleClient
*PartitionStat
*PartitionLocation
*IntermediaryDataManager
This PR adds a new property druid.router.sql.enable which allows the
Router to handle SQL queries when set to true.
This change does not affect Avatica JDBC requests and they are still routed
by hashing the Connection ID.
To allow parsing of the request object as a SqlQuery (contained in module druid-sql),
some classes have been moved from druid-server to druid-services with
the same package name.
* Better logging for lookups
The default pollPeriod of 0 means that lookups are loaded once only at startup
Add a warning message to warn operators about this. I suspect that most
operators using jdbc or uri would expect eventual consistency with the source
of the lookups if using jdbc or uri. So make this a warning to make it easier
to debug if an operator notices a data inconsistency issue.
* oops