Currently we assign the buckets by record partition path which could
cause hotspot if the partition field is datetime type. Changes to assign
buckets by grouping the record whth their key first, the assignment is
valid if only there is no conflict(two task write to the same bucket).
This patch also changes the coordinator execution to be asynchronous.
Current we did a soft delete for DELETE row data when writes into hoodie
table. For streaming read of MOR table, the Flink reader detects the
delete records and still emit them if the record key semantics are still
kept.
This is useful and actually a must for streaming ETL pipeline
incremental computation.
* [HUDI-1653] Add support for composite keys in NonpartitionedKeyGenerator
* update NonpartitionedKeyGenerator to support composite record keys
* update NonpartitionedKeyGenerator
* [HUDI-845] Added locking capability to allow multiple writers
1. Added LockProvider API for pluggable lock methodologies
2. Added Resolution Strategy API to allow for pluggable conflict resolution
3. Added TableService client API to schedule table services
4. Added Transaction Manager for wrapping actions within transactions
* [HUDI-1552] Improve performance of key lookups from base file in Metadata Table.
1. Cache the KeyScanner across lookups so that the HFile index does not have to be read for each lookup.
2. Enable block caching in KeyScanner.
3. Move the lock to a limited scope of the code to reduce lock contention.
4. Removed reuse configuration
* Properly close the readers, when metadata table is accessed from executors
- Passing a reuse boolean into HoodieBackedTableMetadata
- Preserve the fast return behavior when reusing and opening from multiple threads (no contention)
- Handle concurrent close() and open readers, for reuse=false, by always synchronizing
Co-authored-by: Vinoth Chandar <vinoth@apache.org>
In order to support object storage, we need these changes:
* Use the Hadoop filesystem so that we can find the plugin filesystem
* Do not fetch file size until the file handle is closed
* Do not close the opened filesystem because we want to use the
filesystem cache
- introduce configs to control how compaction is triggered
- Compaction can be triggered using time, number of delta commits and/or combinations
- Default behaviour remains the same.
This is the #step 2 of RFC-24:
https://cwiki.apache.org/confluence/display/HUDI/RFC+-+24%3A+Hoodie+Flink+Writer+Proposal
This PR introduce a BucketAssigner that assigns bucket ID (partition
path & fileID) for each stream record.
There is no need to look up index and partition the records anymore in
the following pipeline for these records,
we actually decide the write target location before the write and each
record computes its location when the BucketAssigner receives it, thus,
the indexing is with streaming style.
Computing locations for a batch of records all at a time is resource
consuming so a pressure to the engine,
we should avoid that in streaming system.
* Added HoodieConcatHandle to skip merging for "insert" operation when the corresponding config is set
Co-authored-by: Sivabalan Narayanan <sivabala@uber.com>
Addresses leaks, perf degradation observed during testing. These were regressions from the original rfc-15 PoC implementation.
* Pass a single instance of HoodieTableMetadata everywhere
* Fix tests and add config for enabling metrics
- Removed special casing of assumeDatePartitioning inside FSUtils#getAllPartitionPaths()
- Consequently, IOException is never thrown and many files had to be adjusted
- More diligent handling of open file handles in metadata table
- Added config for controlling reuse of connections
- Added config for turning off fallback to listing, so we can see tests fail
- Changed all ipf listing code to cache/amortize the open/close for better performance
- Timelineserver also reuses connections, for better performance
- Without timelineserver, when metadata table is opened from executors, reuse is not allowed
- HoodieMetadataConfig passed into HoodieTableMetadata#create as argument.
- Fix TestHoodieBackedTableMetadata#testSync
- Adds field to RollbackMetadata that capture the logs written for rollback blocks
- Adds field to RollbackMetadata that capture new logs files written by unsynced deltacommits
Co-authored-by: Vinoth Chandar <vinoth@apache.org>
* [HUDI-1479] Use HoodieEngineContext to parallelize fetching of partition paths
* Adding testClass for FileSystemBackedTableMetadata
Co-authored-by: Nishith Agarwal <nagarwal@uber.com>
* [HUDI-1481] add structured streaming and delta streamer clustering unit test
* [HUDI-1399] support a independent clustering spark job to asynchronously clustering
* [HUDI-1399] support a independent clustering spark job to asynchronously clustering
* [HUDI-1498] Read clustering plan from requested file for inflight instant (#2389)
* [HUDI-1399] support a independent clustering spark job with schedule generate instant time
Co-authored-by: satishkotha <satishkotha@uber.com>
* [HUDI-1276] [HUDI-1459] Make Clustering/ReplaceCommit and Metadata table be compatible
* Use filesystemview and json format from metadata. Add tests
Co-authored-by: Satish Kotha <satishkotha@uber.com>
- Syncing to metadata table, setting operation type, starting async cleaner done in preWrite()
- Fixes an issues where delete() was not starting async cleaner correctly
- Fixed tests and enabled metadata table for TestAsyncCompaction
- TestHoodieBackedMetadata#testSync etc now run for MOR tables
- HUDI-1502 is still pending and has issues for MOR/rollbacks
- Also addressed bunch of code review comments.
* [HUDI-1469] Faster initialization of metadata table using parallelized listing which finds partitions and files in a single scan.
* MINOR fixes
Co-authored-by: Vinoth Chandar <vinoth@apache.org>
- Introduced an internal metadata table, that stores file listings.
- metadata table is kept upto date with
- Fixed handling of CleanerPlan.
- [HUDI-842] Reduce parallelism to speed up the test.
- [HUDI-842] Implementation of CLI commands for metadata operations and lookups.
- [HUDI-842] Extend rollback metadata to include the files which have been appended to.
- [HUDI-842] Support for rollbacks in MOR Table.
- MarkerBasedRollbackStrategy needs to correctly provide the list of files for which rollback blocks were appended.
- [HUDI-842] Added unit test for rollback of partial commits (inflight but not completed yet).
- [HUDI-842] Handled the error case where metadata update succeeds but dataset commit fails.
- [HUDI-842] Schema evolution strategy for Metadata Table. Each type of metadata saved (FilesystemMetadata, ColumnIndexMetadata, etc.) will be a separate field with default null. The type of the record will identify the valid field. This way, we can grow the schema when new type of information is saved within in which still keeping it backward compatible.
- [HUDI-842] Fix non-partitioned case and speedup initial creation of metadata table.Choose only 1 partition for jsc as the number of records is low (hundreds to thousands). There is more overhead of creating large number of partitions for JavaRDD and it slows down operations like WorkloadProfile.
For the non-partitioned case, use "." as the name of the partition to prevent empty keys in HFile.
- [HUDI-842] Reworked metrics pusblishing.
- Code has been split into reader and writer side. HoodieMetadata code to be accessed by using HoodieTable.metadata() to get instance of metdata for the table.
Code is serializable to allow executors to use the functionality.
- [RFC-15] Add metrics to track the time for each file system call.
- [RFC-15] Added a distributed metrics registry for spark which can be used to collect metrics from executors. This helps create a stats dashboard which shows the metadata table improvements in real-time for production tables.
- [HUDI-1321] Created HoodieMetadataConfig to specify configuration for the metadata table. This is safer than full-fledged properties for the metadata table (like HoodieWriteConfig) as it makes burdensome to tune the metadata. With limited configuration, we can control the performance of the metadata table closely.
[HUDI-1319][RFC-15] Adding interfaces for HoodieMetadata, HoodieMetadataWriter (apache#2266)
- moved MetadataReader to HoodieBackedTableMetadata, under the HoodieTableMetadata interface
- moved MetadataWriter to HoodieBackedTableMetadataWriter, under the HoodieTableMetadataWriter
- Pulled all the metrics into HoodieMetadataMetrics
- Writer now wraps the metadata, instead of extending it
- New enum for MetadataPartitionType
- Streamlined code flow inside HoodieBackedTableMetadataWriter w.r.t initializing metadata state
- [HUDI-1319] Make async operations work with metadata table (apache#2332)
- Changes the syncing model to only move over completed instants on data timeline
- Syncing happens postCommit and on writeClient initialization
- Latest delta commit on the metadata table is sufficient as the watermark for data timeline archival
- Cleaning/Compaction use a suffix to the last instant written to metadata table, such that we keep the 1-1
- .. mapping between data and metadata timelines.
- Got rid of a lot of the complexity around checking for valid commits during open of base/log files
- Tests now use local FS, to simulate more failure scenarios
- Some failure scenarios exposed HUDI-1434, which is needed for MOR to work correctly
co-authored by: Vinoth Chandar <vinoth@apache.org>
