Databricks Inc.

Prakhar Jain

Bart Samwel

Frederick Ryan Johnson

Burak Yavuz

Tathagata Das

Juliusz Sompolski

Lars Kroll

Yijia Cui

Tom Van Bussel

Vijayan Prabhakaran

Terry Kim

The present application discloses a method, system, and computer system for managing a data in a storage system. The method includes receiving a first transaction that modifies or deletes first data stored in a storage system, determining that the first data is subject to an intervening re-arrangement transaction, and in response to determining that the first data is subject to the intervening re-arrangement transaction, rolling back the re-arrangement transaction at least with respect to the first data and committing the first transaction.

Data maintenance transaction rollbacks

A data processing service generates a data classifier tree for managing data files of a data table. The data classifier tree may be configured as a KD-classifier tree and includes a plurality of nodes and edges. A node of the data classifier tree may represent a splitting condition with respect to key-values for a respective key. A node of the data classifier tree may be associated with one or more data files assigned to the node. The data files assigned to the node each include a subset of records having key-values that satisfy the conditions represented by the node and parent nodes of the node. The data processing service may efficiently cluster the data in the data table while reducing the number of data files that are rewritten when data is modified or added to the data table.

Data file clustering with KD-classifier trees

A data processing service uses minor compactions for committing transactions to a data table. The service may receive requests to commit transactions to a data table and write metadata for the transactions to log files, and generate a checkpoint file aggregating the transactions described in the log files to compute a data table state at a first time. The service may receive requests to commit a set of transactions and write metadata for the set of transactions to a set of log files. The service may determine that a number of log files in the set of log files reaches a threshold commit number, generate a minor compaction file aggregating the set of transactions, and generate a second checkpoint file aggregating the data table state at the first time with information from the minor compaction file to compute the data table state at a second time.

Generating minor compactions to capture aggregated actions for commit ranges to data files

A data tree for managing data files of a data table and performing one or more transaction operations to the data table is described. The data tree is configured as a KD-epsilon tree and includes a plurality of nodes and edges. A node of the data tree may represent a splitting condition with respect to key-values for a respective key. A leaf node of the data tree may correspond to a data file for a data table that includes a subset of records having key-values that satisfy the condition for the node and conditions associated with parent nodes of the node. A parent node may correspond to a file including a buffer that stores changes to data files reachable by this parent node, and also includes dedicated storage to pointers of the child nodes. By using the data tree, the data processing system may efficiently cluster the data in the data table while reducing the number of data files that are rewritten.

Data file clustering with KD-epsilon trees

Faucet

Data ingestion using data file clustering with KD-epsilon trees

A system for clustering data into corresponding files comprises one or more processors and a memory. The one or more processors is/are configured to: 1) determine to cluster a set of data into a set of files; 2) determine a set of split points in a corresponding set of dimensions of the set of data to determine the set of files, wherein each file of the set of files has an approximate target size; and 3) store one or more items of the set of data into a corresponding file of the set of files based at least in part on the set of split points. The memory is coupled to the one or more processors and configured to provide the processor with instructions.

K-D tree balanced splitting

A method, system, and computer system for performing an operation with respect to a target table are disclosed. The method includes performing first and second jobs, obtaining one or more other resulting files based at least in part on unmatched rows, and obtaining a set of processed files based at least in part on performing a post-processing operation with respect to the set of resulting files. The set of processed files has less files than the set of resulting files. Performing the first job includes determining a set of matching target table files and storing target table information indicating for each of the set of matching target table files, a particular set of rows having matching rows. Performing the second job includes performing a matching action based on matched rows and obtaining the second job resulting file(s).

Growing community of inventors

Sunnyvale, CA, United States of America

Prakhar Jain