Snowflake Inc.

Snowflake Lnc.

Prasanna Rajaperumal

Jiaqi Yan

Thierry Cruanes

Igor Demura

William H Waddington

Abdul Q Munir

Jeffrey Rosen

Benoit Dageville

Varun Ganesh

Libo Wang

Allison Waingold Lee

Tyler Jones

Tyler Arthur Akidau

Daniel E Sotolongo

Yongsik Yoon

Daniel Mills

Shrainik Jain

Boyung Lee

Leon Papke

Fabian Hueske

Saiyang Gou

Wenhao Song

Systems, methods, and devices for sharing materialized views in multiple tenant database systems. A method includes defining a materialized view over a source table that is associated with a first account of a multiple tenant database. The method includes defining cross-account access rights to the materialized view to a second account such that that second account can read the materialized view without copying the materialized view. The method includes modifying the source table for the materialized view. The method includes identifying whether the materialized view is stale with respect to the source table by merging the materialized view and the source table.

Sharing materialized views in multiple tenant database systems

Systems, methods, and devices for incrementally refreshing a materialized view are disclosed. A method includes generating a materialized view based on a source table. The method includes merging the source table and the materialized view to generate a merged table to identify whether an update has been executed on the source table that is not reflected in the materialized view. The method includes, in response to detecting an update made to the source table that is not reflected in the materialized view, applying the update to the materialized view.

Incremental refresh of a materialized view

Embodiments of the present disclosure may provide a dynamic query execution model. This query execution model may provide acceleration by scaling out parallel parts of a query (also referred to as a fragment) to additional computing resources, for example computing resources leased from a pool of computing resources. Execution of the parts of the query may be coordinated by a parent query coordinator, where the query originated, and a fragment query coordinator.

Scalable query processing

Provided herein are systems and methods for a database object (e.g., materialized table) configuration including using the database object for incremental maintenance of query results. For example, a method includes creating a materialized table using a table definition. The table definition includes a query on one or more base tables and a lag duration value. The lag duration value indicates a maximum time period that a result of a prior refresh of the query can lag behind a current time instance. A delta set for the one or more base tables is determined. The delta set includes at least one change applied to the one or more base tables since the prior refresh of the query. The delta set is merged with the materialized table to perform an incremental update of the materialized table.

Incremental maintenance of query results

Provided herein are systems and methods for a database object (e.g., materialized table) configuration including scheduling refreshes of the materialized table. For example, a method includes determining a dependency graph for a first MT. The dependency graph comprises a second MT from which the first MT depends. The first MT includes a query on one or more base tables and a lag duration value. The lag duration value indicates a maximum time period that a result of a prior refresh of the query can lag behind a current time instance. A tick period is selected for a set of ticks based on the lag duration value. The set of ticks corresponds to a set of aligned time instances. Refresh operations are scheduled for the first and second MTs at corresponding time instances from the set of aligned time instances. The corresponding time instances are separated by the tick period.

Execution and consistency model for materialized tables

Embodiments of the present disclosure may provide dynamic and fair assignment techniques for allocating resources on a demand basis. Assignment control may be separated into at least two components: a local component and a global component. Each component may have an active dialog with each other; the dialog may include two aspects: 1) a demand for computing resources, and 2) a total allowed number of computing resources. The global component may allocate resources from a pool of resources to different local components, and the local components in turn may assign their allocated resources to local competing requests. The allocation may also be throttled or limited at various levels.

Flexible computing

A cluster view method of a database to perform compaction and clustering of database objects, such as database materialized view is shown. The database can comprise a cache to store changes to storage units of tables of the database objects. The cluster view method can implement clustering to remove data based on the cache and clustering to group the data of the materialized view.

Maintenance of clustered materialized views on a database system

Disclosed herein are embodiments of systems and methods for selecting partitions for reclustering based on distribution of overlapping partitions. In an example, a database platform makes a determination to at least partially recluster a database table that includes data stored across a plurality of partitions. The database platform responsively selects a subset of the partitions. The selecting of the subset includes identifying a point on a domain of a clustering key that corresponds to a local maximum of overlapping partitions, and also includes selecting the subset from among a group of overlapping partitions. The group includes at least one partition that overlaps the identified point on the domain of the clustering key. Each partition in the selected subset is above a reduction goal of overlapping partitions. The database platform at least partially reclusters the selected subset based on the clustering key.

Selecting partitions for reclustering based on distribution of overlapping partitions

Queries executed against a materialized view can execute up to orders of magnitude faster than equivalent queries on a source (or base) table. However, although a query can reference a materialized view directly, a user (e.g., query author) may not know about a relevant materialized view. Moreover, if a source table has multiple materialized views generated, the user may not know which materialized view to reference in the query. Thus, embodiments of the present disclosure provide techniques for automatically rewriting queries directed to a source table to utilize existing materialized views.

Transient materialized view rewrite

Resource provisioning systems and methods are described. In an embodiment, a system includes a plurality of shared storage devices collectively storing database data, an execution platform, and a compute service manager. The compute service manager is configured to determine a task to be executed in response to a trigger event and determine a query plan for executing the task, wherein the query plan comprises a plurality of discrete subtasks. The compute service manager is further configured to assign the plurality of discrete subtasks to one or more nodes of a plurality of nodes of the execution platform, determine whether execution of the task is complete, and in response to determining the execution of the task is complete, store a record in the plurality of shared storage devices indicating the task was completed.

Resource provisioning in database systems

Embodiments of the present disclosure may provide a dynamic query execution model with fault tolerance and failure recovery techniques. Embodiments of the present disclosure may utilize checkpoints to map processed output files to their corresponding input files. Therefore, if an error occurs in processing one or more files, the system may only need to reschedule processing of selected file(s).

Checkpoints in batch file processing

Disclosed herein are systems and methods for incremental reclustering of database tables based on local maxima of partition overlap. In an embodiment, a database platform makes a determination, based on one or more incremental-reclustering criteria, to incrementally recluster a database table, which has a clustering key and which is stored across a plurality of partitions. In response to making the determination, the database platform selects a subset of the partitions, and at least incrementally reclusters the selected subset. The selecting of the subset includes identifying a local maximum of a quantity of overlapping partitions in the plurality of partitions with respect to a domain of the clustering key of the table, where the overlapping partitions overlap with respect to the clustering key.

Growing community of inventors

San Mateo, CA, United States of America

Prasanna Rajaperumal