Abstract: Optimistic concurrency may be implemented for multiple writers to a database. Writes to a database from different writers may succeed unless a conflict is detected at the data store for the database. For detected conflicts, a request to resolve the conflict between writes may be submitted to a primary writer, which may determine and provide a resolution of the conflict to the conflicting writers. The data store may then be updated to commit the selected write identified by the resolution of the conflict.

Abstract: Change data capture is performed asynchronously for direct transfer to a stream processor. When updates to a database as part of database transactions are performed, a transaction log is generated that describes the database transactions. The transaction log may be evaluated to determine indications of changes included in the database transactions. The indications of the changes may be stored in memory according to the database transactions that include the indications of the changes. When a database transaction is determined to be committed to the database, the change indications may be sent directly from the memory to the stream processor.

Abstract: Methods, systems, and computer-readable media for dynamic prefetching for database queries are disclosed. A query of a database is started according to a first prefetch policy. Before completing the query, the first prefetch policy is changed to a second prefetch policy. A portion of the query is performed according to the second prefetch policy.

Abstract: Queries to a consistent view of a data set across different types of query engines are performed. Portions of a query received at a first type of query engine that can be performed at a different type of a query engine can be identified. A consistent view of a data set stored in a common data store can be provided by the first type of query engine to the second query engine. When the different type of query engine performs the portion of the query, the different type of query engine applies the consistent view of the data set. A result for the query can be returned by the first type of query engine based on results from the different type of query engine.

Abstract: Updates for transactions to a replication log for a database may be amortized. As updates are received as part of an active transaction, replication log records may be generated. When the size of replication log records exceeds a transaction size threshold, the replication log records may be stored in a remote data store instead of a local data store. When a request to commit the active transaction is received, an replication log for the database may be updated to include the replication log records from a transaction data object in the remote data store.

Abstract: Methods, systems, and computer-readable media for dynamic prefetching for database queries are disclosed. A query of a database is started according to a first prefetch policy. Before completing the query, the first prefetch policy is changed to a second prefetch policy. A portion of the query is performed according to the second prefetch policy.

Abstract: Techniques for making machine learning inference calls for database query processing are described. In some embodiments, a method of making machine learning inference calls for database query processing may include generating a first batch of machine learning requests based at least on a query to be performed on data stored in a database service, wherein the query identifies a machine learning service, sending the first batch of machine learning requests to an input buffer of an asynchronous request handler, the asynchronous request handler to generate a second batch of machine learning requests based on the first batch of machine learning requests, and obtaining a plurality of machine learning responses from an output buffer of the asynchronous request handler, the machine learning responses generated by the machine learning service using a machine learning model in response to receiving the second batch of machine learning requests.

Abstract: Techniques for a context preservation during server restart are described. A restart of a database server having at least one open communications session with a database client is initiated. The database server can be executed by a compute instance hosted by a computer system of a provider network. A restarted database server is launched. The restarted database server obtains session state data to resume the at least one open communications session with the database client, the session state data including one or more session variables. The restarted database server receives a transaction from the database client.

Abstract: Storage level query processing may be implemented for processing database queries. Nodes that can access a database may perform parallel processing for at least a portion of a database query. An indication may be received that specifies parallel processing for the database query. The nodes can then be caused to perform the portion of the query as part of providing a result in response to the database query instead of a node, such as a query engine node, that received the database query.

Abstract: A service provider network implements seamless scaling via proxy replay of session state. Upon a trigger, such as a determination to scale a server, a scaled server may be spun up and an identifier of the scaled server provided to a first (existing) server. The first server sends the identification of the second server, and session state information for each of the connections between the first server and the request router, to the request router. For each of the connections, the request router establishes a new connection between the request router and the second (scaled) server, and replays the session state information for the connection to the second server. The request router then routes traffic between each existing client connection (e.g., the same existing client connection which carried traffic delivered to the first server) and the corresponding new connection to the second server.

Abstract: Proxy-based scaling may be performed for databases. A proxy may be implemented for a database that can establish a connection between the proxy and a database engine to perform a database queries received from a client at the proxy. A scaling event may be detected for the database responsive to which the proxy may establish a connection with a new database engine which may, in some embodiments, have different capabilities or resources that address the features or criteria that triggered the scaling event. Session state may be copied from the database engine to the new database engine so that the new database engine may be able to provide access to the database on behalf of requests received from the client through the proxy.

Abstract: Alternative portions of a query plan to perform a query may be selected for processing partial results of a query. A query engine may generate a plan to perform a query that includes an operation that selects between different portions of the plan according to input data. The input data may be a partial result of the query generated by the data store instead of the query engine and provided to the query engine. The input data may be an individual item of the data set that is modified to a different version within a consistent view of the data set for the query.

Abstract: Techniques for making machine learning inference calls for database query processing are described. In some embodiments, a method of making machine learning inference calls for database query processing may include generating a first batch of machine learning requests based at least on a query to be performed on data stored in a database service, wherein the query identifies a machine learning service, sending the first batch of machine learning requests to an input buffer of an asynchronous request handler, the asynchronous request handler to generate a second batch of machine learning requests based on the first batch of machine learning requests, and obtaining a plurality of machine learning responses from an output buffer of the asynchronous request handler, the machine learning responses generated by the machine learning service using a machine learning model in response to receiving the second batch of machine learning requests.

Abstract: Proxy-based scaling may be performed for databases. A proxy may be implemented for a database that can establish a connection between the proxy and a database engine to perform a database queries received from a client at the proxy. A scaling event may be detected for the database responsive to which the proxy may establish a connection with a new database engine which may, in some embodiments, have different capabilities or resources that address the features or criteria that triggered the scaling event. Session state may be copied from the database engine to the new database engine so that the new database engine may be able to provide access to the database on behalf of requests received from the client through the proxy.

Abstract: Systems and methods are disclosed to implement, in a database system, a database engine node that generates write tracking transactions to atomically enforce cross-page constraints for write operations. In embodiments, the database system may include multiple database engine nodes that optimistically perform page writes without checking for potential conflicts at a shared storage system. The storage system is configured to detect and report conflicting writes on the same page. To cause the storage system to report cross-page conflicts for a write, the database engine node creates a transaction that includes the write and a write tracker (WT) operation for a tracked page. The WT operation will cause the storage system to report any conflicting writes to the tracked page before the actual write is accepted by the storage system. If a reported conflict is not favorably resolved, the database engine node will abort the entire transaction, including the actual write.

Abstract: Batch processing of undo log records may be performed for rolling back a database transaction. When a rollback operation for a database transaction is performed, undo log records from an undo log to rollback the database transaction are divided into different groups of undo log records. Different requests for the different undo log records are sent to a storage system to be performed. If a conflict is detected for an undo log record in a group, an indication of the conflicting undo log record may be provided to indicate the conflict. The group of undo log records may be resent with a request to reapply the undo log records in the group. When all groups have been identified as successfully completed, then the rollback operation may be considered completed.

Abstract: This application relates to multi-level replication filtering for a distributed database. A primary node of the database can generate log records that describe write requests performed to the database. The primary node can, in some embodiments, identify a portion of the data of the database that is stored at a second node, such as data stored by a read replica node that can be used in responding to read requests from clients. The primary node can select log records that are applicable to this identified portion of the data to send to the second node. Log records are not selected based on a determination that those unselected log records are not applicable to the identified portion of the data. The primary node can then send the selected log records to the second node, for the second node to apply the log records to its stored database data.

Abstract: A hybrid query engine using different query engines may be implemented to provide access to a database. A database query to a database may be received at a hybrid query engine. A plan to perform the database query may be generated that when performed by the hybrid query engine causes one query engine to perform an assigned portion of the plan by accessing data that describes the database. Another portion of the plan assigned to another query engine may be performed by the hybrid query engine to cause the other query engine to perform the other portion of the plan by accessing the same data that describes the database.

Abstract: Transactional scans of portions of a database may be performed. When a query is received that causes a scan of a database, different portions of the database may be scanned within separate transactions. Operations applicable to perform the scan the records of the different portions of the database may be invoked once within the transactions instead of invoking the operations individually for each of the records in the different portions. Results of the query based on the scanned portions may then be provided in response to the query.

Abstract: A storage engine may implement dynamic splitting of contentious data pages. Data pages may store data for a table of a data store as part of an indexing structure for the table. Access to the table may be provided by locating the corresponding data pages via the indexing structure. Access contention for different data pages may be monitored. Data pages may be identified for splitting based on the monitoring. A split operation for an identified data page may be formed to store the data on the identified data page on two different data pages so that subsequent access requests for the data are divided between the two data pages. Monitoring of access contention may also be performed to identify data pages for merging in order to consolidate access requests to a single data page.