Abstract: Methods, systems, and devices for techniques to balance log structured merge trees are described. A computing system may rebalance a tree structure having an ordered set of leaf nodes by splitting or joining leaf nodes of the tree structure. To split a leaf node, the computing system may select a key to evenly partition key and value data stored in the leaf. The computing system may place each key block of the leaf node having keys less than or equal to the selected key in a first new leaf node, and may place each key block of the leaf node having keys greater than the selected key in a second new leaf node. To join leaf nodes of the tree structure, the computing system may place each key block and each value block of the leaf node and the adjacent leaf node in a new leaf node.

Abstract: A system includes a memory device a processing device, operatively coupled to the memory device. The processing device is configured to receive a request to identify a target key in a key-value store based on a specified key; identify, in at least one of a plurality of sequences of memory keys, the target key based on the specified key, where the plurality of sequences of memory keys includes a sequence of memory keys and a sequence of media keys that comprises designated media keys, where each of the designated media keys is designated as being deleted, where the identifying comprises comparing each of the designated media keys to the specified key; and perform a database operation using the target key.

Abstract: Methods, systems, and devices for techniques to balance log structured merge trees are described. A computing system may rebalance a tree structure having an ordered set of leaf nodes by splitting or joining leaf nodes of the tree structure. To split a leaf node, the computing system may select a key to evenly partition key and value data stored in the leaf. The computing system may place each key block of the leaf node having keys less than or equal to the selected key in a first new leaf node, and may place each key block of the leaf node having keys greater than the selected key in a second new leaf node. To join leaf nodes of the tree structure, the computing system may place each key block and each value block of the leaf node and the adjacent leaf node in a new leaf node.

Abstract: An event trigger to flush key-value data from a volatile memory device is detected. In response to detection of the event trigger, a determination is made whether the key-value data comprises uncommitted data for one or more active database transactions. If the key-value data comprises uncommitted data for one or more active database transactions, the uncommitted data is moved to a hold aside buffer in the volatile memory device, and committed data in the key-value data is stored on a non-volatile memory device.

Abstract: An example method of cursor seek operations using deleted record spans in memory sub-systems comprises: receiving, by a processing device, a request to search for a first data element of a key-value data store, the request specifying a first key associated with the first data element; determining that the first key is within a range of deleted data elements comprising a plurality of deleted data elements of the key-value data store; identifying a tail key associated with a tail element of the range of deleted data elements; identifying a second data element in the key-value data store, wherein the second data element is associated with a second key that follows the tail key in a specified order of keys; and providing the second data element in response to the request.

Abstract: A request to retrieve data from a memory device of a memory sub-system can be received from a machine learning (ML) framework executing on a host system, where the data comprises a plurality of logical partitions. A set of parallel I/O threads can be initiated to retrieve the data from the memory device, where each I/O thread of the set of parallel I/O threads retrieves a different portion of the data from a different corresponding logical partition and stores the different portion of the data in a I/O buffer of a set of I/O buffers corresponding to the set of I/O threads in parallel. The different portion of the data can be successively provided from each I/O buffer to the ML framework, where the set of parallel I/O threads is to continually retrieve the data from the memory device until all of the data from the logical partitions has been provided to the ML framework.

Abstract: A key-value pair comprising a key and a value associated with the key is stored in a current segment of a data container. The key comprises a segment identifier identifying the current segment. The data container comprises a plurality of segments residing on a non-volatile memory device. Responsive to determining that a size of the current segment exceeds a threshold, a new segment is added to the container. The plurality of key-value pairs having respective current segment identifiers referencing one or more least recently created segments of the data container is deleted.

Abstract: A request to retrieve data from a memory device of a memory sub-system can be received from a machine learning (ML) framework executing on a host system, where the data comprises a plurality of logical partitions. A set of parallel I/O threads can be initiated to retrieve the data from the memory device, where each I/O thread of the set of parallel I/O threads retrieves a different portion of the data from a different corresponding logical partition and stores the different portion of the data in a I/O buffer of a set of I/O buffers corresponding to the set of I/O threads in parallel. The different portion of the data can be successively provided from each I/O buffer to the ML framework, where the set of parallel I/O threads is to continually retrieve the data from the memory device until all of the data from the logical partitions has been provided to the ML framework.

Abstract: A prefix probe component receives a request to perform a cursor operation to search for one or more data elements of a key-value data store, the request comprising a key identifier associated with the one or more data elements, and wherein the key-value data store comprises a tree structure with a plurality of nodes; traverses a portion of the plurality of nodes to identify data elements in the key-value data store that match the key identifier; determines whether a number of the data elements that match the key identifier satisfies a threshold condition; and responsive to determining that the number of data elements satisfies the threshold condition, performs the cursor operation for the data elements that match the key identifier.

Abstract: A request to retrieve data from a memory device of a memory sub-system can be received from a machine learning (ML) framework executing on a host system, where the data comprises a plurality of logical partitions. A set of parallel I/O threads can be initiated to retrieve the data from the memory device, where each I/O thread of the set of parallel I/O threads retrieves a different portion of the data from a different corresponding logical partition and stores the different portion of the data in a I/O buffer of a set of I/O buffers corresponding to the set of I/O threads in parallel. The different portion of the data can be successively provided from each I/O buffer to the ML framework, where the set of parallel I/O threads is to continually retrieve the data from the memory device until all of the data from the logical partitions has been provided to the ML framework.

Abstract: A key-value pair comprising a key and a value associated with the key is stored in a current segment of a data container. The key comprises a segment identifier identifying the current segment. The data container comprises a plurality of segments residing on a non-volatile memory device. Responsive to determining that a size of the current segment exceeds a threshold, a new segment is added to the container. The plurality of key-value pairs having respective current segment identifiers referencing one or more least recently created segments of the data container is deleted.

Abstract: Aspects of the present disclosure provide for operations of a key-value tree data structure that: merge key-value sets within a given node by merging and rewriting key blocks of the key-value sets while rewriting or deferring rewrite of value blocks of the merged key-value sets based on whether one or more child nodes of the given node comprise a leaf node; and move one or more portions of the merged key-value set into one or more child nodes of the given node.

Abstract: An example method of cursor seek operations using deleted record spans in memory sub-systems comprises: receiving, by a processing device, a request to search for a first data element of a key-value data store, the request specifying a first key associated with the first data element; determining that the first key is within a range of deleted data elements comprising a plurality of deleted data elements of the key-value data store; identifying a tail key associated with a tail element of the range of deleted data elements; identifying a second data element in the key-value data store, wherein the second data element is associated with a second key that follows the tail key in a specified order of keys; and providing the second data element in response to the request.

Abstract: A request to retrieve data from a memory device of a memory sub-system can be received from a machine learning (ML) framework executing on a host system, where the data comprises a plurality of logical partitions. A set of parallel I/O threads can be initiated to retrieve the data from the memory device, where each I/O thread of the set of parallel I/O threads retrieves a different portion of the data from a different corresponding logical partition and stores the different portion of the data in a I/O buffer of a set of I/O buffers corresponding to the set of I/O threads in parallel. The different portion of the data can be successively provided from each I/O buffer to the ML framework, where the set of parallel I/O threads is to continually retrieve the data from the memory device until all of the data from the logical partitions has been provided to the ML framework.

Abstract: Aspects of the present disclosure provide for operations of a key-value tree data structure that: merge key-value sets within a given node by merging and rewriting key blocks of the key-value sets while rewriting or deferring rewrite of value blocks of the merged key-value sets based on whether one or more child nodes of the given node comprise a leaf node; and move one or more portions of the merged key-value set into one or more child nodes of the given node.