Abstract: A method and apparatus are described for assigning mastership of nodes to data blocks. A method involves connecting each session of a plurality of sessions to a particular node of a cluster of nodes based on services associated with the plurality of sessions. Each session of the plurality of sessions is associated with a respective service of a plurality of services. The method also involves collecting service-based access statistics aggregated by service and ranges of data block addresses. Each range corresponds to one or more contiguous subrange of data block addresses. The method further involves assigning mastership of the nodes to the data blocks having addresses within said ranges of data block addresses based on services associated with the nodes and the service-based access statistics.

Abstract: Techniques related to instance recovery using Bloom filters are disclosed. A multi-node node database management system (DBMS) includes a first database server instance and a second database server instance. A recovery set includes a set of data blocks that have been modified by a first database server instance and not persisted. A Bloom filter is generated to indicate whether data blocks are excluded from the recovery set. The Bloom filter is sent to the second database server instance, which determines whether the Bloom filter indicates that a particular data block is excluded from the recovery set. Based on determining that the Bloom filter indicates that the particular data block is excluded from the recovery set, access to the particular data block is granted.

Abstract: Techniques are described herein for quick identification of a set of units of data for which recovery operations are to be performed to redo or undo changes made by the failed node. When a lock is requested by an instance, lock information for the lock request is replicated by another instance. If the instance fails, the other instance may use the replicated lock information to determine a set of data blocks for recovery operations. The set of data blocks is available in memory of a recovery instance when a given node fails, and does not have to be completely generated by scanning a redo log.

Abstract: Techniques related to cache recovery for failed database instances are disclosed. A first database instance and a second database instance share a primary persistent storage and a secondary persistent storage. Each database instance stores, in volatile memory, a respective primary cache of a respective set of data stored on the primary persistent storage. Each database instance also stores, in volatile memory, a respective set of header data. Further, each database instance moves the respective set of data from the respective primary cache to a respective secondary cache on the secondary persistent storage. Still further, each database instance stores, on the secondary persistent storage, a respective set of persistent metadata. When the first database instance becomes inoperative, the second database instance retrieves, from the secondary persistent storage, persistent metadata corresponding to data stored in a secondary cache of the first database instance.

Abstract: According to embodiments, reconfiguration of lock masters for a cluster of nodes is optimized using coordinated hash indexes to index into the master hash table and into local hash tables stored on the nodes of the cluster. A DBMS uses a hybrid hash index, a portion of which represents a corresponding master hash index, to index into both the master hash table and into a local hash table for a given lock. The hash index used to store lock metadata in a particular local hash table bucket, on a particular node, encodes the lock master index, for a master hash table, to which the locks in the local hash table bucket correspond. Only the portions of the local hash table on the lock master that correspond to the index of the master hash table bucket need to be scanned in order to perform needed tasks for lock master reconfiguration.

Abstract: Techniques are described herein for quick identification of a set of units of data for which recovery operations are to be performed to redo or undo changes made by the failed node. When a lock is requested by an instance, lock information for the lock request is replicated by another instance. If the instance fails, the other instance may use the replicated lock information to determine a set of data blocks for recovery operations. The set of data blocks is available in memory of a recovery instance when a given node fails, and does not have to be completely generated by scanning a redo log.

Abstract: According to one aspect of the invention, a read-write instance of a multiple-instance database system receives a database statement for execution. The multiple-instance database system comprises a database shared by a plurality read-mostly instances that request only read access to a plurality of user database objects in the database and a plurality of read-write instances that request both read access and write access to the plurality of user database objects. The read-write instance selects one or more read-mostly instances, in the plurality of read-mostly instances, to perform one or more operations that require no write access to the plurality of user data objects.

Abstract: Techniques related to instance recovery using Bloom filters are disclosed. A multi-node node database management system (DBMS) includes a first database server instance and a second database server instance. A recovery set includes a set of data blocks that have been modified by a first database server instance and not persisted. A Bloom filter is generated to indicate whether data blocks are excluded from the recovery set. The Bloom filter is sent to the second database server instance, which determines whether the Bloom filter indicates that a particular data block is excluded from the recovery set. Based on determining that the Bloom filter indicates that the particular data block is excluded from the recovery set, access to the particular data block is granted.

Abstract: Techniques related to cache recovery for failed database instances are disclosed. A first database instance and a second database instance share a primary persistent storage and a secondary persistent storage. Each database instance stores, in volatile memory, a respective primary cache of a respective set of data stored on the primary persistent storage. Each database instance also stores, in volatile memory, a respective set of header data. Further, each database instance moves the respective set of data from the respective primary cache to a respective secondary cache on the secondary persistent storage. Still further, each database instance stores, on the secondary persistent storage, a respective set of persistent metadata. When the first database instance becomes inoperative, the second database instance retrieves, from the secondary persistent storage, persistent metadata corresponding to data stored in a secondary cache of the first database instance.

Abstract: According to embodiments, reconfiguration of lock masters for a cluster of nodes is optimized using coordinated hash indexes to index into the master hash table and into local hash tables stored on the nodes of the cluster. A DBMS uses a hybrid hash index, a portion of which represents a corresponding master hash index, to index into both the master hash table and into a local hash table for a given lock. The hash index used to store lock metadata in a particular local hash table bucket, on a particular node, encodes the lock master index, for a master hash table, to which the locks in the local hash table bucket correspond. Only the portions of the local hash table on the lock master that correspond to the index of the master hash table bucket need to be scanned in order to perform needed tasks for lock master reconfiguration.

Abstract: A method and apparatus are described for assigning mastership of nodes to data blocks. A method involves connecting each session of a plurality of sessions to a particular node of a cluster of nodes based on services associated with the plurality of sessions. Each session of the plurality of sessions is associated with a respective service of a plurality of services. The method also involves collecting service-based access statistics aggregated by service and ranges of data block addresses. Each range corresponds to one or more contiguous subrange of data block addresses. The method further involves assigning mastership of the nodes to the data blocks having addresses within said ranges of data block addresses based on services associated with the nodes and the service-based access statistics.

Abstract: Processes, computer-readable media, and machines are disclosed for reducing a likelihood that active functional components fail in a computing system. An active monitoring component receives metrics associated with different active functional components of a computing system. The different active functional components contribute to different functionalities of the system. Based at least in part on the metrics associated with a particular active functional component, the active monitoring component determines that the particular active functional component has reached a likelihood of failure but has not failed. In response to determining that the particular active functional component has reached the likelihood of failure but has not failed, the active monitoring component causes a set of actions that are predicted to reduce the likelihood of failure.

Abstract: Techniques for taking one or more nodes of a multi-node database system offline are provided. Initially, the multi-node database system uses a first data access management system (e.g., a distributed lock management system) to manage access to data stored in a database. In response to determining that one or more nodes need to be taken offline or otherwise become unavailable, the remaining nodes implement a second data access management system (e.g., an affinity-based access system) to manage access to data stored in the database. After the offline nodes are brought back online, all the nodes in the multi-node database system implement the first data access management system again. The transitions between the two data access management system may involve multiple phases that involve communicating different types of information between nodes that are taken offline and nodes that are not taken offline.

Abstract: Techniques are provided for managing caches in a system with multiple caches that may contain different copies of the same data item. Specifically, techniques are provided for coordinating the write-to-disk operations performed on such data items to ensure that older versions of the data item are not written over newer versions, and to reduce the amount of processing required to recover after a failure. Various approaches are provided in which a master is used to coordinate with the multiple caches to cause a data item to be written to persistent storage. Techniques are also provided for transferring data items and locks associated with the data items from one node to another.

Abstract: Techniques for mastering resources in a cluster of nodes are provided. A global backup lock manager (GBLM) is maintained for a cluster of nodes that implement distributed lock management. Before a server instance is taken down, for example, for maintenance purposes, such as installing a new version of the server instance code, the mastership information that the server instance stores is reflected in the mastership information maintained by the GBLM. Thus, shutting down the server instance does not involve remastering the resources mastered by the server instance. As a result, shutting down the server instance may take minimal time.

Abstract: Techniques for taking one or more nodes of a multi-node database system offline are provided. Initially, the multi-node database system uses a first data access management system (e.g., a distributed lock management system) to manage access to data stored in a database. In response to determining that one or more nodes need to be taken offline or otherwise become unavailable, the remaining nodes implement a second data access management system (e.g., an affinity-based access system) to manage access to data stored in the database. After the offline nodes are brought back online, all the nodes in the multi-node database system implement the first data access management system again. The transitions between the two data access management system may involve multiple phases that involve communicating different types of information between nodes that are taken offline and nodes that are not taken offline.

Abstract: Techniques for mastering resources in a cluster of nodes are provided. A global backup lock manager (GBLM) is maintained for a cluster of nodes that implement distributed lock management. Before a server instance is taken down, for example, for maintenance purposes, such as installing a new version of the server instance code, the mastership information that the server instance stores is reflected in the mastership information maintained by the GBLM. Thus, shutting down the server instance does not involve remastering the resources mastered by the server instance. As a result, shutting down the server instance may take minimal time.

Abstract: Techniques for monitoring resources of a computer system are provided. A monitoring process collects and reports utilization data for one or more resources of a computer system, such as CPU, memory, disk I/O, and network I/O. Instead of reporting just an average of the collected data over a period of time (e.g., 10 seconds), the monitoring process at least reports individually collected resource utilization values. If one or more of the utilization values exceed specified thresholds for the respective resources, then an alert may be generated. In one approach, the monitoring process is made a real-time priority process in the computer system to ensure that the memory used by the monitoring process is not swapped out of memory. Also, being a real-time priority process ensures that the monitoring process obtains a CPU in order collect resource utilization data even when the computer system is in a starvation mode.

Abstract: Processes, computer-readable media, and machines are disclosed for reducing a likelihood that active functional components fail in a computing system. An active monitoring component receives metrics associated with different active functional components of a computing system. The different active functional components contribute to different functionalities of the system. Based at least in part on the metrics associated with a particular active functional component, the active monitoring component determines that the particular active functional component has reached a likelihood of failure but has not failed. In response to determining that the particular active functional component has reached the likelihood of failure but has not failed, the active monitoring component causes a set of actions that are predicted to reduce the likelihood of failure.

Abstract: Processes, computer-readable media, and machines are disclosed for reducing a likelihood that active functional components fail in a computing system. An active monitoring component receives metrics associated with different active functional components of a computing system. The different active functional components contribute to different functionalities of the system. Based at least in part on the metrics associated with a particular active functional component, the active monitoring component determines that the particular active functional component has reached a likelihood of failure but has not failed. In response to determining that the particular active functional component has reached the likelihood of failure but has not failed, the active monitoring component causes a set of actions that are predicted to reduce the likelihood of failure.