Abstract: An online system stores data obtained from various users of the online system. For example, the online system maintains databases for various users, with a database including data received from the user. As users provide data to the online system for storage, the online system applies data integrity checks received from users that verifies received data satisfies one or more criteria. To facilitate creation and execution of data integrity checks, the online system tunes a large language model (LLM) using executable instructions for previously generated data integrity checks and metadata describing execution of the previously generated data integrity checks. After tuning, the online system obtains one or more parameters that are input as prompts to the LLM to generate executable instructions for performing a data integrity check using the parameters.

Abstract: An online system manages various internal services using network resources or computing resources. Managing the internal services involves generating executable instructions for provisioning new services or for changing or monitoring existing services. To generate executable instructions for allocating or for monitoring network resources, the online system maintains a database of previously generated executable instructions for provisioning resources along with information about various previously generated instructions, such as comments on the executable instructions or past performance information for the previously generated instructions. To generate instructions for a new internal service, the online system tunes a large language model (LLM) with the database and provides prompts to LLM to generate executable instructions for the internal service based the prompts.

Abstract: Systems and methods for analyzing the usage of a set of workloads in a hyper-converged infrastructure are disclosed. A neural network model is trained based upon historical usage data of the set of workloads. The neural network model can make usage predictions of future demands on the set of workloads to minimize over-allocation or under-allocation of resources to the workloads.

Abstract: Disclosed herein are system, method, and computer program product embodiments for non-blocking backup for tertiary initialization in a log replay only node. An embodiment operates by performing a standard log replay on a secondary server and briefly suspending the standard log replay in response to tertiary initialization. Further, the secondary server may determine backup block information and perform a page-aligned backup process from the secondary server to a tertiary server. Additionally, the secondary server may determine log replay block information, and perform a modified log replay concurrently with the backup process based on the backup block information.

Abstract: Disclosed herein are system, method, and computer program product embodiments for utilizing an extended cache to access an object store efficiently. An embodiment operates by executing a database transaction, thereby causing pages to be written from a buffer cache to an extended cache and to an object store. The embodiment determines a transaction type of the database transaction. The transaction type can a read-only transaction or an update transaction. The embodiment determines a phase of the database transaction based on the determined transaction type. The phase can be an execution phase or a commit phase. The embodiment then applies a caching policy to the extended cache for the evicted pages based on the determined transaction type of the database transaction and the determined phase of the database transaction.

Abstract: A system and method for scaling resources of a secondary network for disaster recovery uses a disaster recovery notification from a primary resource manager of a primary network to a secondary resource manager of the secondary network to generate a scale-up recommendation for additional resources to the secondary network. The additional resources are based on latest resource demands of workloads on the primary network included in the disaster recovery notification. A scale-up operation for the additional resources is then executed based on the scale-up recommendation from the secondary resource manager to operate the secondary network with the additional resources to run the workloads on the secondary network.

Abstract: A method of restarting a virtual machine (VM) running in a cluster of hosts in a first data center, in a second data center, includes: transmitting an image of the VM to the second data center; in response to determining that a host in the cluster in which the VM was running has failed, determining whether or not there are sufficient resources in the cluster to run the VM in another host in the cluster; and upon determining that there are not sufficient resources in the cluster to run the VM in another host in the cluster, setting the VM to be inactive in the first data center, and communicating with a control plane in the second data center to set as active, and power on, a VM in the second data center using the image of the VM that has been transmitted to the second data center.

Abstract: Disclosed herein are system, method, and computer program product embodiments for utilizing an extended cache to access an object store efficiently. An embodiment operates by executing a database transaction, thereby causing pages to be written from a buffer cache to an extended cache and to an object store. The embodiment determines a transaction type of the database transaction. The transaction type can a read-only transaction or an update transaction. The embodiment determines a phase of the database transaction based on the determined transaction type. The phase can be an execution phase or a commit phase. The embodiment then applies a caching policy to the extended cache for the evicted pages based on the determined transaction type of the database transaction and the determined phase of the database transaction.

Abstract: A system and method for automatically generating recurrent neural networks for log anomaly detection uses a controller recurrent neural network that generates an output set of hyperparameters when an input set of controller parameters is applied to the controller recurrent neural network. The output set of hyperparameters is applied to a target recurrent neural network to produce a child recurrent neural network with an architecture that is defined by the output set of hyperparameters. The child recurrent neural network is then trained, and a log classification accuracy of the child recurrent neural network is computed. Using the log classification accuracy, at least one of the controller parameters used to generate the child recurrent neural network is adjusted to produce a different input set of controller parameters to be applied to the controller recurrent neural network so that a different child recurrent neural network for log anomaly detection can be generated.

Abstract: Disclosed are various implementations of cluster capacity management for infrastructure updates. In some examples, cluster hosts for a cluster can be scheduled for an update. A datacenter level workload can invoke an enter cluster maintenance mode component of a datacenter level resource scheduler by identifying a specified cluster of a datacenter. The datacenter level workload can receive an add host command to add a host for the host level update. Once the host level update is performed on hosts of the specified cluster, the datacenter level workload invokes an exit cluster maintenance mode component and implements a cluster scaling decision.

Abstract: Disclosed are various implementations of cluster capacity management for infrastructure updates. In some examples, cluster hosts for a cluster can be scheduled for an update. A component of a datacenter level resource scheduler can analyze cluster specific resource usage data to identify a cluster scaling decision for the cluster. The datacenter level resource scheduler transmits an indication that the resource scheduler is successfully invoked. Cluster hosts can then be updated.

Abstract: A method of restarting a virtual machine (VM) running in a cluster in a first data center, in a second data center, includes: transmitting images of VMs, including a first VM, running in the cluster of hosts at a first point in time to the second data center for replication in the second data center; generating difference data representing a difference in an image of the first VM at a second point in time and the image of the first VM at the first point in time; transmitting the difference data to the second data center; setting the first VM to be inactive in the first data center; and communicating with a control plane in the second data center to set as active, and power on, a VM in the second data center using the replicated image of the first VM updated with the difference data.

Abstract: Disclosed herein are system, method, and computer program product embodiments for utilizing an extended cache to access an object store efficiently. An embodiment operates by executing a database transaction, thereby causing pages to be written from a buffer cache to an extended cache and to an object store. The embodiment determines a transaction type of the database transaction. The transaction type can a read-only transaction or an update transaction. The embodiment determines a phase of the database transaction based on the determined transaction type. The phase can be an execution phase or a commit phase. The embodiment then applies a caching policy to the extended cache for the evicted pages based on the determined transaction type of the database transaction and the determined phase of the database transaction.

Abstract: Disclosed are various implementations of cluster capacity management for infrastructure updates. In some examples, cluster hosts for a cluster can be scheduled for an update. A component of a datacenter level resource scheduler can analyze cluster specific resource usage data to identify a cluster scaling decision for the cluster. The datacenter level resource scheduler transmits an indication that the resource scheduler is successfully invoked. Cluster hosts can then be updated.

Abstract: A system and method for scaling resources of a secondary network for disaster recovery uses a disaster recovery notification from a primary resource manager of a primary network to a secondary resource manager of the secondary network to generate a scale-up recommendation for additional resources to the secondary network. The additional resources are based on latest resource demands of workloads on the primary network included in the disaster recovery notification. A scale-up operation for the additional resources is then executed based on the scale-up recommendation from the secondary resource manager to operate the secondary network with the additional resources to run the workloads on the secondary network.

Abstract: Systems and methods for analyzing a customer deployment in a converged or hyper-converged infrastructure are disclosed. A machine learning model is trained based upon historical usage data of other customer deployments. A k-means clustering is performed to generate a prediction as to whether a deployment is configured for optimal failover. Recommendations to improve failover performance can also be generated.

Abstract: A method of restarting a virtual machine running in a cluster of hosts in a first data center, in a second data center, wherein each virtual machine is assigned a priority level, includes: transmitting virtual machines images running in the cluster at a first time to the second data center; selecting virtual machines to be restarted in the second data center according to priority levels assigned; and for each selected virtual machine, (a) generating difference data in an image of the selected virtual machine at a second time and at the first time, (b) transmitting the difference data to the second data center, (c) setting the virtual machine inactive in the first data center, and (d) communicating with the second data center to set as active; and power on, a virtual machine in the second data center using the image of the virtual machine transmitted to the second data center.

Abstract: Disclosed herein are system, method, and computer program product embodiments for non-blocking backup for tertiary initialization in a log replay only node. An embodiment operates by performing a standard log replay on a secondary server and briefly suspending the standard log replay in response to tertiary initialization. Further, the secondary server may determine backup block information and perform a page-aligned backup process from the secondary server to a tertiary server. Additionally, the secondary server may determine log replay block information, and perform a modified log replay concurrently with the backup process based on the backup block information.

Abstract: A number of hosts in a logical cluster is adjusted up or down in an elastic manner by tracking membership of hosts in the cluster using a first data structure and tracking membership of hosts in a spare pool using a second data structure, and upon determining that a triggering condition for adding another host is met and that all hosts in the cluster are being used, selecting a host from the spare pool, and programmatically adding an identifier of the selected host to the first data structure and programmatically deleting the identifier of the selected host from the second data structure.

Abstract: Disclosed herein are system, method, and computer program product embodiments for non-blocking backup for tertiary initialization in a log replay only node. An embodiment operates by performing a standard log replay on a secondary server and briefly suspending the standard log replay in response to tertiary initialization. Further, the secondary server may determine backup block information and perform a page-aligned backup process from the secondary server to a tertiary server. Additionally, the secondary server may determine log replay block information, and perform a modified log replay concurrently with the backup process based on the backup block information.