Abstract: A node within a group of participant nodes begins an election by sending a vote request to the other nodes in the group. The vote request sets an input term argument to a future term value without incrementing the actual current term value. The current term value at each participant node is only incremented in response to a successful leadership change. At startup time, a candidate node issues a vote request with a non-disruptive election type. An established leader automatically rejects a non-disruptive vote request. A heartbeat loss vote request is rejected by each receiving node if its own heartbeat timeout does not exceed a predetermined limit. A mandatory vote request informs the leader node that it should stop requesting new workload. This is used in manual leadership transition to make sure that the old leader does not accept new transactions during the leadership transition.

Abstract: A consensus protocol-based replication approach is provided. For each change operation performed by a leader server on a copy of the database, the leader server creates a replication log record and returns a result to the client. The leader does not wait for consensus for the change operation from the followers. For a commit, the leader creates a commit log record and waits for consensus. Thus, the leader executes database transactions asynchronously, performs replication of change operations asynchronously, and performs replication of transaction commits synchronously.

Abstract: A lead-sync log record is used to synchronize the replication logs of follower shards to the leader shard. In response to a failure to determine that there is a consensus for a database transaction commit operation after a shard server becomes a new leader, the new leader shard performs a sync operation using the lead-sync log record to synchronize replication logs of the follower shards to the replication log of the new leader. A shard server identifies a first transaction having a first log record but not a post-commit log record in the replication log, defines a recovery window in the replication log starting at the first log record of the identified first transaction and ending at the lead-sync log record, identifies a set of transactions to be recovered, and performs a recovery action on the set of transactions to be recovered.

Abstract: A consensus protocol-based replication approach is provided. Chunks are grouped into replication units (RUs) to optimize replication efficiency. Chunks may be assigned to RUs based on load and replication throughput. Splitting and merging RUs do not interrupt concurrent user workload or require routing changes. Transactions spanning chunks within an RU do not require distributed transaction processing. Each replication unit has a replication factor (RF), which refers to the number of copies/replicas of the replication unit, and an associated distribution factor (DF), which refers to the number of servers taking over the workload from a failed leader server. RUs may be placed in rings of servers, where the number of servers in a ring is equal to the replication factor, and quiescing the workload can be restricted to a ring of servers instead of the entire database.

Abstract: Techniques are described for modeling variations in correlation to facilitate analytic operations. In one or more embodiments, at least one computing device receives first metric data that tracks a first metric for a first target resource and second metric data that tracks a second metric for a second target resource. In response to receiving the first metric data and the second metric data, the at least one computing device generates a time-series of correlation values that tracks correlation between the first metric and the second metric over time. Based at least in part on the time-series of correlation data, an expected correlation is determined and compared to an observed correlation. If the observed correlation falls outside of a threshold range or otherwise does not satisfy the expected correlation, then an alert and/or other output may be generated.

Abstract: Techniques are described for modeling variations in correlation to facilitate analytic operations. In one or more embodiments, at least one computing device receives first metric data that tracks a first metric for a first target resource and second metric data that tracks a second metric for a second target resource. In response to receiving the first metric data and the second metric data, the at least one computing device generates a time-series of correlation values that tracks correlation between the first metric and the second metric over time. Based at least in part on the time-series of correlation data, an expected correlation is determined and compared to an observed correlation. If the observed correlation falls outside of a threshold range or otherwise does not satisfy the expected correlation, then an alert and/or other output may be generated.

Abstract: Techniques are described for modeling variations in correlation to facilitate analytic operations. In one or more embodiments, at least one computing device receives first metric data that tracks a first metric for a first target resource and second metric data that tracks a second metric for a second target resource. In response to receiving the first metric data and the second metric data, the at least one computing device generates a time-series of correlation values that tracks correlation between the first metric and the second metric over time. Based at least in part on the time-series of correlation data, an expected correlation is determined and compared to an observed correlation. If the observed correlation falls outside of a threshold range or otherwise does not satisfy the expected correlation, then an alert and/or other output may be generated.

Abstract: Techniques are described for performing cluster analysis on a set of data points using tri-point arbitration. In one embodiment, a first cluster that includes a set of data points is generated within volatile and/or non-volatile storage of a computing device. A set of tri-point arbitration similarity values are computed where each similarity value in the set of similarity values corresponds to a respective data point pair and is computed based, at least in part, on a distance between the respective data point pair and a set of one or more arbiter data points. The first cluster is partitioned within volatile and/or non-volatile storage of the computing device into a set of two or more clusters. A determination is made, based at least in part on the set of similarity values in the tri-arbitration similarity matrix, whether to continue partitioning the set of data points.

Abstract: Techniques are described for modeling variations in correlation to facilitate analytic operations. In one or more embodiments, at least one computing device receives first metric data that tracks a first metric for a first target resource and second metric data that tracks a second metric for a second target resource. In response to receiving the first metric data and the second metric data, the at least one computing device generates a time-series of correlation values that tracks correlation between the first metric and the second metric over time. Based at least in part on the time-series of correlation data, an expected correlation is determined and compared to an observed correlation. If the observed correlation falls outside of a threshold range or otherwise does not satisfy the expected correlation, then an alert and/or other output may be generated.

Abstract: Techniques are described for performing cluster analysis on a set of data points using tri-point arbitration. In one embodiment, a first cluster that includes a set of data points is generated within volatile and/or non-volatile storage of a computing device. A set of tri-point arbitration similarity values are computed where each similarity value in the set of similarity values corresponds to a respective data point pair and is computed based, at least in part, on a distance between the respective data point pair and a set of one or more arbiter data points. The first cluster is partitioned within volatile and/or non-volatile storage of the computing device into a set of two or more clusters. A determination is made, based at least in part on the set of similarity values in the tri-arbitration similarity matrix, whether to continue partitioning the set of data points.