Distributed transaction database log with immediate reads and batched writes

A distributed transactional database comprising:

• a plurality of data replica servers in a database cluster,
• a data set partitioned among the plurality of data replica servers in the database cluster;
• a global log, replicated and partitioned among a plurality of log replica servers in the database cluster;
• a query coordinator configured to forward transactions to one of the plurality of log replica servers;
• a consistency model providing strict serializability for transactions, using a consensus algorithm to achieve consensus with other log replica servers to record transactions in a pre-determined order;
• each data replica server configured to process, in the pre-determined order, an ordered series of batches of transactions, on only the data in its partition;
• and
• each data replica server configured to process read-only transactions immediately, separately from transaction batching and logging, as time-stamped, lockless read operations without coordination with other servers;
• wherein write-only and read/write transactions are restricted to a single logical database and performed at strictly serializable consistency, while read-only transactions may be performed at serializable or read-committed consistency.

A distributed transactional database comprising:

• a plurality of data replica servers in a database cluster, having a data set partitioned among the plurality of data replica servers in the database cluster;
• a global log, replicated and partioned among a plurality of log replica servers in the database cluster;
• a query coordinator configured to forward transactions to one of the plurality of log replica servers;
• a consistency model providing strict serializability for transactions, using a consensus algorithm to achieve consensus with other log replica servers to record transactions in a pre-determined order;
• each data replica server configured to process, in the pre-determined order, an ordered series of batches of transactions, on only the data in its partition;
• each data replica server configured to process read only transactions immediately, separately from transaction batching and logging, as time-stamped, lockless read operations without coordination with other servers;
• a plurality of database drivers that maintain a high watermark of global log position of a last request, wherein the plurality of database drivers are configured to guarantee a monotonically advancing view of a global transaction order, such that all transactions, including immediate read-only transactions, are performed with serializable consistency;
• and
• a distributed storage layer with a transaction resolution algorithm, wherein the transaction resolution algorithm uses Raft logs to replicate;
• wherein write-only and read/write transactions are restricted to a single logical database, and performed at strictly serializable consistency, while read-only transactions may be performed at serializable or read-committed consistency;
• wherein the distributed transactional database uses optimistic locking for the database transactions, in order to support arbitrary read dependencies within transactions;
• wherein the distributed transactional database is configured to provide a row-level access control system and row level security, identity and query isolation and management;
• wherein the query coordinator is configured to pre-process transactions with read dependencies by pushing read predicates to data replicas that own the underlying data, and accumulate potential transaction read and write effects in a write buffer;
• and
• wherein the query coordinator is configured to use a NoSQL query language.

A distributed transactional database comprising:

• a plurality of data replica servers in a database cluster, with a data set partitioned among the plurality of data replica servers;
• a global log, replicated and partitioned among a plurality of log replica servers in the database cluster;
• a query coordinator configured to forward transactions to one of the plurality of log replica servers;
• a consistency model providing strict serializability for transactions, using a consensus algorithm to achieve consensus with other log replica servers to record transactions in a pre-determined order;
• each data replica server configured to process, in the pre-determined order, an ordered series of batches of transactions, on only the data in its partition;
• wherein write-only and read/write transactions are restricted to a single logical database and performed at strictly serializable consistency, while read-only transactions may be performed at a serializable or read-committed consistency;
• each data replica server configured to process read only transactions immediately, separately from transaction batching and logging, as time-stamped, lockless read operations without coordination with other servers;
• a recursive multi-tenancy engine configured to assign each tenant a separate instance of the distributed transactional database;
• a query isolation engine including a recursive process executor for recursive scheduling;
• a data storage engine including data types that support a variety of query models, including any of search, graph, temporal, geospatial, key/value, document, and analytics queries, configured such that any such query model may be used to access any data.