Consistent query execution in hybrid DBMS

A computer-implemented method, comprising, in a data processing system including a first database engine maintaining a first database, a second database engine maintaining a second database, a table existing as a first instance in the first database and as a second instance in the second database, and a replication module:

• executing, by the first engine, a plurality of write transactions, thereby respectively changing the first instance;
• asynchronously pooling, by the replication module, the changes of the first instance introduced by the write transactions;
• in response to receiving a query against the table, determining whether to execute the query against the first instance or against the second instance;
• and
• responsive to determining to execute the query against the second instance:
• identifying, by the first engine, a first time, the first time being the commit time of the most recent of the write transactions before the receipt of the query,
• defining, by the first engine, a current replication batch that selectively comprises the pooled changes of the ones of the transactions having committed at or before the first time,
• asynchronously replicating, by the replication module, the current replication batch, thereby propagating the changes of the first instance specified in the current replication batch to the second instance,
• storing, by the second engine, each change that is specified in the current replication batch in the second instance,
• assigning, by the first engine, a batch-ID of the current replication batch to the query,
• sending, by the first engine, the query and the assigned batch-ID to the second engine,
• determining, by the second engine, if the replication of the current replication batch was completed,
• responsive to the current replication batch having been completed, executing, by the second engine, the query on the second instance, thereby using the batch-ID assigned to the query as a filter to exclude any changes having assigned a batch-ID of a replication batch having committed later than the current replication batch, and
• returning, by the second engine, results of the query execution on the second instance.