RocksDB stress test is done by creating a database with a large number of rows (1M) in a number of column families(10). We also create an array of large vectors to store the values of these rows in memory. The rows are partitioned between multiple threads(32). Each thread is assigned to operate on a fixed contiguous portion of the rows. Each thread does a number of operations on its piece of the database and then verifies it with the in memory copy of the database which is also being updated as changes are made to the database. As configured, there can be different number of operations, threads, rows, column families, operations per batch, iterations, and operations between database reopen. Different database options can also be modified which results in various configurations and features being tested. eg. checksums, time to live, universal/level compaction, compression types to use at different levels, maximum number of write buffers, minimum number of buffers to flush.
Documentation license here.
本文档中文部分由 johnzeng 翻译完成，在此表示感谢。 欢迎大家贡献力量，加QQ群联系。