Log Engine Family
These engines were developed for scenarios when you need to quickly write many small tables (up to about 1 million rows) and read them later as a whole.
Engines of the family:
Store data on a disk.
Append data to the end of file when writing.
Support locks for concurrent data access.
INSERTqueries, the table is locked, and other queries for reading and writing data both wait for the table to unlock. If there are no data writing queries, any number of data reading queries can be performed concurrently.
Do not support mutations.
Do not support indexes.
This means that
SELECTqueries for ranges of data are not efficient.
Do not write data atomically.
You can get a table with corrupted data if something breaks the write operation, for example, abnormal server shutdown.
TinyLog engine is the simplest in the family and provides the poorest functionality and lowest efficiency. The
TinyLog engine does not support parallel data reading by several threads in a single query. It reads data slower than other engines in the family that support parallel reading from a single query and it uses almost as many file descriptors as the
Log engine because it stores each column in a separate file. Use it only in simple scenarios.
StripeLog engines support parallel data reading. When reading data, ClickHouse uses multiple threads. Each thread processes a separate data block. The
Log engine uses a separate file for each column of the table.
StripeLog stores all the data in one file. As a result, the
StripeLog engine uses fewer file descriptors, but the
Log engine provides higher efficiency when reading data.