Concurrent Inserts
The MyISAM storage engine supports concurrent inserts. This feature allows SELECT statements to be executed during INSERT operations, reducing contention.
Whether concurrent inserts can be used or not depends on the value of the concurrent_insert server system variable:
NEVER(0) disables concurrent inserts.AUTO(1) allows concurrent inserts only when the target table has no free blocks (no data in the middle of the table has been deleted after the last OPTIMIZE TABLE). This is the default.ALWAYS(2) always enables concurrent inserts, in which case new rows are added at the end of a table if the table is being used by another thread.
If the binary log is used, CREATE TABLE ... SELECT and INSERT ... SELECT statements cannot use concurrent inserts. These statements acquire a read lock on the table, so concurrent inserts will need to wait. This way the log can be safely used to restore data.
Concurrent inserts are not used by replicas with the row based replication (see binary log formats).
If an INSERT statement contain the HIGH_PRIORITY clause, concurrent inserts cannot be used. INSERT ... DELAYED is usually unneeded if concurrent inserts are enabled.
LOAD DATA INFILE uses concurrent inserts if the CONCURRENT keyword is specified and concurrent_insert is not NEVER. This makes the statement slower (even if no other sessions access the table) but reduces contention.
LOCK TABLES allows non-conflicting concurrent inserts if a READ LOCAL lock is used. Concurrent inserts are not allowed if the LOCAL keyword is omitted.
Notes
The decision to enable concurrent insert for a table is done when the table is opened. If you change the value of concurrent_insert it will only affect new opened tables. If you want it to work for also for tables in use or cached, you should do FLUSH TABLES after setting the variable.
