How to сomprehend the reason of the deadlock - namely, how to find out which transactions have captured which locks?
I have the engine.log file with the following deadlock:
------------------------
LATEST DETECTED DEADLOCK
------------------------
170327 11:09:53
*** (1) TRANSACTION:
TRANSACTION 4 2719072253, ACTIVE 5 sec, OS thread id 26215 starting index read
...
INSERT INTO... (the first transaction)
*** (1) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 0 page no 36025889 n bits 96 index `PRIMARY` of table `mydb`.`mytable` trx id 4 2719072253 lock mode S locks rec but not gap waiting
...
*** (2) TRANSACTION:
TRANSACTION 4 2719072205, ACTIVE 35 sec, OS thread id 25564 starting index read, thread declared inside InnoDB 485
UPDATE ... (the second transaction)
*** (2) HOLDS THE LOCK(S):
RECORD LOCKS space id 0 page no 36025889 n bits 96 index `PRIMARY` of table `mydb`.`mytable` trx id 4 2719072205 lock_mode X locks rec but not gap
Record lock, heap no 27 PHYSICAL RECORD: n_fields 72; compact format; info bits 0
...
*** (2) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS space id 0 page no 42767646 n bits 120 index `PRIMARY` of table `mydb`.`mytable` trx id 4 2719072205 lock_mode X locks rec but not gap waiting
...
*** WE ROLL BACK TRANSACTION (1)
And my vision of what is described in the logs is the following:
1. Transaction №2 initially has one lock (and the type of the lock isn't clear from logs):
*** (2) HOLDS THE LOCK(S)
RECORD LOCKS space id 0 page no 36025889 n bits 96 index PRIMARY of table mydb.mytable trx id 4 2719072205 lock_mode X locks rec but not gap
Record lock, heap no 27 PHYSICAL RECORD: n_fields 72; compact format; info bits 0
2. Transaction №1 is trying to get the lock of S type:
*** (1) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS ... trx id 4 2719072253 lock mode S locks rec but not gap waiting
and after unsuccessful attempt it begins to wait for releasing of transaction №2 lock;
3. Then transaction №2 is trying to get the lock of X type:
*** (2) WAITING FOR THIS LOCK TO BE GRANTED:
RECORD LOCKS ... trx id 4 2719072205 lock_mode X locks rec but not gap waiting
and after unsuccessful attempt it begins to wait until transaction №1 gets the S lock and release it.
Do I correctly understand logs, or is my interpretation wrong?
Your interpretation is almost correct. A few thoughts to add:
In innodb there are two basic types of locks: shared (S) and exclusive (X). If a transaction holds an X lock on a record, then neither S or X locks can be granted on the same records until the transaction completes. If a transaction holds an S lock on a record, then another S lock can be granted immediately, but an X lock request must wait until the 1st transaction completes.
The 2nd transaction being an update holds an exclusive (X) lock on the index records that are being updated. Therefore the 1st transaction cannot get hold of an S lock on the same records. However, the 2nd transaction is waiting for X lock to be granted on a different set of records (page no is different from the other locks').
The excerpt does not tell us what transaction locked the records that the 2nd transaction is waiting for. We can only presume that it was the 1st transaction - otherwise it would not be a deadlock.