Concurrency control with locks
----------------------------------

Two phase locking/Strict two phase locking

Deadlocks:

   -> Periodically check for deadlocks: wait-for graph
   -> Avoid deadlocks:

---------------------

What are we locking?
---------------------

Relation

Page

Tuple


Trade off: high concurrency - high overhead

--------------------------

Lock Granularity
-----------------

IS
IX
S
X

PHANTOM UPDATE
-----------------

Read uncommitted  -> possible to read uncommitted data
Read committed    -> possible to read the same data twice and get different results
Repeatable read   -> possible to have phantom updates
Serializable      -> not  possible to have phantom updates


T1:
SELECT count(*) FROM episodes WHERE year = 2020;

Locked all tuples using S (or IS)

T2:
INSERT INTO episodes VALUES(....) ;

<< possible since no lock on a new tuple!>>


To implement serializability: lock a condition!