what is the significance of checkpoint in a transaction ?

Question

I am confused regrading the significance of checkpoint i.e. is it that whenever we place checkpoint does that mean that all the data items before it have been written onto the data base , if not then why do we place it below the transactions which have simply written the data item but not yet committed ?

Answer 1

checkpoints are inserted in order to make the recovery process more simpler. instead of writing everything to the stable storage at the end, we issue checkpoint in different occassion. Once a check point is encountered all the dirty pages of all transaction ( committed and non committed ) is flushed to the stable storage. Also the transaction log is flushed to storage

Now if there is any need of any recovery, if we need to do a redo , then we need to do it only after the checkpoing

if it is an undo also we need to do it for before the check  point (for transaction which were not committed at the time of checkpoint)
those transactions which were committed before checkpoint, there is no need to do anything.

 

please note in reality some transaction/ schecules may take a very long time to complete. so it is always better to use checkpoints

Comments

Consider a simple checkpointing protocol and the following set of operations in the log. (start, T4); (write, T4, y, 2, 3); (start, T1); (commit, T4); (write, T1, z, 5, 7); (checkpoint); (start, T2); (write, T2, x, 1, 9); (commit, T2); (start, T3); (write, T3, z, 7, 2); If a crash happens now and the system tries to recover using both undo and redo operations, what are the contents of the undo list and the redo list

 answ is :Undo: T3, T1; Redo: T2

 

Now I am not getting that when T2 is committed then why are we actually performing Redo operation on it ,Also when T1 has not written anything onto the disk neither it has committed, then why did we use checkpoint after it ?

---

T2 committed means the result should be durable. But it does not mean we have stored it into a stable storage medium. The changes in the stable storage is done at checkpoint. so once a crash occurs, we should not loose the effect of committed transactions. So whatever operations done by committed transactions should be redone.

Check points can be inserted anywhere. so the checkpoint after T1 does not matter. But since there is checkpoint we are storing the new z written  by T1 to storage. But finally before end of transaction crash happens. so we need to undo the operations performed by T1.  But T4 was committed before last checkpoint, so all changes are successfully recorded in the storage, so no need of any undo/ redo for T4

hope the explanation is clear

---

Just last confusion sir , that we put checkpoints so that the recovery procedure does not search in the entire log file which transactions have been committed and which not , so even if we placed checkpoint after T1 , it is actually of no use , God knows but still I am stucked with this point ,on wikipedia it explained that " For performance benefits all the database servers doing changes in memory , once Changes done in memory of database are checked as per last checkpoint and then all the dirty pages of databases are written to the disk"

Now this is even more confusing ,please help me sir with this concept , I am stucked up badly .