Postgres Pro
Downloads
Home   >   mailing lists

Thread: [HACKERS] GSoC 2017 Proposal for predicate locking in hash index

[HACKERS] GSoC 2017 Proposal for predicate locking in hash index

From
Shubham Barai
Date:
Hi, 

Now that hash index support write-ahead logging, it will be great if we add support for predicate locking to it. 
Implementation of predicate locking in hash index seems very simple.
I have already made changes in the code. I am currently working on testing.

Here is my approach 

1) PredicateLockPage()

->_hash_first()

During a scan operation, acquire a predicate lock on the primary page of a bucket.

2) CheckForSerializableConflictIn()

->_hash_doinsert()

During an insert operation, check if there is any predicate lock on the primary page of a bucket.


3) PredicateLockPageSplit()

In case of a bucket split, copy predicate lock from the primary page of an old bucket to the primary page of a new bucket.

Any suggestions or corrections will be appreciated.

Regards,
Shubham



Sent with Mailtrack

Re: [HACKERS] GSoC 2017 Proposal for predicate locking in hash index

From
Alvaro Herrera
Date:
Shubham Barai wrote:
> Hi,
> 
> Now that hash index support write-ahead logging, it will be great if we add
> support for predicate locking to it.
> Implementation of predicate locking in hash index seems very simple.
> I have already made changes in the code. I am currently working on testing.

So if I understand correctly, this would only cause a false positive if
two transactions have a rw/ww conflict in different tuples in the same
bucket.  Is that what we expect?

-- 
Álvaro Herrera                https://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services

Re: [HACKERS] GSoC 2017 Proposal for predicate locking in hash index

From
Shubham Barai
Date:
Hi,

On 22 June 2017 at 21:12, Alvaro Herrera <alvherre@2ndquadrant.com> wrote:
Shubham Barai wrote:
> Hi,
>
> Now that hash index support write-ahead logging, it will be great if we add
> support for predicate locking to it.
> Implementation of predicate locking in hash index seems very simple.
> I have already made changes in the code. I am currently working on testing.

So if I understand correctly, this would only cause a false positive if
two transactions have a rw/ww conflict in different tuples in the same
bucket.  Is that what we expect?

Yes, I think so. Is there any way to further reduce false positives in the same bucket?

Regards,
Shubham


Sent with Mailtrack