Sequential disk access during VACUUM for GiST

From PostgreSQL wiki
Jump to navigationJump to search

Contact

name - Constantine Kuznetsov

via email - chapaev28@ya.ru


Introduction

I work on this idea later. http://www.postgresql.org/message-id/CAPpHfdtxftKcM43EBbj=KteW-3LTTRTkm_ABEt0pMvkfVwQwSA@mail.gmail.com

But then there were problems with "jump back" to rescan parent page.


Now i propose implementation with 2 steps:

1. physical order scan. Creation(child,parent) of hash that needed to fast search parent page.

2. scan of page that be needed rescan after 1 step.


Expected project schedule

1. Implement new gistbulkdelete function. 2 weeks.

2. Test this implementation and writing regression test. 2 weeks.

3. Getting the first result and discussion with mentor. 1 week.

4. Add all memory limitation(maintenance_work_mem) to implementation. 1 week.

5. If there are problems then solve their. 2 weeks.

6. Cleaning of code, writing of coments and preparing to patch. 1 week.