Обсуждение: Checkpoint versus Background Writer
They seem to do similar things: clear dirty buffers from shared_buffers to disk.
So why have 2 processes with seperate semantics (seperate set of config partms) ?
Assuming PG is multi-threaded, can't we simply have multiple threads of the checkpoint achieve the same result as (checkpoint + bg)
Shiv
Checkpoint is different to background writer though works together
BGW continuously write buffer to disk but checkpoint is interval default to 5min at which the process call BGW to flush the entire buffer to disk
BGW continuously write buffer to disk but checkpoint is interval default to 5min at which the process call BGW to flush the entire buffer to disk
Thanks,
Sunday Olutayo
Sunday Olutayo
From: "Shiv Sharma" <shiv.sharma.1835@gmail.com>
To: pgsql-novice@postgresql.org
Sent: Friday, December 27, 2013 8:46:42 PM
Subject: [NOVICE] Checkpoint versus Background Writer
To: pgsql-novice@postgresql.org
Sent: Friday, December 27, 2013 8:46:42 PM
Subject: [NOVICE] Checkpoint versus Background Writer
They seem to do similar things: clear dirty buffers from shared_buffers to disk.
So why have 2 processes with seperate semantics (seperate set of config partms) ?
Assuming PG is multi-threaded, can't we simply have multiple threads of the checkpoint achieve the same result as (checkpoint + bg)
Shiv
>Checkpoint is different to background writer though works together
>BGW continuously write buffer to disk but checkpoint is interval default to 5min at which the process call >BGW to flush the entire buffer to disk
>BGW continuously write buffer to disk but checkpoint is interval default to 5min at which the process call >BGW to flush the entire buffer to disk
But with 8.3 and checkpoint_completion_target, is not the checkpoint spread out too? (or can be spread out).
On Fri, Dec 27, 2013 at 10:58 AM, SUNDAY A. OLUTAYO <olutayo@sadeeb.com> wrote:
Checkpoint is different to background writer though works together
BGW continuously write buffer to disk but checkpoint is interval default to 5min at which the process call BGW to flush the entire buffer to diskThanks,
Sunday OlutayoFrom: "Shiv Sharma" <shiv.sharma.1835@gmail.com>
To: pgsql-novice@postgresql.org
Sent: Friday, December 27, 2013 8:46:42 PM
Subject: [NOVICE] Checkpoint versus Background WriterThey seem to do similar things: clear dirty buffers from shared_buffers to disk.So why have 2 processes with seperate semantics (seperate set of config partms) ?Assuming PG is multi-threaded, can't we simply have multiple threads of the checkpoint achieve the same result as (checkpoint + bg)Shiv
I do not understand your statement
Thanks,
Sunday Olutayo
Sunday Olutayo
From: "Shiv Sharma" <shiv.sharma.1835@gmail.com>
To: "SUNDAY A. OLUTAYO" <olutayo@sadeeb.com>
Cc: pgsql-novice@postgresql.org
Sent: Friday, December 27, 2013 9:41:58 PM
Subject: Re: [NOVICE] Checkpoint versus Background Writer
To: "SUNDAY A. OLUTAYO" <olutayo@sadeeb.com>
Cc: pgsql-novice@postgresql.org
Sent: Friday, December 27, 2013 9:41:58 PM
Subject: Re: [NOVICE] Checkpoint versus Background Writer
>Checkpoint is different to background writer though works together
>BGW continuously write buffer to disk but checkpoint is interval default to 5min at which the process call >BGW to flush the entire buffer to disk
>BGW continuously write buffer to disk but checkpoint is interval default to 5min at which the process call >BGW to flush the entire buffer to disk
But with 8.3 and checkpoint_completion_target, is not the checkpoint spread out too? (or can be spread out).
On Fri, Dec 27, 2013 at 10:58 AM, SUNDAY A. OLUTAYO <olutayo@sadeeb.com> wrote:
Checkpoint is different to background writer though works together
BGW continuously write buffer to disk but checkpoint is interval default to 5min at which the process call BGW to flush the entire buffer to diskThanks,
Sunday OlutayoFrom: "Shiv Sharma" <shiv.sharma.1835@gmail.com>
To: pgsql-novice@postgresql.org
Sent: Friday, December 27, 2013 8:46:42 PM
Subject: [NOVICE] Checkpoint versus Background WriterThey seem to do similar things: clear dirty buffers from shared_buffers to disk.So why have 2 processes with seperate semantics (seperate set of config partms) ?Assuming PG is multi-threaded, can't we simply have multiple threads of the checkpoint achieve the same result as (checkpoint + bg)Shiv
On Fri, Dec 27, 2013 at 11:46 AM, Shiv Sharma <shiv.sharma.1835@gmail.com> wrote: > They seem to do similar things: clear dirty buffers from shared_buffers to > disk. > > So why have 2 processes with seperate semantics (seperate set of config > partms) ? AFAIU, they serve for completely different purposes. Background writer is for performance. Each round, based on shared memory usage statistics and LRU data, it estimates how many new pages user backend processes will require before the next round, and flushes necessary amount of dirty pages on disk. So user backend processes will have enough non-dirty pages to operate with, and wont need to spend time on doing flushing themselves, that allows to return control to users faster. Checkpoints is a part of reliability mechanism. It applies changes accumulated in WAL files, that have not been applied earlier by background writer or user backends, to data files, keeping the database consistent before performing rotation. Correct me if I missed or misunderstood something. -- Kind regards, Sergey Konoplev PostgreSQL Consultant and DBA http://www.linkedin.com/in/grayhemp +1 (415) 867-9984, +7 (901) 903-0499, +7 (988) 888-1979 gray.ru@gmail.com
Sergey Konoplev <gray.ru@gmail.com> writes:
> On Fri, Dec 27, 2013 at 11:46 AM, Shiv Sharma
> <shiv.sharma.1835@gmail.com> wrote:
>> They seem to do similar things: clear dirty buffers from shared_buffers to
>> disk.
>>
>> So why have 2 processes with seperate semantics (seperate set of config
>> partms) ?
> AFAIU, they serve for completely different purposes.
We used to have a single process trying to serve both purposes, but
it didn't work tremendously well, because the criteria for when to
flush buffers are so different.
regards, tom lane