Thread: Checkpoints - what happens actually?

Checkpoints - what happens actually?

From
KÖPFERL Robert
Date:
Hi,

i went across a checkpoint problematic (90% load, near real time app). I've
read in pg-documentation what exists about CHECKPOINT and the configuration
parameters. But there're still many open questions as:

What actually happens if a checkpoint occours?
OK, all dirty pages of data files get written to disk...but

Are read requests (simple selects) still possible and quickly answered?

What happens to wite-operations? Do they get queued or can they be written
to xlog while the fsync is running?
Is it maybe really the case but wave of pending statements arises from the
reduced i/o capacity which is left due to the fsync?

How about increasing the interval more and more? Will this produce more and
more dirty pages or is it just that a possible restart will take a longer
time since the xlog is longer?

Questions over questions. Can you please give me a clearing little insight.

Re: Checkpoints - what happens actually?

From
KÖPFERL Robert
Date:
Empric tests have shown that read rrequests seem to still get proceeded
whilst write requests get queued until all dirty pages have been written.

That's however just testing and looking. What actually happens is a secret.


|-----Original Message-----
|From: KÖPFERL Robert
|Sent: Montag, 27. Juni 2005 15:22
|To: pgsql-admin@postgresql.org
|Subject: [ADMIN] Checkpoints - what happens actually?
|
|
|Hi,
|
|i went across a checkpoint problematic (90% load, near real
|time app). I've
|read in pg-documentation what exists about CHECKPOINT and the
|configuration
|parameters. But there're still many open questions as:
|
|What actually happens if a checkpoint occours?
|OK, all dirty pages of data files get written to disk...but
|
|Are read requests (simple selects) still possible and quickly answered?
|
|What happens to wite-operations? Do they get queued or can
|they be written
|to xlog while the fsync is running?
|Is it maybe really the case but wave of pending statements
|arises from the
|reduced i/o capacity which is left due to the fsync?
|
|How about increasing the interval more and more? Will this
|produce more and
|more dirty pages or is it just that a possible restart will
|take a longer
|time since the xlog is longer?
|
|Questions over questions. Can you please give me a clearing
|little insight.
|
|---------------------------(end of
|broadcast)---------------------------
|TIP 5: Have you checked our extensive FAQ?
|
|               http://www.postgresql.org/docs/faq
|