Comments on: Video and Slides: How InnoDB works

By: Sheeri Cabral

Sheeri Cabral — Sat, 05 Jun 2010 14:57:38 +0000

Chris — I’m not sure what you mean….As the first paragraph implies, I went through SHOW ENGINE INNODB STATUS in this presentation, and in doing so ended up teaching a lot about how InnoDB works.

Is there a “better” storage engine for MySQL? Sure, if you’re doing lots of INSERTs, few or no UPDATEs and few or no DELETEs, and some reads, it’s best to use MyISAM with concurrent_insert=2. Similarly, if you want to do blob streaming PBXT is better than InnoDB.

I could go on, but “better” is relative for what you want to do with a particular table.

By: Chris Huntley

Chris Huntley — Sat, 05 Jun 2010 07:52:12 +0000

I can’t tell from your comments if you’re just reviewing how innodb works impartially or if you’re a fan. That being said, do you think there’s a better storage engine for mysql?

By: Kayra Otaner

Kayra Otaner — Fri, 20 Nov 2009 19:13:26 +0000

Sheeri,
Thank you, I believe we agree on that.
I just wanted to confirm there is no direct relationship between dirty pages & log flushes as that sentence I quoted sounded as if there is one.

Thank you again.

By: Sheeri

Sheeri — Fri, 20 Nov 2009 17:27:18 +0000

Kayra —

I’m not sure what the question is…you commented on me saying that dirty pages are not flushed to disk, even when logs are…you quoted me saying:

“modified db pages are pages not flushed to disk yet (>0 even when flush_logs_at_trx_commit=0).”

so I think we’re agreeing here….flushing logs has nothing to do with flushing dirty pages…..

Unless I missed something in your original question?

By: Kayra Otaner

Kayra Otaner — Thu, 19 Nov 2009 21:35:40 +0000

Sorry I should have been more clear with my question above.
My question was more like to understand relationship you came up with between dirty page count and innodb_flush_log_at_trx_commit variable.
AFAIK innodb_flush_log_at_trx_commit=1 guarantees 1 commit – 1 flush. Any other variation will (almost) provide flush every second. As you said, flushes are for logs only, so when you flush, you actually flush into log file, not to InnoDB table space. I believe flushing into log, whether its with every commit or every second should not reduce dirty pages in memory as they’re still not identical to the pages on InnoDB table space.

By: Sheeri

Sheeri — Thu, 19 Nov 2009 20:24:05 +0000

Kayra —

Yes, I can say that.

innodb_flush_log_at_trx_commit flushes logs to disk, not dirty pages. If the database crashes, InnoDB’s crash recovery will redo the writes on the dirty pages using what’s in the logs.

The flushes are for logs only, not the actual data pages. That tripped me up when I first thought about it too.

By: Kayra Otaner

Kayra Otaner — Thu, 19 Nov 2009 18:05:30 +0000

Can you really say :
“modified db pages are pages not flushed to disk yet (>0 even when flush_logs_at_trx_commit=0).”

Only innodb_flush_log_at_trx_commit=1 (default/Acid compliant) mode guarantees flushes with every commit. So anything other than default value 1, will flush logs to disk. But still this isn’t directly related to dirty pages?

By: Shlomi Noach

Shlomi Noach — Fri, 16 Oct 2009 19:31:31 +0000

Sheeri,

Haven’t watched it yet. Will find the time after the weekend :)

By: Sheeri Cabral

Sheeri Cabral — Fri, 16 Oct 2009 19:30:02 +0000

Shlomi — did you like the video too?

By: Sheeri Cabral

Sheeri Cabral — Fri, 16 Oct 2009 19:29:43 +0000

Shlomi — yep! Usually people just want the notes, so I use the trick of opening the first page to the first page of the notes. (mostly it bothered me when I printed, b/c when I print I want *only* the notes. So for a 10-slide deck, I print pages 11-20).

But I appreciate your trick!