OK, I see that we are all ultimately saying the same thing: Know your setup and customize for it.

I agree with you 100%. InnoDB will go down less, especially if you set it up correctly, and for system critical setups you should definitely have replicas, and backups, and load balancers, etc. When the data is important enough, the money will be there. And yes, sometimes you have to slap the in-house developers.

I also agree with Steve Meyers on the other emails in this thread. InnoDb has a lot of neat stuff in it, and is a lot better than it used to be (there are reasons Oracle bought it, some even have to do with the code). When it makes sense, I definitely use it. Customer tables, yes; general business data, yes; high volume audit log, no. Customer's pet blog, depends on the software.

In the past, I have done a lot of work with ISPs and shared hosting, etc, and most of their setups are single server, doing mysql for their pet blog. A lot of this opensource stuff used InnoDB simply because it looks cool or the book they bought says to use it, and it causes a lot of issues due to not being setup very well, which causes issues for other users as MySQl doesn't expect to have thousands of databases all doing InnoDB poorly, and when the inevitable crash comes it doesn't bother to tell you which user, which database, or which table, just a memory pagedump, which can be very annoying playing "to whom does the wordpress data belong to".

-Steve

On 2/8/12 4:17 PM, Jason Van Patten wrote:
On 2/8/2012 4:02 PM, Steve Alligood wrote:
Why would you run performance critical really large data sets on
spindles? If it is performance critical, put your db on SSDs.

This entire discussion has been from a DBA standpoint.  From a
SysAdmin standpoint, splitting out your tables into files makes life
MUCH better when things go south.  In fact, MyISAM tables are MUCH
better when things go south, as they are both easier and faster to
repair and get back online.

In fact, if you have a lot of databases (or even more than one) using
InnoDB tables and one of them has issues bad enough to where the
autorepair doesn't work, you have to dump them all and re-import them
all to be sure you got the bad one, as the page dump MySQL so nicely
gives you does squat for telling you *which* database or even table is
corrupt and is causing all of MySQL do stay offline.  Or you can try
hide and seek and randomly try them until MySQL comes back online.
That one is fun, too.

I have also found that MyISAM is considerably faster for relatively
simple datasets, especially if you are mostly doing inserts and not a
lot of selects.  Constraints, checkpoints, row locking, etc, all add
overhead.  In fact, if write speed is important, don't even do more
indexes that you specifically need.

Basically, if you don't specifically need what InnoDB brings to the
party, stick with MyISAM every time and you will have far less
problems in a repair situation, simpler management, and faster writes.

-Steve
And then your socket volume spikes and your down. Once you get to the
point that you are having crashes you should at least look at switching.
That and/or slap your developers for not optimizing queries. as for
fixing things when they go south yes myisam is easier, but your risk of
going down in the first place is higher. Furthermore a competent backup
system and recovery code can make innodb down time negligible.

Jason

On 2/8/12 3:44 PM, Jason Van Patten wrote:
On 2/8/2012 3:13 PM, Paul Seamons wrote:
I'd second that, but I'd add a "be sure to add innodb_file_per_table=1
to the my.cnf
Be very careful with that entry. It can wreck your seek times once you
get really large data sets and can make load balancing moot.
Jason
Paul

On 02/08/2012 12:24 PM, Jason Van Patten wrote:
On 2/8/2012 11:50 AM, Merrill Oveson wrote:
MyISAM


That might be the reason mysql locked in the first place. You get one
rouge query that takes too long and every other query hitting  that
table and all joins has to wait for it. InnoDB is a little slower, but
it ill only lock rows affected by the query instead of the whole
table/tables. Unless you are index text columns i would recommend
switching to Innodb or setting up a watchdog on your process list.
Most
times the watchdog is a good move just so you have time to get innodb
running (select * statements will all need to get paired down to the
minimum)




Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

/*
PLUG: http://plug.org, #utah on irc.freenode.net
Unsubscribe: http://plug.org/mailman/options/plug
Don't fear the penguin.
*/

Reply via email to