OSEC

Neohapsis is currently accepting applications for employment. For more information, please visit our website www.neohapsis.com or email hr@neohapsis.com
 
Re: "Load Data Infile ... Repair with KeyCache" --WAY SLOW!

From: mos (mos99fastmail.fm)
Date: Thu Mar 03 2005 - 14:13:19 CST


At 12:39 PM 3/3/2005, Harrison Fisk wrote:
>Hi,
>
>On Mar 3, 2005, at 11:32 AM, mos wrote:
>
>>At 10:07 PM 3/2/2005, you wrote:
>>>Hello.
>>>
>>>You may use ALTER TABLE .. DISABLE KEYS to speed up the loading process
>>>on the MyISAM table.
>>
>>That may work provided I can get the keys rebuilt later using FileSort
>>and not KeyCache.
>>
>>You see the problem isn't in loading the data into the table which occurs
>>reasonably fast that because running "Load Data Infile" on an empty table
>>will disable the keys until all the data is loaded, so explicitly
>>disabling before hand them won't help. After the data is loaded, Load
>>Data Infile will then rebuild the keys and will choose to use "Repair
>>with keycache" whereas "Repair with filesort" would be 100x-1000x faster.
>>There doesn't seem to be any way to get it to use "Repair with filesort".
>>
>>So I could use disable keys as you had suggested and then rebuild them
>>manually with myisamchk "repair with sort" provided it doesn't reload all
>>the data into a temporary table (which Alter Table usually does). If it
>>does create a temporary table it will physically reloads the 500 million
>>rows a second time and I will need another 100g of free disk space.
>>
>>So is there a solution to this paradox?
>
>It chooses which method to use based on your myisam_sort_buffer_size,
>myisam_max_sort_file_size and myisam_max_extra_sort_file_size. Have you
>increased the size of these? Keep in mind these are SESSION variables, so
>they can be set on the connection right before you LOAD DATA INFILE.
>
>You always have to tell MySQL how much extra memory/disk it can use, it
>can't assume that you want it to use it. That is why it will use Repair
>by keycache, since it doesn't use any extra resources. If you increase
>the amount of diskspace or memory, then it should use a Repair by sort instead.

That's what I thought. I have myisam_max_sort_file_size set to 15.7GB and
myisam_max_extra_sort_file_size set to 2GB. I confirmed this with Show
Variables. I thought this would be enough to sort the indexes. I have
confirmed there is enough free space on the hard drive to handle this both
in the database directory and the mysql temp directory.

>Also keep in mind that Repair by sort doesn't work for UNIQUE or PRIMARY KEYs.

So the solution is to keep using Load Data Infile with even larger max_sort
variables and more disk space? There are only 4 indexes on the table and
the largest index is around 50 bytes. This is running on an AMD3500+ with
1gb RAM.

My problem is it takes about 6 hours to load in the rows using Load Data
Infile to the empty table (which is fine) then by doing a "Show
ProcessList" I'll know if it is using FileSort or KeyCache. That is a lot
of time to waste if it ends up using KeyCache because then I know it could
be a few days to index the table.

So how do people force Load Data Infile to use FileSort and not KeyCache?
In other words how do they know ahead of time what to set
myisam_max_extra_sort_file_size set and myisam_max_sort_file_size to
without guessing? To me it seems a lot like hit and miss guesswork. I would
rather see a parameter "BySort" or "ByKeyCache" added to Load File command
that forces it to use either FileSort or KeyCache rather than having the
developer cross his fingers and hope the system variables are set correctly.

Mike

>Regards,
>
>Harrison
>
>--
>Harrison C. Fisk, Trainer and Consultant
>MySQL AB, www.mysql.com
>
>Get a jumpstart on MySQL Cluster --
>http://www.mysql.com/consulting/packaged/cluster.html

--
MySQL General Mailing List
For list archives: http://lists.mysql.com/mysql