in reply to A database table size issue

sophate:

You've already gotten a good answer, so I'll just digress a bit.

I don't often see a task to process *every* row in a table where using SQL isn't better choice. It's pretty expensive to serialize each row and ship it to the other machine which does some processing, and then repeats the process to get the data back into the database. Since you haven't mentioned your level of experience with databases, I thought I'd mention a couple things, just in case.

  1. Can your task be done with SQL? It's a pretty simple language, and since the database server is optimized to use it, it's frequently faster than anything you'll be able to do locally (especially as you already have the overhead of shipping the data across a pipe).
  2. Even if your modifications are too complicated for a simple update, it can be better to make a temporary table and then do a set of transformations in SQL before updating the original table.
  3. If you find that you really need to get the data to the local desktop, be sure to get only the data you need to work on (via the WHERE clause, appropriate joins, etc.) so you can minimize the impact to your application, and all other users of the database.
  4. If you need to process all the rows and it's going to be a frequent operation, you might consider using the bulk-loader capabilities of your database to export the table into a flat file, which you can process easily with perl, and then bulk load the results back into the database. The bulk-loading tools for the databases I use (Sybase, MSSQL server, Oracle) are pretty efficient for whacking large amounts of data (and 30GB certainly qualifies).

If you want, feel free to share the types of transformations you're wanting and I'll try to help with the SQL.

Update: Typo fix (s/to/too/).

...roboticus

When your only tool is a hammer, all problems look like your thumb.

Replies are listed 'Best First'.
Re^2: A database table size issue
by nikosv (Deacon) on Apr 27, 2012 at 14:54 UTC
    ++ for the good answer but just to add to it, use set based logic rather than procedural row by row processing;Use the server resources rather that the client's. Furthermore even opening just firehose cursors for fetching that amount of rows will place loads of stress over the connection line if not optimised for batch fetching
[reply]
      Use the server resources rather that the client's.
      Really? Why? Typically, the number of servers is far less than the number of clients. Using server resources that benefit one client hurt all the clients.

      I'm not saying that in this case the task should be done on the client (we've no idea about the task, nor about the setup), but I disagree with the blanket "use server resources rather than the clients". In way too many cases that isn't scalable to do without thinking.

[reply]

        I set to write about why you should be using the dbms's power or resources instead of the client's,mainly because of query optimization,execution plan caching,parallel query execution,set based operations instead of procedural row by row,etc, but then I've noticed the OP's remark Re^2: A database table size issue,which acts as straightforward reply to your question :

        Thanks for the suggestion. The script was really slow when I processed the table row by row. I took your advice and rewrote the SQL to let the DB to handle part of the processing. The script now runs much faster than before

        So I have no futher comments
[reply]
Re^2: A database table size issue
by sophate (Beadle) on Apr 29, 2012 at 00:10 UTC

    Thanks for the suggestion. The script was really slow when I processed the table row by row. I took your advice and rewrote the SQL to let the DB to handle part of the processing. The script now runs much faster than before :-)

[reply]

In Section Seekers of Perl Wisdom