• (nodebb)

    This code is ACID-compliant - when you see it, your eyes are burned to the point of going blind.

  • (nodebb)

    Back in the day (TM) I was a developer of two different 'database' systems that did not have transactions. The one you are more likely have heard of was the Pick System. (Named after Dick Pick) If you don't have transactions you have to do interesting things to make sure that your data doesn't get corrupted. It doesn't seem like the author of this TDWTF was ever taught that transactions existed. Or exceptions. Or log files.

  • (nodebb) in reply to Rick

    It doesn't seem like the author of this TDWTF was ever taught that transactions existed.

    Maybe the said author was working with MySQL before it added transactions...

    Or exceptions.

    The article is clear that the whole situation broke because stuff suddenly started throwing exceptions for what were, in fact, given the overall "design" of the system, actually real errors. Of course, the system shouldn't have been designed like that, but with that "feature" in its design, they are real errors. It's much too long since I did any work with SQL of any flavour that I'm willing to propose an alternative, unfortunately. (Or fortunately, perhaps...)

  • (nodebb)

    I dunno what to write about this article because the implementation seems so crazy, I can't even imagine what they try to solve.

  • George (unregistered)

    Devs: 100% sure nobody will use this s**t, just limit the import to 10k. If this hits 10k I will eat my hat!

  • (nodebb)

    10,000 bottles of beer in the database, clearly too drunk to succeed.

  • Tim (unregistered)

    Although SELECT COUNT is better than SELECT *, it still forces the DB engine to count the rows. I normally use SELECT DISTINCT 'X' because at least that tells the DB you don't care how many rows and you only want to know whether there are any.

    obviously if the normal case is no rows then that probably makes no difference

  • gg (unregistered) in reply to Tim

    I normally use SELECT DISTINCT 'X' because at least that tells the DB you don't care how many rows and you only want to know whether there are any.

    On a sufficiently dumb DBMS this will scan the table anyway. Why not simply LIMIT 1 (FIRST 1 depending on dialect)?

  • (nodebb)

    Best is the simple SELECT MAX(id) FROM data_import

  • (nodebb) in reply to Tim

    There is a few DB out there (MsSQL & PostgresSQL AFAIR) where the DB actually stores the row count, so SELECT COUNT(*) is basically a noop. Still have the round trip over the network, so SELECT LIMIT is the way to go for that one in an transaction scope with the DELETE.

  • Richard Brantley (unregistered) in reply to George

    "If this hits 10k I will eat my hat!"

    By the time this hits 10k, I will be long gone from here and it will be someone else's problem!

  • (nodebb) in reply to mynameishidden

    Take one down, pass it to ROUND(), give it permission to read

  • (nodebb)

    It sounds as if that smart boss has been around since there were only 9 items in the database.

  • (nodebb)

    Well, transactions or no transactions, row counts or not, and for anyone that didn't figure it out from the original article, the right way to implement this thing is, of course, to add an extra column with the import status...

    Or hey, do as they do in the system I work with... keep a completely separate import-database where you have table's (with the same name as the original) you can insert in and then have triggers on those tables do the importing on the fly........ (NO! I'm kidding... don't do that!)

  • mihi (unregistered) in reply to MaxiTB

    A simple COUNT() will not help you if you initially had less than 5000 rows, incremented the IDs and then the process broke before more 5000 rows were inserted. The COUNT() will still be less than 10000, but your crude error check should throw. And a COUNT(*) WHERE id > 10000 will not benefit from this optimization.

  • Loren Pechtel (unregistered)

    I think this code evolved.

    Originally it simply replaced the data with the new data. Something went wrong one day and they wrote this to provide a rollback without altering the table. Yeah, that's what transactions are for, but we see plenty of things where someone obviously wasn't aware of some feature and did what they could.

  • Conradus (unregistered) in reply to Rick

    "(Named after Dick Pick)"

    Yes, really. You can't make this stuff up, folks.

  • NobodySpecific (unregistered) in reply to Loren Pechtel
    I think this code evolved.
    "Evolved" is not the word I'd use here; it implies that this code improved from its original shape in some way. I think a better way to describe its change is mutated.
  • (nodebb)

    I've worked with systems in decades past that needed this kind of thinking. And with devs who did this kind of thinking. I've been the sort of dev who did something like this ... and hated that I had to. Or didn't know any better way.

  • Gnasher729 (unregistered)

    Boss was right. I’d have changed from 10,000 to 1,000,000 and added a visible warning if the count is > 20000 to get some idea of the behaviour.

Leave a comment on “The First 10,000”

Log In or post as a guest

Replying to comment #:

Log Out

« Return to Article