Posts tagged "cf_tip mysql"

Safely Selecting the Last Inserted ID, Part II

Well, as it turns out, this isn’t
an issue at all. I ran two tests. The first followed these steps:

  1. Using a CFQUERY tag, I executed an insert statement.
  2. Used Thread.sleep() to make the page hang for 10 seconds.
  3. Inserted ten additional rows from the command line (using the MySQL command
    line tool).
  4. After the Thread.sleep(), used LAST_INSERT_ID() to get the last inserted ID,
    and displaed it.

I expected the result to be 11, but it was actually 1.  The 10 rows I inserted
between the initial insert and when I selected the last inserted ID had no effect.  I
figured this was because ColdFusion was considered one client, and the command
line administrator was considered a different client, and since the scope of
the last inserted ID is the client/connection, I was protected.  So I ran this
test, instead:

  1. Executed an insert statement in ColdFusion.
  2. Slept for 10 seconds.
  3. Ran a second CFM page that inserted 10 rows while the first page was hung.
  4. After the thread in the original page woke up, selected the last inserted id
    and displayed it.

Keep in mind I was using no transactions or locks since I was trying to get a
wrong result before trying to figure out how to get the correct one, however
the result was once again 1.  As it turns out, no transactions or locks are
needed.  Why?  Apparently because each request gets its own database connection,
and that one connection gets reused for the duration for the request, regardless
of how many database operations you perform.  Other
requests can happen simultanously, of course, but they all get their own connections,
and hense do not affect the first connection’s last inserted ID value.  In
other words, it seems to work exactly how you would want it to!  ColdFusion
does it again!

Disclaimer: This test was run using JRun, CFMX 6.1 (with the
latest updater) and MySQL
version 4.0.18.  Before you rely on this data, you might want to run some tests
yourself, though I will try to get confirmation from the ColdFusion team that
this technique should work across all databases and future versions of ColdFusion.

Safely Selecting the Last Inserted ID

I run into this issue occasionally, and always find some kind of work-around,
but I’ve finally decided to address it head-on. The scenario is this:

  1. You insert
    a record into the database.
  2. You need the ID of the row you just inserted so you can insert additional
    rows in other tables referencing the first row you inserted.
  3. You’re worried about your code being thread-safe, so you consider
    using locks, transactions, UUIDs, etc. But what is the best way?

The easiest and most database-independent way of doing this is to generate a UUID
and use it for the primary key for the first insert, and the foreign key for any
subsequent inserts. That’s fine for certain apps, but for the app I’m working on
now, I need to use sequential, auto incremented integers as keys. I’m using the
MySQL function LAST_INSERT_ID() to get the last inserted ID, however the ID’s scope
is the current connection, so if I use two CFQUERY tags (one to insert the row,
and one to get the last inserted ID), I believe it’s possible I could get the wrong
ID if another row is inserted between the time I do my insert and the time I select
the last inserted ID. So I’m going to try:

  1. Sending two queries to MySQL at the same time. One to do the insert, and one
    to do the select. Both statements will use the same connection, and problem
    solved. I haven’t had any luck with this so far, though. I’m not sure MySQL supports
    this as I keep getting SQL syntax errors.
  2. Putting both CFQUERY tags in a transaction (they actually already are in a
    transaction), and using the isolation level of serializable. I’m potentially
    sacrificing some performance, but it should be miniscule. This is a very
    good solution if it works, but I’m not 100% positive that it will work, though
    I am hopeful.
  3. Using an exclusive lock. If all else fails, this will definitely work, although
    the disadvantage is that I will only be taking database operations that
    originate from one CF server into account. In other words, another server that
    is unaffected by the lock can slip a new row into the database at just the right
    moment and trip me up. I would much rather the lock occur at a lower level — actually
    in the database, if possible.

I’ll post my results. In the meantime, anyone have any additional insight?