Last week, when I was working on my jQuery Photo Tagger plugin (for Flickr-style photo annotation), I was using CFLock to create thread-safe cache updates in the ColdFusion aspect of the code. As I was doing that, it occurred to me - I wasn't 100% sure what would happen if the timeout (indicated by the CFLock tag) was exceeded and no error was thrown? In the CFLock tag, you can use the ThrowOnTimeout attribute to determine the action taken by the thread if the a lock cannot be successfully obtained in the allotted time; but, what would that action actually be? Would the paused thread simply enter the CFLock tag body? Or, would it skip it?
| || || || || |
| || |
| || || |
I decided to get to the bottom of this by creating two tags that would compete for the same named lock. One thread pauses before the CFLock; one thread pauses whilst inside the CFLock. In this way, they get to it at different times and compete head-to-head for ultimate victory!. Here is the first page (a.cfm) which is the one that will have to wait for the CFLock:
- A: Before the CFLock tag.<br />
- Pause this thread to make sure the other thread gets
- to the CFLock first.
- Get an exclusive lock (this named lock is shared with
- the other thread). Notice that we are asking the page
- NOT to throw an error if the lock cannot be made within
- the given timeout.
- A: Inside the CFLock tag.<br />
- A: After the CFLock tag.
Notice that the CFLock tag specifies that no error should be thrown if the lock cannot be gotten. The second page (b.cfm) gets to the CFLock tag first and then pauses, once inside, for a time that is greater than A's specified timeout:
- B: Before the CFLock tag.<br />
- Get an exclusive lock (this named lock is shared with
- the other thread).
- B: Inside the CFLock tag.<br />
- Once inside the tag, let's sleep the thread so that
- the lock will be held for a while (preventing the other
- page from entering within the given selected timeout).
- duration="#(3 * 1000)#"
- B: After the CFLock tag.
Notice that this page pauses its lock for 3 seconds, which is greater than the one second timeout specified in page A.
I then put these two pages head-to-head in a Frameset (I won't bother showing the code for that) to see what would happen. When I ran the two pages at the same time, I got the following output from A:
A: Before the CFLock tag.
A: After the CFLock tag.
... and the following output from B:
B: Before the CFLock tag.
B: Inside the CFLock tag.
B: After the CFLock tag.
As you can see, when page A failed to get a lock in the specified timeout, it simply skipped over the CFLock tag altogether. This is good; this is the outcome that I was hoping for (and one that I have coded against) and it's nice to see that it is what actually happens. At least I don't have to go back and fix my code.
So, why would one not want to throw an error if a lock was not obtained? I think this is definitely a rare case, but one that I have used a few times. Typically, I will use this if I am inserting / mutating large amounts of database records containing very non-critical information. In such a case, I might not care if periods of intense performance fail to insert a few rows here and there.
Another place where I actually use this a lot is within a scheduled task. If I have a scheduled task that might take a while to run, I will typically wrap the entire task in a named CFLock tag with a timeout of one second. This way, if the scheduled task tries to execute while the previous instance is still running, the scheduled task will simply skip over itself.
After you've written a bunch of code under a given assumption, it's always a bit scary to realize that you've never tested to confirm the behavior you were expecting. When it comes to CFLock, it's nice to see that when a thread cannot obtain a lock in the given timeout, it skips over the tag altogether.
I'd never thought to use a named lock to guarantee that frequently running scheduled tasks wait for themselves! Genius.
I can't believe that the minimum timeout you can use is 1 second. That's like 10 minutes in Internet years.
Thanks my man - it's a method I've found very useful for long running tasks.
Why would you NOT want it to execute the code in the CFLOCK tag? If you don't care if it is executed, then don't include the code on the page.
Also, can you explain this to me? "In such a case, I might not care if periods of intense performance fail to insert a few rows here and there."
When would there be an instance when you didn't care if it failed to insert a few rows here and there? That seems to raise a red flag on data integrity.
This kind of technique is actually great for scheduled tasks that might hand or potentially overlap in a way that is not appreciated:
Question. What would you consider a good standard setting for the timeout value?
We have an issue with a new CF Unix server that is failing load tests constantly (it is configured to the bare minimum specs required in terms of memory, and would take an Act of God to get more memory) and the hardware folks are claiming that our usage of CFLOCK is the issue for the failures.
I am of the opinion that exclusive locks are the only kind needed, that readonly are nice to have but not critical. We have exclusive locks in place anytime an application or session variable is set. We have a 10 second timeout as a default value. They want us to change it to 2-3 seconds. Every CFLOCK will throw an error, btw.
Basically my opinion is that if the server cannot handle needing 10 seconds to acquire a lock, shrinking the value will only make things worse. Also I don't think 10 is an unreasonable value, but I'd like to hear what others think.