I just posted this to the forum thread on this topic: --------------- Folks,
It seems like this continues to be a problem, so I thought I would share a couple of ideas on where the underlying problem might be. While this won't get any of you _directly_ closer to solving your problems, it might be of some help to Macromedia in narrowing down the problem. Understand that this is all just theory offered to try and assist an eventual solution, so please don't attack me if I'm off base. Here goes... 1) The client.properties file is performing the same function as an Oracle Sequence object declared with CACHE = 100. This is a very efficient model for high transaction systems, as the requester simply queries a lightweight object pinned in memory that simply gets the next pre-cached value from an array, increments the index into the array, checks if the new index value is equal to the CACHE value (or directly tests the issued value against the value at the end of the array), and if so updates the persistent "placeholder" for the sequence value, which is in the form of a text file. 2) I assume that the LastID is "forward-looking" to the beginning of the next sequence series being cached, rather than "backward-looking" to the start of the current sequence series cached, since after a restart the mechanism must enable CFMX to pick right up where it left off without having to check anything other than this one persistent file. Anything else would cause overlapping values to be issued from the sequence, which would in turn overlap CFMX client IDs and throw everything off. 3) Since the sequence object pinned in memory must modify internal values every time it is called, all access to it would be synchronized through an exclusive lock on the object. If the system must suffer the mechanical bottleneck of an exclusive lock with each request, it must do so in the least resource-intensive manner possible. This is the main reason why this whole client.properties/sequence object mechanism exists in the first place: to prevent the performance bottleneck that would occur if the system had to exclusively lock and manipulate a file each time an ID request was made. So instead, CFMX just locks an existing object in memory that already contains 100 requests' worth of data and one simple index value that must be modified. Simultaneous requests of the object are queued, but not for very long (in terms of nanoseconds). Except for the request that gets the last value in the array, as this triggers the modification of the client.properties file. 4) Because access to the client.properties file is managed by the OS, OS-level mutex objects (locks) handle sequential I/O from this file, which is certainly slower than a mutex on an object pinned in memory. If CFMX has a problem with its "acquire lock/read file/write file/release lock" procedure on the client.properties file, then that would cause one of two things: either an empty client.properties file, or an unchanged client.properties file. 5) This is where their internal exception handling model comes into play, because what do you do if the modified file cannot be written to disk? What if CFMX couldn't acquire an exclusive lock in the time Macromedia's programmers allotted? Everything pivots on the two critical steps of acquiring an exclusive file lock and writing a properly-formatted file while the lock is in place. Keep this in mind as we discuss the subject further. 6) There is an anomaly in the system that causes the ID to return a value that is not one but _two_ more than the previous issued value. This makes me focus on the exact mechanism used to trigger the modification of client.properties. Is the test a match of the index value to the size of the array, or is it a match of the value retrieved to the last value in the array? If the former, this anomaly will cause some of the next batch of sequence values issued by the object to overlap those it had already issued from the previous batch. If the latter, we need to ask another question: is the test an equality test, or a "greater than or equal to" test. If a strict equality test, this anomaly might cause the test to sometimes fail. What happens to the next ID request after this "end of array" test fails? Is the result a NULL value, perhaps? 7) So now we're back to updating client.properties. With what value? If I were the engineer, I wouldn't take the time to read the file, parse the contents, extract the string containing the LastID value, convert it to an integer, and increment it. I would simply take the value I have in hand (issued by the object), increment that by 100, construct the new string contents of the client.properties file, and overwrite the file. So, if this issued value is NULL, and you try to concatenate that NULL value with any other value, the result will be NULL, right? So the file that is written to disk (if not checked for empty content) is empty. 8) Or what if an error _is_ thrown trying to update the client.properties file? If you were the engineer, how would you handle such a problem so central to the entire functionality of the system? You can handle the error as an exception, then route that exception to a retry of the update, but you'd have to quit retrying after a certain amount of time or retries, otherwise you could put the system into an endless loop. Perhaps retrying with a longer allowable time limit to acquire the exclusive lock on client.properties? Anyway, all this is just theory, but I figured that _any_ input from anywhere at this point might help somewhat. Maybe this discussion sparks an idea in the mind of a CFMX-based ISP who observed some seemingly unrelated behavior that might have bearing on this issue. Anyway, I would investigate further by asking all of you to ask yourselves these questions: * Do you ISPs see this problem equally on shared servers and dedicated servers? * Do any of you see this problem happening on some of your servers but not on others? If so, what are the differences between them? * Does the frequency of this problem parallel the load on the system having the problem? * Can you reproduce this problem by turning off cookies, closing the browser, re-launching the browser, and going back to the site? If so, does this also trigger problems afterward for other clients who have not turned off their cookies? * Does the problem _always_ disappear after deleting cookies, closing the browser, cycling the CFMX service, re-launching the browser, and making a new request? * Can you reproduce this problem by modifying the CFID value in the URL? * Have you observed any pattern or semi-pattern for the problem appearing or disappearing? Guys, I hope this discussion can somehow lead to a solution. If I'm in error on any of this then please understand that I'm only theorizing as to how it *might* work behind the scenes and how that might lead to the problem itself -- and therefore help Macromedia find a solution. Respectfully, Adam Phillip Churvis Advanced Intensive ColdFusion MX Training http://www.ColdFusionTraining.com E-mail: [EMAIL PROTECTED] Phone: 770-446-8866 Team Macromedia Volunteer for ColdFusion http://www.macromedia.com/support/forums/team_macromedia/ ----- Original Message ----- From: "Mark Leder" <[EMAIL PROTECTED]> To: "CF-Talk" <[EMAIL PROTECTED]> Sent: Monday, October 28, 2002 10:06 AM Subject: What in the world is this? > In the application.cfm, right after the CFAPPLICATION tag (running on > Win CFMX) I get the following NULL - however the line in question is a > comment tag???: > ========================== > null > > The error occurred in C:\Inetpub\website\admin\Application.cfm: line 24 > > 22 : <!--- This page creates global variables for use throughout the > site ---> > 23 : > 24 : <!--- Toggles the security check 0=off 1=on ---> > 25 : <CFSET REQUEST.securitycheck = "0"> > 26 : > ========================== > > Thanks, > Mark > > ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~| Archives: http://www.houseoffusion.com/cf_lists/index.cfm?forumid=4 Subscription: http://www.houseoffusion.com/index.cfm?sidebar=lists&body=lists/cf_talk FAQ: http://www.thenetprofits.co.uk/coldfusion/faq Structure your ColdFusion code with Fusebox. Get the official book at http://www.fusionauthority.com/bkinfo.cfm
