Page 1 of 1
Is ReadersBypassWriters a valid setting in tm1s.cfg?
Posted: Sun Sep 27, 2009 2:19 pm
by jonathan.d
Hi All,
Wondering if ReadersBypassWriters=T/F is a valid setting in tm1 9.4 fp2?
Apparently, it was one in tm1 8.x release?
Nothing found in existing documentation.
Also, what does the setting offer?
Any clues?
Thanks!
Re: Is ReadersBypassWriters a valid setting in tm1s.cfg?
Posted: Sun Sep 27, 2009 7:10 pm
by Alan Kirk
jonathan.d wrote:Hi All,
Wondering if ReadersBypassWriters=T/F is a valid setting in tm1 9.4 fp2?
Apparently, it was one in tm1 8.x release?
Nothing found in existing documentation.
Also, what does the setting offer?
Any clues?
I can answer the last question but not the first with certainty; however I
suspect that it may have been made redundant after the new locking model came in in 9.1.
The parameter appears in the Admin guides for version 8.2 to 9.0. However in the 9.0 manual it has a honking great "CAUTION" icon next to it which warns that:
The Tm1s.cfg parameters identified by a {honking great "CAUTION"} icon are restricted to Applix employees only. Any unauthorized use of these parameters in a production system disqualifies your system for support under your Applix support agreement.
As for the parameter itself:
Under normal circumstances, a client attempting to write to the TM1 database waits until all prior read requests are executed, and then applies a write lock to block any subsequent read requests, until the process of writing values to the database is complete.
When ReadersBypassWriters is included in the Tm1s.cfg file and set to T, a client attempting to write to the TM1 database waits until all prior read requests are executed, allows any subsequent read requests to execute, and begins writing only when no incoming read requests are detected. After the client attempting to write to the TM1 database determines that there are no pending read requests, a write lock is applied to the server and the client is allowed to write values to the TM1 database.
Also the parameter looks like it has always been a bit flaky. This from the 8.2.12 release notes:
TM1 Server Crashes when TurboIntegrator Process LockOff Function is Used in Conjunction with ReadersBypassWriters Parameter (297484)
In previous versions of TM1, when the ReadersBypassWriters parameter was enabled on the TM1 server and the LockOff function was executed from a TurboIntegrator process, it was possible for the process to prevent other users from acquiring a WRITE lock indefinitely. This was possible because ReadersBypassWriters dictates that readers always take precedence over writers; as long as the process continued to issue READ requests, all WRITE requests would be placed the end of the processing queue. A customer encountered just such a situation that led to a TM1 server crash.
In TM1 8.2.12, when the ReadersBypassWriters parameter is enabled on the TM1 server and the LockOff function is executed from a TurboIntegrator process, the process will suspend execution every 100 records to check for pending WRITE requests in queue. If any pending WRITE requests are detected, the process relinquishes its lock to the WRITE request, and then places a READ lock request in queue.
Re: Is ReadersBypassWriters a valid setting in tm1s.cfg?
Posted: Mon Sep 28, 2009 7:16 am
by jonathan.d
Definitely flaky.
As always, appreciate your advice Alan.
Thanks.