Planning Analytics

 View Only
  • 1.  Massive No lock messages when saving data happens concurrently

    Posted Tue March 17, 2020 09:59 PM
    I know it is warning but there are massive "No lock" info in the sever log when two saving processes running at the same time. Is there a way to suppress this? 
    13428 [1f43] WARN 2020-03-18 11:09:33.160 TM1.Lock Attempt to release lock on 0x00000191F23D1A68(0x00000191F23B9A10). No lock held by module c:\jenkins\j12\workspace\build-tm1_main-tm1server-win64-production\src\tm1server\prod\tm1_r7s\tm1serverimpl.cpp at Line 16703
    13428 [1f43] WARN 2020-03-18 11:09:33.162 TM1.Lock Attempt to release lock on 0x00000191F23D1A68(0x00000191F23B9A10). No lock held by module c:\jenkins\j12\workspace\build-tm1_main-tm1server-win64-production\src\tm1server\prod\tm1_r7s\tm1serverimpl.cpp at Line 16703
    1


    ------------------------------
    mvp morgan
    ------------------------------

    #PlanningAnalyticswithWatson


  • 2.  RE: Massive No lock messages when saving data happens concurrently

    Posted Wed March 18, 2020 12:09 PM
    Hello,
    You can check the log properties file (tm1s-log.properties) and see if TM1.Lock line is set to something. Otherwise, try adding the line below then save to see if it suppresses these.
    log4j.logger.TM1.Lock=ERROR

    ------------------------------
    Tom Kim
    ------------------------------



  • 3.  RE: Massive No lock messages when saving data happens concurrently

    Posted Wed March 18, 2020 06:58 PM
    Hi 

    Am I correct in understanding that you have two processes running at the same time, both doing a SaveDataAll to write data to disk? If so, then as far as I am aware the standard IBM advice is not to do this. It is best to have SaveDataAll in a Chore that runs periodically where the process that runs the SaveDataAll is the only process in that Chore. 

    We have had huge problems with SaveDataAll. We do not run it during the day during busy month end periods, as in the past it has caused crashes. Your situation may be different but we typically have 100+ users all entering data and running processes.. 

    Not flushing data to disk has its risks if the server crashes, but if running a SaveDataAll causes a crash then you are possibly even worse off.

    Regards

    Paul SImon

    ------------------------------
    Paul Simon
    ------------------------------



  • 4.  RE: Massive No lock messages when saving data happens concurrently

    Posted Thu March 19, 2020 10:41 AM

    I came across this from QueBit a while ago.

    SaveDataAll runs in serial however you can speed it up by running CubeSaveData in parallel. 


    https://quebit.com/askquebit/IBM/tm1-save-data-to-tm1-cube-save-data/



    ------------------------------
    Robby Meyers
    ------------------------------



  • 5.  RE: Massive No lock messages when saving data happens concurrently

    Posted Thu March 19, 2020 06:53 PM
    Hi

    The problem withe CubeSaveData is that in some cases having some cubes saved and not others can make it harder to recover from an outage than having all cubes in a consistent state. However, it will depend on the particular model. Most models I have seen have a few large cubes with most of the data, so the reduction in locking by saving them individually may not be that great. 

    The issue on locking is interesting it might be behind some of crashes that we have had, which is why we know only save cubes to disk out of hours.

    Regards

    Paul Simon

    ------------------------------
    Paul Simon
    ------------------------------



  • 6.  RE: Massive No lock messages when saving data happens concurrently

    Posted Thu March 19, 2020 07:50 PM

    I think the concept was that instead of having to save all cubes in serial, you could build a parallel process to save some or all cubes using multiple threads.

     

    In theory you could run a parallel process that incremented through  X cubes at a time.  So if you ran 8 cube saves at once and kept processing until all cubes had completed you should roughly 8x faster than a CubeSaveDataAll.

     

     

     

     

     






  • 7.  RE: Massive No lock messages when saving data happens concurrently

    Posted Thu March 19, 2020 09:21 AM
    The messages are there as a WARN because they indicate TM1 server mishandled its bookkeeping of thread/lock state which might be a bigger issue.
    Trying to release an extra lock that's not held won't itself cause issues, but it can indicate a previous issue that may or may not be more serious.

    If you can reproduce it consistently, I would suggest to open a case with IBM and attach the testcase, so we can take a look at it and potentially fix it.

    ------------------------------
    VLADIMIR KIM
    ------------------------------



  • 8.  RE: Massive No lock messages when saving data happens concurrently

    Posted Fri March 20, 2020 03:35 AM
    Hi, guys
    Thanks for your advice. Yes, I probably will raise the ticket with IBM and for the moment, yest, CubeSaveData would be a good option and I haven't used it a lot to be honest.

    ------------------------------
    mvp morgan
    ------------------------------