Planning Analytics

Planning Analytics

Get AI-infused integrated business planning

 View Only
Back to discussions
Expand all | Collapse all

TM1 adminhost redundancy issue

  • 1.  TM1 adminhost redundancy issue

    Posted yesterday

    Hello community,

    In my company we have around 40 TM1 applications in production. Adminhosts are a key element of this architecture, and we have decided to implement redundancy for this component. Declaring two adminhosts for each and every TM1 instance is supported by the product (semi-column separated server names in tm1s.cfg). However we have diagnosed a misbehavior regarding this configuration, the adminhosts sometimes not responding with the same instance list... I reported the incident to IBM, the answer was "works as designed". I am sharing the issue here with you to collect your feedback, and maybe have IBM folks revise their answer...

    Issue summary: if one of the two adminhosts declared in tm1s.cfg is temporarily unavailable, a database being restarted may never be referenced there.
    Context:
    -  The vanilla SDATA instance
    -  Two adminhosts ah1 and ah2, on two different machines
    -  SDATA is declared on the two adminhosts in tm1s.cfg: AdminHost=ah1;ah2
     
    Scenario:
    1-  We start ah1 (ah2 is stopped)
    2-  We start SDATA
    3-  SDATA ignores ah2 and starts without an error, we notice a line "Starting Admin Server poller thread. Thread Id: 9636. For Admin server: ah1, LocalServer: False"
    4-  SDATA is visible on ah1
    5-  We stop and restart ah1
    6-  After a few seconds, SDATA is visible on ah1 (polling mechanism from SDATA). This behavior is expected
    7-  We start ah2
    8-  BUG - SDATA never appears on ah2
    9-  A REST call is documented on this community in order to add SDATA to ah2. It indeed works, but does not resist a restart of ah2 because this call does not trigger any polling mechanism from SDATA to ah2.
     
    Our conclusions:
    1. If only one adminhost, the database does not start successfully until the adminhost is available (retry mechanism is implemented, tested successfully)
    2. If two adminhosts, one of the adminhosts may be unavailable, this does not prevent the database from starting (this is a nice and expected behavior)
    3. Proposed resolution: in case of redundancy, a "poller" should be activated on each of the declared adminhosts, even if one of them is temporarily off.


    ------------------------------
    Francois-Xavier Menard
    ------------------------------


Related Content