Menu

#25 chkconfig doesn't always check for single instance services

closed
None
5
2004-04-26
2004-03-25
No

The default configuration should be "S initnode Y" The
only time that the "all" or "node=1,2,..." will fail is
if the "S" is specified in /etc/rc.d/ssiconfig. The
configuration should be determined (ssiconfig: in rc,
read from /etc/rc.d/ssiconfig, or default values), then
the checking done if "S" is the type. Right now the
type value is ignored except in the case in which it is
found in /etc/rc.d/ssiconfig.

Discussion

  • David Zafman

    David Zafman - 2004-03-25

    Logged In: YES
    user_id=297844

    Per my description above an M type specified in "ssiconfig:"
    line in RC script doesn't over-ride S type specified in
    /etc/rc.d/ssiconfig.

     
  • Brian J. Watson

    Brian J. Watson - 2004-04-05
    • assigned_to: nobody --> kishoreks
     
  • Vinod Vijayarajan

    • assigned_to: kishoreks --> vvinod
     
  • Vinod Vijayarajan

    Logged In: YES
    user_id=945844

    I have been looking at the chkconfig code. And I don't find
    a check anywhere in the code where the incompatibility
    between 'S' and 'all/node=x,y,..' is being verified.

    The readSSIserviceInfo function does not do anything with
    the value of type (S/M) read in either from
    /etc/rc.d/ssiconfig or the individual rc-files. What should
    it be doing with it anyway?

    Further, at what point in the execution of chkconfig should
    this incompatibility check be done and an error thrown?

     
  • David Zafman

    David Zafman - 2004-04-20

    Logged In: YES
    user_id=297844

    The function validateSSIServiceInfo() is the only one that I
    found which looks at the S vs. M value and only from the
    ssiconfig file. It set a variable "single_instance_svc"
    which is later used to produce an error if an improper node
    change is attempted. This needs to be fixed to perform as I
    described. I don't think that validateSSIServiceInfo()
    should be the point in which this value is read, it should
    be read into the "struct service" structure just like the
    other fields which behave as I described. Then the validate
    code can just check that value from the single_instance field.

     
  • Vinod Vijayarajan

    Logged In: YES
    user_id=945844

    A patch that extends current checks for compatibility
    between S/M values and nodeclass has been checked in. This
    introduces a new field 'instances' into struct service. This
    value is updated in readSSIServiceInfo and verified in
    validateSSIServiceInfo. Should fix the problem

     
  • Vinod Vijayarajan

    • status: open --> closed
     

Log in to post a comment.