//
// Some System Reliability, Availability and Serviceability (RAS) Checks
//

delta = 20 sec;

//
// For Origin systems, sequence number errors are not indicative of
// a problem, but persistent checkbit and/or retry errors may indicate
// a CrayLink interconnect problem.
//
some_inst ( all_sample (
    hw.router.perport.cb_errors @0..2 > 0 || 
    hw.router.perport.retry_errors @0..2 > 0
) )
    ->  alarm 30mins "CrayLink SN and/or Retry errors: " "%i ";


