A regression due to work in Bug 1028624.

Only one MeasurementData per unique timestamp, per measurement report, is forwarded for alerting evaluation.

So, if a MeasurementReport contains multiple datums with the same timestamp, only one will make it to alerting for evalaution.  In this situation if there is an alert definition dependent on the omitted data, it will not fire as needed.

This issue is exacerbated by many metrics being set for the same collection schedule.  So, to some degree the issue can be alleviated by having relevant metrics collected on a schedule that does not coincide with other collections performed by the same plugin (not so easy to do with a lot of metrics being collected)

To simulate the issue you can try the following:
1) Create an alert def on Free Swap Space for a platform. 
   ** Choose a threshold value that should always cause the alert to fire.
2) Set collection interval for Free Swap Space to 1 minute
   ** note - I am assuming all default collection intervals, so all of the
      other metrics should have much slower collection times.
   ** you should start to see an alert every minute
3) After several alerts have fired then select all of the active metrics,
   including Free Swap Space (so it is synced up), and set collection the
   intervals for all of them to 1 minute (all together, not one at a time)
   ** The alerting will likely stop, as now all of those metrics likely have
      the same timestamp and only one is making it to alerting.
4) After a pause, reset all by Free Swap to 20 minutes.
   ** Alerting should resume.