What memory usage percentage values should we be alarming at on Linux machines running Cassandra?
I've seen from one question on here that we should be disabling swap memory.
From the Three percentages on the Linux protocol for 'physical memory usage', 'actual physical memory usage' and 'real memory usage', from reading up on it the only one (especially with swap disabled) that we should really care about is 'actual physical memory usage' as it excludes cached memory, or do we need to alarm on the others as well still, and if so what values?
As this question has now been inactive for a long time, I will close it. If you still want more information about this, could you post a new question?
Hi Philip,
I believe the answer greatly depends on your setup, usage and situation. Hence, it's difficult to give a definitive answer.
If you have no clear limit in mind on when action should be taken, it might be more interesting to look at Anomaly detection to generate alarms. Here, DataMiner will look at the trend of your parameter and create alarms if some unforeseen anomaly occurs. You can for example have an alarm trigger when a sudden level shift of 20% occurs, regardless of the absolute values at the time. This also keeps expected patterns in mind, so only the true anomalous occurrences trigger an alarm.
This feature is documented here, and specific information on how to set up alarm templates using this can be found here.
Note that this feature requires DataMiner to learn the pattern of a parameter, so it does require a week or so of data before it can recognize an unexpected change.
I see that this question has been inactive for some time. Do you still need help with this? If not, could you select the answer (using the ✓ icon) to indicate that the question is resolved?