Le mardi 13 juin 2017 à 11:14 -0400, Amar Tumballi a écrit : > All, > > Please update the github issue [1], with what are the metrics you need to > see periodically. These may be metrics which helps you to understand the > health of the process, or the counters which gives insight into things to > monitor bottlenecks. > > I know there exists 'statedump' feature already. It provides some > information, but it also does provide more than required info like dump of > all inode table entries etc. What I am looking here is more of metrics > based on which, we can get a timebased graph. > > A simple example would be what is the number of malloc/free we have done > till now, and what is the total 'in-use' buffers (like you have info in > mem-pool), so you can see depending on work load how the memory usage > varies. A sample implementation i have looks like this [2]. > > Feel free to ask questions, add pointers, and suggestions. This is not > about the tool for plotting graph, more of what should get in the graph. So the first question is: - who is gonna consume the stats ? A sysadmin will not want the same stuff as someone focused on having a SLA to fullfill (like "all request must respond under X seconds"). A team lead or a manager will not care about the same stuff (like, number of client served, to show to $upper_management that the systemm is used). And a developper will not want the same stuff either, as I am quite sure that they are likely the only ones caring about malloc/free, along with people focused on optimisation. -- Michael Scherer Sysadmin, Community Infrastructure and Platform, OSAS
Attachment:
signature.asc
Description: This is a digitally signed message part
_______________________________________________ Gluster-devel mailing list Gluster-devel@xxxxxxxxxxx http://lists.gluster.org/mailman/listinfo/gluster-devel
