On 03/25/09 05:25, James Carlson wrote: > Sangeeta Misra writes: > >>> I agree that there's some duplication here. Having good event logging >>> features and debug dtrace probes would (I expect) remove much of the >>> need for a separate active monitoring command. >>> >>> >>> >> Jim, >> A user will need special privs to use dtrace probe . So if we are to >> abandon "ilbadm monitor" in favor of dtrace probe, would ILB project >> need to provide another authorization( like " >> solaris.network.ilb.dtrace") for this purpose? Or do we simply state >> that the user has to be root to monitor ilbd events ? >> > > Note that I said "good event logging features and debug dtrace > probes." > > If the goal of the person doing this monitoring is to debug the > operation of the load balancer itself, then requiring that user to > gain privileges on his own doesn't seem like a bad thing at all. > He'll need privileges to run mdb and other debug tools. > > If the goal is to measure and deal with system performance issues, > then dtrace is already the tool of choice, so integrating with it > makes things easier, not harder. > > If the goal is for an end user to find problems with a given > configuration, then that's different. End users typically don't care > about internal states and often don't have the ability (or time) to > monitor things in real time as they fail. They need logs that explain > what happened during some failure. > > If the goal is to understand how a proposed configuration would work > in production, then I think the user will need checking and "what if" > tools, sort of like tcpdchk and tcpdmatch. > > The bottom line on this is that I don't quite understand what is > accomplished by real-time monitoring of internal events in this one > subsystem. I wanted to borrow the tracefile notion from in.routed daemon implementation. I recall we used that facility a lot to debug things when a certain action did not yeild a specific result, and the tracefile was used to find out what expected sequence of steps did not occur. But you are right that in a ILB case it will the events are interelated between various subsystems ( servers, backup loadbalancer) .
Sangeeta > A real deployment would likely involve multiple systems, > each running many different services, and it's unclear to me how a > single command-line utility monitoring just one of the components > would provide enough context to solve problems. It seems like an > unusual approach. > > If it makes sense to you, though, then drive on. > > -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://mail.opensolaris.org/pipermail/ilb-dev/attachments/20090325/c70eb047/attachment.html>
