Peter Memishian wrote: > > You're reducing (or changing) the interfaces that can be used > > to get network interface statistics. > > > > While the APIs and ABIs are unchanged, software that works > > today using libkstats may stop working after a "dladm rename". > >That's just life with kstats -- the API is stable but the kstats >themselves are not. This is discussed in the PSARC kstats best >practice (http://sac.eng/cgi-bin/bp.cgi?NAME=kstat.bp): > > Advice > > The kstat API is relatively stable but the data it exports is tagged > by a name and the set of names, and the types and meanings of the data >-> exported with them can be very unstable. It can and does change in >-> patches, let alone OS releases. The names can spontaneously disappear, > change type or be rearranged. In practice most of them don't change > very often, but some do. For example, the addition of some extra > paging stats to cpu_vminfo in Solaris 7, broke sar binary datafile > compatability between Solaris releases. There are other examples. >-> Network interface kstats are in particular are highly unstable, as are >-> some device drivers which are written by 3rd parties. > >
There is no "intro" man page for 3kstats so none of this information is public. Having looked at the manual pages for the relevant kstats library functions, I did not find any mention of anything being private or unstable, only that the library interfaces themselves are stable. This doesn't lead people to draw very good conclusions about how reliable kstats info is :/ Darren _______________________________________________ networking-discuss mailing list [email protected]
