Kstat_lookup() failing: "Not enough space"

Question

Hi,
I have a peculiar problem. I'm looking up CPU statistics using kstat(), and I'm getting a failure in kstat_lookup() on some machines.
Here is the relevant code:
kstat_ctl_t* pKstatCtl;
kstat_t* pCpuStats;
if ((pKstatCtl = kstat_open()) == NULL)
// Handle error...
if ((pCpuStats = kstat_lookup(pKstatCtl, NULL, -1, "cpu_stat0")) == NULL)
// Handle error...
... // kstat_read(), etc.
I am compiling this program on Solaris 2.6, and it works fine on every 2.6 machine I've used. It is interesting to note that, when I run "truss" on the program and watch the results of iostat(KSTAT_IOC_READ), it returns a number in the range of 600-800. I think this is the number of bytes read.
When I run this program on a Solaris 2.8 machine (without recompiling it there, for reasons which are too complicated to explain here), it works most of the time. However, one of these 2.8 machines is having problems. The kstat_lookup() call fails, and strerror(errno) returns "Not enough space". Note that it's not the kstat_read() that's out of space, it's kstat_lookup(). I tried running the program on another 2.8 machine in the same cluster, and it worked. However, truss told me that ioctl(KSTAT_IOC_READ) was returning 264752.
What's going on here? Why is kstat_lookup() failing with "Not enough space"? Why does this work on one machine and not another?
Thanks,
Chris

Answer 1

Jason - the discs may be advertised as having 4.7GB of space, but in fact this is a bit of an urban legend - it all depends on how you count the bits, in groups of 1000 or in groups of 1024. Disc manufacturers like to count in 1000's, which gives a larger capacity, whereas your computer will count in groups of 1024...
If you allow 4.37GB per disc, you'll be much nearer the mark.