Hi David,

Thanks for reporting this.  It turns out this is the expected behavior when accessing these PCP metrics from kernels that do not provide the underlying instrumentation in /proc/vmstat.  It's a moving feast with this procfs file; some fields are transient depending on kernel version, architecture and kernel-build-options.

cheers.

Hi David,

the PCP Linux PMDA code is scanning for these strings :

$ grep \"pgsteal proc_vmstat.c
    { .field = "pgsteal_dma",
    { .field = "pgsteal_dma32",
    { .field = "pgsteal_high",
    { .field = "pgsteal_movable",
    { .field = "pgsteal_normal",

which are mapped to these metrics :

$ pminfo | grep pgsteal
mem.vmstat.pgsteal_dma
mem.vmstat.pgsteal_dma32
mem.vmstat.pgsteal_high
mem.vmstat.pgsteal_movable
mem.vmstat.pgsteal_normal

but /proc/vmstat on my system (Fedora-25) only has the following "pgsteal" metrics available :

$ grep pgsteal /proc/vmstat
pgsteal_kswapd 1172576
pgsteal_direct 1126

On your system you have :

[root@fireball ~]# grep pgsteal /proc/vmstat
pgsteal_kswapd_dma 0
pgsteal_kswapd_dma32 0
pgsteal_kswapd_normal 0                                            <-- metric here
pgsteal_kswapd_movable 0
pgsteal_direct_dma 0
pgsteal_direct_dma32 0
pgsteal_direct_normal 0                                               <-- metric here
pgsteal_direct_movable 0

Since you're running pmval mem.vmstat.pgsteal_normal which needs "pgsteal_normal" in
/proc/vmstat the result is No Values Available, which is correct.

I acknowledge there are quite a few combinations of the pgsteal metrics that PCP is
not reporting, e.g. "pgsteal_kswapd_normal" and "pgsteal_direct_normal", and others.
So we could add support for those missing metrics. But the bug as it stands is NOTABUG
because the values you're after are not present in /proc/vmstat on your system.

TO me it looks like pgsteal_normal may have been split into two kswapd_normal and
direct_normal in newer kernels ... I haven't checked this in the kernel but can if
you like. If that's the case, we'd rather just add support for the new metrics but
keep the current pgsteal_normal (and similarly named metrics) because they would
still be valid on older kernels.

Regards
-- Mark Goodwin

Will work on this when I get time, as per Comment #4 (add the missing metrics).

Mark, Yes that sounds great.
Thank You

With the help of our friends at Fujitsu, this work has been done in the upstream PCP community.  Just a bit of automated regression testing work remains to complete it but that should be no problem at all - I'll get that done.

Reproduced against pcp-3.10.6-2.el7 on rhel-7.2 and verified against pcp-3.12.2-5.el7 on rhel-7.5 (kernel-3.10.0-843.el7).

Since the problem described in this bug report should be
resolved in a recent advisory, it has been closed with a
resolution of ERRATA.

For information on the advisory, and where to find the updated
files, follow the link below.

If the solution does not work for you, open a new bug report.

https://access.redhat.com/errata/RHBA-2018:0926