Re: [PATCH v10 09/12] perf record: implement -z,--compression_level[=<n>] option

From: Arnaldo Carvalho de Melo
Date: Fri May 17 2019 - 11:03:47 EST


Em Wed, May 15, 2019 at 06:44:29PM +0300, Alexey Budankov escreveu:
> On 15.05.2019 15:59, Arnaldo Carvalho de Melo wrote:
> > Em Wed, May 15, 2019 at 11:43:30AM +0300, Alexey Budankov escreveu:
> >> On 15.05.2019 0:46, Arnaldo Carvalho de Melo wrote:
> >>> Em Tue, May 14, 2019 at 05:20:41PM -0300, Arnaldo Carvalho de Melo escreveu:
> >>>> Em Mon, Mar 18, 2019 at 08:44:42PM +0300, Alexey Budankov escreveu:
> >
> >>>>> Implemented -z,--compression_level[=<n>] option that enables compression
> >>>>> of mmaped kernel data buffers content in runtime during perf record
> >>>>> mode collection. Default option value is 1 (fastest compression).
> >
> >>> <SNIP>
> >
> >>>> [root@quaco ~]# perf record -z2
> >>>> ^C[ perf record: Woken up 1 times to write data ]
> >>>> 0x1746e0 [0x76]: failed to process type: 81 [Invalid argument]
> >>>> [ perf record: Captured and wrote 1.568 MB perf.data, compressed (original 0.452 MB, ratio is 3.995) ]
> >
> >>>> [root@quaco ~]#
> >
> >>> So, its the buildid processing at the end, so we can't do build-id
> >>> processing when using PERF_RECORD_COMPRESSED, otherwise we'd have to
> >>> uncompress at the end to find the PERF_RECORD_FORK/PERF_RECORD_MMAP,
> >>> etc.
> >
> >>> [root@quaco ~]# perf record -z2 --no-buildid sleep 1
> >>> [ perf record: Woken up 1 times to write data ]
> >>> [ perf record: Captured and wrote 0.020 MB perf.data, compressed (original 0.001 MB, ratio is 2.153) ]
> >>> [root@quaco ~]# perf report -D | grep PERF_RECORD_COMP
> >>> 0x4f40 [0x195]: failed to process type: 81 [Invalid argument]
> >>> Error:
> >>> failed to process sample
> >>> 0 0x4f40 [0x195]: PERF_RECORD_COMPRESSED
> >>> [root@quaco ~]#
> >
> >>> I'll play with it tomorrow.
> >
> >> Applied the whole patch set on top of the current perf/core
> >> and the whole thing functions as expected.
> >
> > It doesn't, see the reported error above, these three lines, that
> > shouldn't be there:
> >
> > 0x4f40 [0x195]: failed to process type: 81 [Invalid argument]
> > Error:
> > failed to process sample
> >
> > That is because at this point in the patch series a record was
> > introduced that is not being handled by the build id processing done, by
> > default, at the end of the 'perf record' session, and, as explained
> > above, needs fixing so that when we do 'git bisect' looking for a non
> > expected "failed to process type: 81" kind of error, this doesn't
> > appear.
> >
> > I added the changes below to this cset and will continue from there:
> >
> > diff --git a/tools/perf/builtin-record.c b/tools/perf/builtin-record.c
> > index d84a4885e341..f8d21991f94c 100644
> > --- a/tools/perf/builtin-record.c
> > +++ b/tools/perf/builtin-record.c
> > @@ -2284,6 +2284,12 @@ int cmd_record(int argc, const char **argv)
> > "cgroup monitoring only available in system-wide mode");
> >
> > }
> > +
> > + if (rec->opts.comp_level != 0) {
> > + pr_debug("Compression enabled, disabling build id collection at the end of the session\n");
> > + rec->no_buildid = true;
> > + }
> > +
> > if (rec->opts.record_switch_events &&
> > !perf_can_record_switch_events()) {
> > ui__error("kernel does not support recording context switch events\n");
> >
> > ---------------------------------------------------------------------------
> >
> > [acme@quaco perf]$ perf record -z2 sleep 1
> > [ perf record: Woken up 1 times to write data ]
> > [ perf record: Captured and wrote 0.001 MB perf.data, compressed (original 0.001 MB, ratio is 2.292) ]
> > [acme@quaco perf]$ perf record -v -z2 sleep 1
> > Compression enabled, disabling build id collection at the end of the session
> > Using CPUID GenuineIntel-6-8E-A
> > nr_cblocks: 0
> > affinity: SYS
> > mmap flush: 1
> > comp level: 2
> > mmap size 528384B
> > Couldn't start the BPF side band thread:
> > BPF programs starting from now on won't be annotatable
> > perf_event__synthesize_bpf_events: can't get next program: Operation not permitted
> > [ perf record: Woken up 1 times to write data ]
> > [ perf record: Captured and wrote 0.001 MB perf.data, compressed (original 0.001 MB, ratio is 2.305) ]
> > [acme@quaco perf]$
> >
> > Will check if its possible to get rid of the following in this patch, to
> > keep bisection working for this case as well:
> >
> > [acme@quaco perf]$ perf report -D | grep COMPRESS
> > 0x1b8 [0x169]: failed to process type: 81 [Invalid argument]
> > Error:
> > failed to process sample
> > 0 0x1b8 [0x169]: PERF_RECORD_COMPRESSED
> > [acme@quaco perf]$
>
> Makes sense. Thanks.

I did it yesterday, all is in my acme/perf/core branch, now testing it
together with the large pile of patches there accumulated while I was in
LSF/MM + vacations :-)

All have already passed through most of my test build containers, with
most of the distros that have libzstd being updated to include it, and
the make_minimal test build target was updated to build explicitely
disabling zstd, i.e. with NO_LIBZSTD=1, so that we test with/without it
in systems where it is installed and also in systems where zstd is not
even available.

- Arnaldo