Changes committed to git://oss.sgi.com/pcp/pcp.git dev
man/man1/pmlogger.1 | 11 +++++++++--
man/man1/pmmgr.1 | 3 ++-
qa/749 | 15 +++++++++++++--
src/pmlogger/pmnewlog.sh | 8 ++++----
src/pmlogger/src/pmlogger.c | 22 +++++++++++++++-------
src/pmmgr/TODO | 3 ++-
src/pmmgr/pmmgr.cxx | 36 ++++++++++++++++++++++++++++++------
src/pmmgr/pmmgr.h | 1 +
8 files changed, 76 insertions(+), 23 deletions(-)
commit c3ef04e0207470b615f1891395e0f643b78d8480
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Wed Feb 26 15:59:00 2014 +1100
Fix pmnewlog getopts string after pmlogger -y/-z switch
commit 2f6780aded0d348adf5b055e94d04b33e5a9eecd
Author: Frank Ch. Eigler <fche@xxxxxxxxxx>
Date: Wed Feb 26 15:52:49 2014 +1100
pmmgr: avoid super-quick daemon failure-retry loops
The '-p N' parameter is also interpreted to limit the rate of retries
for any particular daemon's preparation / startup. Without a limit,
pmmgr's rapid response to signals / failure has led to tight
failure/retry loops, resulting in huge log files and no merriment.
commit cfe5ccb345a6a0f43e4b1a0d6d3ea4845b24ab85
Author: Frank Ch. Eigler <fche@xxxxxxxxxx>
Date: Fri Feb 21 10:19:51 2014 -0500
pmmgr signal handling: setpgrp
Shutting down an entire pmmgr family of processes is tricky, because
we use process-group-based signalling to forward SIGTERM to everything
we started. But in some cases, pmmgr may not be in a process group of
its own, such as if it is started by init scripts, instead of by hand
from a shell running on a pty. To make both situations similar, we
now run a setpgrp() near initialization, to be sure that pmmgr's own
pid can be used (negatively) as a process-group.
pmlogger has had a hardcoded conflicting setpgrp inside it, for its
normal use as a daemon. However, when invoked by hand, or via pmmgr,
it now considers itself in the "foreground", and does not setpgrp away
from the inherited one.
commit a91aa864dfa81f6904b9249cd567ad94e7ff0043
Author: Frank Ch. Eigler <fche@xxxxxxxxxx>
Date: Wed Feb 19 17:21:02 2014 -0500
pmlogger: -z -> -y
Nathan noticed that PCPintro(1)'s general reference to -z is the
opposite polarity ("use remote time zone") to pmlogger's new -z
("use local time zone"). So why not rename pmlogger's to -y.
commit d045bcd12cd738e368271e56a78998cbf66afd5e
Author: Frank Ch. Eigler <fche@xxxxxxxxxx>
Date: Wed Feb 19 16:23:13 2014 -0500
pmlogger: add -z option for host-local timezone support
It is difficult for a pmlogger user to specify an absolute time at
which she'd like the process to shut down. That's because the
pmlogger -T option is interpreted with the remote PMCD's time zone in
mind, which could be an a-priori unknown. This patch adds a -z
option, which lets pmlogger interpret -T / etc. in the originating
local time zone instead.
pmmgr's pmlogmerge-granular mode is updated to use this.
commit 9681ff309d09de29fd92942f89b4405164b4d3f2
Author: Nathan Scott <nathans@xxxxxxxxxx>
Date: Tue Feb 25 10:51:48 2014 +1100
Fix qa/749 when systemtap userspace probes are unavailable
Tweak test 749 using Franks suggestion of verifying up to and
including the compilation phase (-p4) before going ahead with
the complete test. This is to catch missing userspace probe
support on some platforms (s390x tripped this in the Red Hat
QE environment).
|