Both are pretty low risk and unlikely to be seen outside of a PCP QA
environment.
Changes committed to git://oss.sgi.com/kenj/pcp.git dev
src/pmlc/actions.c | 12 +++++++++++-
src/pmlogctl/pmlogger_check.sh | 19 ++++++++++++++-----
2 files changed, 25 insertions(+), 6 deletions(-)
commit eb054ac176ab06bcd978e4b6a3d26e9e2f9c9927
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Wed Oct 28 07:20:28 2009 +1100
pmlogger_check fix small timing window / race condition
There was a race condition that was sometimes exposed by QA test 119
where the attempt to create the lock file would fail because the lock
file already existed, but by the time the status of the lock file was
checked the lock file had been removed. Some convoluted logic
translated this into the confusing messages
Warning: removing lock file older than 30 minutes
ls: cannot access <path>/lock: No such file or directory
Fixed by a guard to test the existence of the lock file after the
pmlock fails.
All of this is safe because it is within a loop that keeps trying to
acquire the lock, and the race condition is most unlikely to be
persistent.
commit 0485576b39c160561c1ae548331be022b098d314
Author: Ken McDonell <kenj@xxxxxxxxxxxxxxxx>
Date: Wed Oct 21 16:42:59 2009 +1100
pmlc sometimes connnects to wrong pmcd
This was causing QA/381 to fail sometimes (depending on the setup
of the remote qa hosts).
After connecting to pmlogger, pmlc may need to connect to the same
pmcd that pmlogger is connected to. When pmlogger is connected to
pmcd on "localhost" and pmlc is run from a different machine, then
pmlc may connect to the wrong pmcd or no pmcd at all, depending on
the state of the host where pmlc is run.
|