Measure and send update duration (and corresponding uptime) to UMA
This patch introduces two new metrics, Installer.UpdateDuration and
Installer.UpdateDurationUptime. The former is the timespan from when
the update was first discovered until it has been downloaded and
applied (including the time the device is suspended or powered
off). The latter is similar, but without taking into account time
spent in suspend or powered off.
For example, if the device is suspended (or powered off) for N seconds
while updating, the Installer.UpdateDuration metric will be N seconds
bigger than Installer.UpdateDurationUptime metric:
Histogram: Installer.UpdateDuration recorded 1 samples, average = 313.0
Histogram: Installer.UpdateDurationUptime recorded 1 samples, average = 251.0
Also remove the existing Installer.UpdateTime metric as this didn't
take process restarts into account and is now superseeded by the
Installer.UpdateDuration metric.
This is done by using the CLOCK_MONOTONIC_RAW clock (available in
Linux 2.6.28 and later) since this clock indeed does not advance when
the system is sleeping.
We use the PayloadState class to persist recorded data across
update_engine process restart (including device reboots).
Since clock_gettime(2) and CLOCK_MONOTONIC_RAW requires linking to the
librt library do this and also request the system header files to
expose the required symbols and defines, i.e. define _POSIX_C_SOURCE
>= 199309L.
Also remove _POSIX_C_SOURCE mangling from update_attempter.cc since
it's actually not needed there and generally it's better to make the
environment the same across all translation units (by putting whatever
is needed in e.g. CCFLAGS).
BUG=chromium:226763
TEST=unit tests, force update, examine chrome://histograms
Change-Id: I883668564b5fa78ff3e19156bd77496ff929ca58
Signed-off-by: David Zeuthen <zeuthen@chromium.org>
Reviewed-on: https://gerrit.chromium.org/gerrit/47928
Reviewed-by: Jay Srinivasan <jaysri@chromium.org>
11 files changed