So I think that there is a race---a losing race, in fact.

Both metrics_daemon.conf and crash-reporter.conf start on "starting system-services".  Then crash-boot-collect.conf starts on "stopped crash-reporter".

crash-reporter.conf runs crash_reporter with the --init flag.  crash-boot-collect.conf runs crash_reporter with the --boot_collect flag.  This execution creates the kKernelCrashDetected file in /run/kernel-crash-detected, whose existence is checked by the metrics daemon when it starts.  But I think that by the time the file is created, the metrics daemon has already looked for it (and not found it).

I'll figure out the best way of fixing it.

The following revision refers to this bug:
  https://chromium.googlesource.com/chromiumos/platform2/+/8c2212674f27afa0417ceee813ad26cd259c658c

commit 8c2212674f27afa0417ceee813ad26cd259c658c
Author: Luigi Semenzato <semenzato@chromium.org>
Date: Mon Jul 31 02:16:07 2017

metrics: postpone start of daemon

Previously the metrics daemon started at the same time as
the crash reporter, but its correct execution depends on
some state that the crash reporter was in a race to create.
Specifically, the crash reporter detects if the previous
session ended in a kernel panic and creates a file in
/run whose existence is checked by the metrics daemon.

BUG= chromium:749291 
TEST=verified that bug is present before, and gone after the change

Change-Id: I9b7d7930e8f146b38b5f8aeb9ef9e25c773cc6c7
Reviewed-on: https://chromium-review.googlesource.com/590929
Commit-Ready: Luigi Semenzato <semenzato@chromium.org>
Tested-by: Luigi Semenzato <semenzato@chromium.org>
Reviewed-by: Grant Grundler <grundler@chromium.org>
Reviewed-by: Mike Frysinger <vapier@chromium.org>

[modify] https://crrev.com/8c2212674f27afa0417ceee813ad26cd259c658c/metrics/init/metrics_daemon.conf

NOTE: this affected a number of other crash-related UMA metrics (see metrics_daemon.cc for details).

Believed fixed.