Optimize service restart logic.
If a non-persistent service is being restarted and it was killed due to
low memory previously, instead of restarting the service in one second,
we now restart it based on its last known oom_adj value. The backoff
logic of consistent restarts remains the same.
Additionally, when the service is due to restart, if the memory pressure
is still considered to be critical, the new logic will further delay the
service restart.
Initial test results: # of restarts dropped by 40-50% on average with
the optimized logic when the memory pressure was gradually increased.
Tested manually via memeater. Increased the memory pressure by 500 mb
at a time via the following command: adb shell memeater 500
On a 4gb device, no significant restarts were noticed when memory
pressure was increased by 2gb in total. When increased by 3gb in total,
services started to restart and the above results were observed via
the logged output over a 5 min period.
Bug: 79848281
Test: manually (steps listed above)
Change-Id: Idcd1300882321c6033e569736150a8bf7889678b
4 files changed