Open mcomella opened 3 years ago
Triage: csadilek feels this could be explained by the device load and isn't too important to investigate so we put it at the bottom of the backlog.
After discussing about this, we've come to the conclusion that testing this in a setting like FNPRMS might lead to some answers that are skewed. We've thought that using telemetry to get a full picture on how it affects our users might be a better solution since reproducing this with our own device could be difficult (i.e: leaving it on, waiting 15 days ,etc...)
In FNPRMS, we noticed a trend where if we left the devices on without restarting them, the fenix cold start up time would trend up over the week and go back down after we restarted the devices:
We're curious to know 1) if this is something users experience and 2) is fenix doing something to cause this and is there something we can change to prevent it? Or is this just an artifact of the testing infrastructure (FNPRMS) or the load we put on the device (100's of daily test iterations between FNPRMS and browsertime).
One thing to note is that bdekoz has continued to restart the devices but we no longer see such peaky behavior – it's unclear why. It's also been hard to interpret the results recently (since mid-Jan) due to regressions and device outages.