ubuntu-phone team mailing list archive

[Colin Ian King <colin.king@xxxxxxxxxxxxx>]

On 05/09/13 10:33, Colin Ian King wrote:
> On 05/09/13 10:23, Evan Dandrea wrote:
>> Stuart Langridge brought up an interesting idea for this on IRC, which
>> I'm copying here rather than having more side discussions:
>>
>> (10:13:20) aquarius: ev, just reading your thread about runaway
>> processes on the touch mailing list (I'm not subscribed to the list,
>> so there's no good way of replying), and I had a thought: it seems to
>> be primarily about *accidental* runaways (I put an infinite loop in my
>> code by mistake), and all the discussion is "how do we know if this
>> process is actually accidental, or if it's deliberately using 100% CPU
>> because it's properly
>> (10:13:20) aquarius: busy doing a lot? The thought is: provide an API
>> call yesIAmReallyBusy which you can call every few seconds while
>> you're busy and then the runaway-killer will know you're not a runaway
>> and will ignore you. This is how screensaver stuff works -- your
>> movie-playing app calls "ignoreTheScreenSaverForAMinute" every minute,
>> and then you don't have to worry about holding locks or anything, and
>> if your app crashes the
>> (10:13:20) aquarius: screensaver doesn't stay disabled.
>> (10:14:22) Evan: aquarius: did you see cking's reply? It seems to be
>> going down this route of "can we tell between accidental runaway and
>> purposeful"
>> (10:15:55) aquarius: ev, I did. The discussion is about being clever
>> around trying to identify from outside the runaway process whether it
>> is runaway or not, which is a useful thing to have if you can do it
>> certainly. What I'm suggesting is making it explicit -- a system
>> service should throw a yesIReallyAmBusy() into its high-CPU processing
>> loop, and then we don't *have* to dwim it; we'll know.
>> (10:20:32) aquarius: ev, I brought it up in case clever people like
>> you and cking might say "we can't do that because $REASONS"; if you
>> think it's worth bringing it up for discussion then I think that's a
>> good idea. Two reasons I can see against it: the first is that if I
>> drop yesIReallyAmBusy() into my main processing loop and that loop has
>> a bug which makes it run infinitely by accident then I'm now immume
>> from being killed, which
>> (10:20:32) aquarius: is bad. Secondly, it suggests that if this comes
>> to the desktop that things like Firefox would need to drop a
>> yesIReallyAmBusy() all over the place, which they ought to do (because
>> they want to be a good citizen on Ubuntu) but probably won't (because
>> being a good citizen on Ubuntu isn't a big enough deal to them).
>>
> 
> A "yesIAmReallyBusy()" API is open to abuse and I guess applications may
> add in thread to do something like:
> 
> 	while (myAppIsRunning) {
> 		yesIAmReallyBusy();
> 		sleep(60);
> 	}
> 
> ..which leads to more wakeups and more battery suckage.
> 

Hrm, I was thinking this would not stop the whole app from being killed,
but I guess it will just stop the thread from being killed, so ignore
that example.

Anyhow, I still think myAppIsRunning() is open to abuse as it will in
time be lavishly added to all sorts of tight loops that suck CPU, which
just adds more overhead on the long run.

I guess a more intelligent way would be a
"myAppRequiresLotsOfCPU(percent, duration)" API that can explicitly
state how much CPU it is expected to use and how long for.  Again, still
open to abuse.

Colin