Grr, I think I messed up the threshold detection ("> progress_goal") on the
denominator.
Take 2:
expr: |
(*(progress_goal - progress_current) > 0)* /
(( predict_linear(*progress_current*{fstype!~"fuse.*|nfs.*"}[12h],
604800) *> progress_goal*) - *progress_current*) * 604800
On Thursday, 18 May 2023 at 08:57:57 UTC+1 Brian Candler wrote:
> > I have seen a bunch of examples for predicting when a disk will fill,
> but these all seem to rely on the assumption that a gauge is trending
> downwards, and we are predicting when it meets zero.
>
> Here is a better recipe as a starting point, using predict_linear():
> https://groups.google.com/g/prometheus-users/c/PCT4MJjFFgI/m/kVfOW069BQAJ
>
> > How can we write an expression that will tell us the time in seconds
> until progress_current meets *progress_goal*?
>
> In principle, I think you should just be able to replace the metric in
> that expression with (progress_goal - progress_current) and see when that
> hits zero. Something like this:
>
> expr: |
> (*(progress_goal - progress_current) > 0)* / (*(progress_goal -
> progress_current)* -
> ((*progress_goal -*
> predict_linear(*progress_current*{fstype!~"fuse.*|nfs.*"}[12h],
> 604800)) < 0)) * 604800
>
> (That's syntactically valid, but otherwise untested. Obviously you should
> remove or change the label selectors on progress_current to suit the metric
> in question).
>
> I see +progress_goal and -progress_goal in the denominator so it should be
> possible to simplify it further, I think to this:
>
> expr: |
> (*(progress_goal - progress_current) > 0)* /
> (( predict_linear(*progress_current*{fstype!~"fuse.*|nfs.*"}[12h],
> 604800) *- progress_current*) > *progress_goal*) * 604800
>
> The value of this expression is intended to return *the time until the
> goal is reached*, using linear interpolation over the past 12 hours, to
> predict where it will be in 7 days. If the goal isn't expected to be
> reached in 7 days then it will return no value.
>
> I can attempt to justify that expression graphically:
>
> | C2 ^
> | / |
> | / |
> G +.^......x.....|....
> | | / D
> | N / |
> | | / |
> | v C1 v
> |
> +---0d------7d----> time
>
> G is the progress_goal, which I'm assuming is constant
> C1 is the value of progress_current now (time = 0d), which is less than G
> C2 is the future predicted value of progress_current (at time = 7d), which
> is greater than G
>
> N (numerator) is G - C1
> D (denominator) is C2 - C1
>
> And geometrically, I think the ratio of the crossover time (0d...x) to the
> full time (0d...7d) is the same as the ratio of N to D. Hence (N/D)*7d
> gives the time to crossover.
>
> Give it a go. Note that if progress_goal is a metric, rather than a
> constant, then it will need to have exactly the same label set as
> progress_current; or you will need to add some on(...) or ignoring(...)
> clauses as appropriate.
>
--
You received this message because you are subscribed to the Google Groups
"Prometheus Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to prometheus-users+unsubscr...@googlegroups.com.
To view this discussion on the web visit
https://groups.google.com/d/msgid/prometheus-users/bcb15414-2ee4-4355-8302-b89913c2a22en%40googlegroups.com.
