Hey,

Sorry; In the next release that should be in -h at least. The easiest way
is to just start it with/without the option and diff the output of `stats
settings` between the two.

On Sun, 13 Nov 2016, Bill Moseley wrote:

> ok, I found the "modern" option by looking at the code.
> In a few simple tests I can see slab memory is reclaimed.   From what I see:
>  *  Memcached will reclaim memory in the slab, but will leave two allocated 
> pages in the slab.
>
>  *  That freed memory can then be used for pages in other slabs.
>
>  *  If all items in the slab are reclaimed then "stats items" no longer shows 
> that slab. I wonder if that is correct because we can no longer see stats on 
> evicted items for
>     that slab.

It's always been that way... just never fixed/changed it I guess?

>  *  If more memory than specified with the -m command line argument was 
> malloced, then memory seems be be given back to the system after a reclaim 
> (down to -m setting).

There's a "cache_memlimit" command that allows you to increase or decrease
the memory limit at runtime. that's how the command works.

>  *  Is there a good way to see how much memory has been reclaimed (and thus 
> available to other slabs)?

| slab_global_page_pool | 32u     | Slab pages returned to global pool for
|
|                       |         | reassignment to other slab classes.
|

>     Is the best way to subtract "bytes" from "total_malloced"?   i.e. to 
> check how much "free" memory there is available for the cache.
> Does the reclaim algorithm move items within the slab?  That is, will it 
> reshuffle items from one page to another page so that the page can be 
> reclaimed?  Or do all items in a
> given page first need to expire before it can be reclaimed?

See above.

To answer your earlier questions: -I 4m should work a lot better with the
new code. It doesn't screw up the smaller slab classes.

>
> On Fri, Nov 11, 2016 at 10:44 PM, Bill Moseley <[email protected]> wrote:
>
>
>       On Thu, Oct 13, 2016 at 11:32 PM, dormando <[email protected]> wrote:
>             I think this is fixed in 1.4.32. Been broken for a very long time
>             unfortunately.
>
>
> Finally getting back to this.  I'm testing with 1.4.33.  Looks good, thanks.
>
>
> I'm trying to understand all the new options.  What does -o modern include?
>
> - modern: Enables 'modern' defaults. See release notes (highly recommended!).
>
> Which release notes should I be looking at?
>
> Or rather, can you provide guidance on what options to use?   We are using 
> Memcached as a general cache so our value sizes and set frequency varies 
> quite a bit.  I
> suspect being able to reallocate slab space would be helpful in our case.
>
> We also have a need to store some larger (>1MB) items, too. (for example, -I 
> 4m)
>
> Thanks,
>
>  
>
>       On Sun, 25 Sep 2016, Bill Moseley wrote:
>
>       > If I understand the documentation correctly, evicted_time should show 
> the seconds since the last access for the more recently evicted item.
>       >
>       > I'm seeing evicted_time for our production servers, but I'm not able 
> to have a value show up when testing.
>       >
>       > I have a slab class that can hold 10 items of the value I'm writing.
>       >
>       > I assume I would see an evicted_time doing this:
>       >  1. write a key and read it back
>       >  2. sleep 5 seconds.
>       >  3. write and read 10 random keys into the slab (quicly)
>       >  4. Notice that there has now been one eviction
>       >  5. read from the original key and get NOT FOUND
>       > I then see:
>       >
>       >       STAT items:32:evicted 1
>       > STAT items:32:evicted_nonzero 0
>       > STAT items:32:evicted_time 0
>       >
>       >
>       > Shouldn't I see evicted_time of 5 seconds since the item evicted was 
> last accessed about 5 seconds before it got evicted?
>       >
>       > I get the same result if I use no expires times or If I set it to, 
> say, 30 seconds.
>       >
>       > VERSION 1.4.25
>       >
>       >
>       > --
>       > Bill Moseley
>       > [email protected]
>       >
> > --
> >
> > ---
> > You received this message because you are subscribed to the Google Groups 
> > "memcached" group.
> > To unsubscribe from this group and stop receiving emails from it, send an 
> > email to [email protected].
> > For more options, visit https://groups.google.com/d/optout.
> >
> >
>
> --
>
> ---
> You received this message because you are subscribed to the Google Groups 
> "memcached" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to [email protected].
> For more options, visit https://groups.google.com/d/optout.
>
>
>
>
> --
> Bill Moseley
> [email protected]
>
>
>
>
> --
> Bill Moseley
> [email protected]
>
> --
>
> ---
> You received this message because you are subscribed to the Google Groups 
> "memcached" group.
> To unsubscribe from this group and stop receiving emails from it, send an 
> email to [email protected].
> For more options, visit https://groups.google.com/d/optout.
>
>

-- 

--- 
You received this message because you are subscribed to the Google Groups 
"memcached" group.
To unsubscribe from this group and stop receiving emails from it, send an email 
to [email protected].
For more options, visit https://groups.google.com/d/optout.

Reply via email to