disable pg_num warning?

disable pg_num warning?

[Wyllys Ingersoll]

Is there a way to disable the warning (shown when doing "ceph -s", for
example) about the number of pgs ("too few pgs per osd") ?

Would it be considered risky to disable it?

Consider when a cluster is initially created with just a single pool,
and you don't anticipate that this pool will ever be more than a very
small percentage of the total capacity (1%, for example).  The new
calculator (http://ceph.com/pgcalc) recommends a small pg_num like
128.  However, for a cluster with 144 OSDs, this will trigger the
warning about the pg num being too small.

It seems that the calculation that triggers the warning is a little
bit too simplistic, and while it might be useful in some cases, it is
also a nuisance warning that can be ignored in some environments.  It
would be nice to tell ceph to ignore it and return a healthy status if
everything else is clean.

-Wyllys Ingersoll

Re: disable pg_num warning?

[Sage Weil]

On Fri, 20 Feb 2015, Wyllys Ingersoll wrote:
> Is there a way to disable the warning (shown when doing "ceph -s", for
> example) about the number of pgs ("too few pgs per osd") ?

All of the PG-related warnings are tuanble:

OPTION(mon_pg_warn_min_per_osd, OPT_INT, 30)  // min # pgs per (in) osd before we warn the admin
OPTION(mon_pg_warn_max_per_osd, OPT_INT, 300)  // max # pgs per (in) osd before we warn the admin
OPTION(mon_pg_warn_max_object_skew, OPT_FLOAT, 10.0) // max skew few average in objects per pg
OPTION(mon_pg_warn_min_objects, OPT_INT, 10000)  // do not warn below this object #
OPTION(mon_pg_warn_min_pool_objects, OPT_INT, 1000)  // do not warn on pools below this object #

> Would it be considered risky to disable it?

Go for it.  They are advisory only.  Just keep in mind you're outside of 
the recommended range.

> Consider when a cluster is initially created with just a single pool,
> and you don't anticipate that this pool will ever be more than a very
> small percentage of the total capacity (1%, for example).  The new
> calculator (http://ceph.com/pgcalc) recommends a small pg_num like
> 128.  However, for a cluster with 144 OSDs, this will trigger the
> warning about the pg num being too small.
> 
> It seems that the calculation that triggers the warning is a little
> bit too simplistic, and while it might be useful in some cases, it is
> also a nuisance warning that can be ignored in some environments.  It
> would be nice to tell ceph to ignore it and return a healthy status if
> everything else is clean.

Hmm, it will go away after the other pool(s) are created that will hold 
the bulk of the data, right?  Unfortunately it's hard to make a helpful 
warning that doesn't make noise during the initial cluster setup...

sage

Re: disable pg_num warning?

[Wyllys Ingersoll]

Great, thanks!

On Fri, Feb 20, 2015 at 11:15 AM, Sage Weil <sage@newdream.net> wrote:
> On Fri, 20 Feb 2015, Wyllys Ingersoll wrote:
>> Is there a way to disable the warning (shown when doing "ceph -s", for
>> example) about the number of pgs ("too few pgs per osd") ?
>
> All of the PG-related warnings are tuanble:
>
> OPTION(mon_pg_warn_min_per_osd, OPT_INT, 30)  // min # pgs per (in) osd before we warn the admin
> OPTION(mon_pg_warn_max_per_osd, OPT_INT, 300)  // max # pgs per (in) osd before we warn the admin
> OPTION(mon_pg_warn_max_object_skew, OPT_FLOAT, 10.0) // max skew few average in objects per pg
> OPTION(mon_pg_warn_min_objects, OPT_INT, 10000)  // do not warn below this object #
> OPTION(mon_pg_warn_min_pool_objects, OPT_INT, 1000)  // do not warn on pools below this object #
>
>> Would it be considered risky to disable it?
>
> Go for it.  They are advisory only.  Just keep in mind you're outside of
> the recommended range.
>
>> Consider when a cluster is initially created with just a single pool,
>> and you don't anticipate that this pool will ever be more than a very
>> small percentage of the total capacity (1%, for example).  The new
>> calculator (http://ceph.com/pgcalc) recommends a small pg_num like
>> 128.  However, for a cluster with 144 OSDs, this will trigger the
>> warning about the pg num being too small.
>>
>> It seems that the calculation that triggers the warning is a little
>> bit too simplistic, and while it might be useful in some cases, it is
>> also a nuisance warning that can be ignored in some environments.  It
>> would be nice to tell ceph to ignore it and return a healthy status if
>> everything else is clean.
>
> Hmm, it will go away after the other pool(s) are created that will hold
> the bulk of the data, right?  Unfortunately it's hard to make a helpful
> warning that doesn't make noise during the initial cluster setup...
>
> sage