objecter incremental osdmap crc check

objecter incremental osdmap crc check

[Xinze Chi (信泽)]

hi, all:

      When osd update the osdmap, it would apply the
OSDMap::Incremental and get the crc for the new osdmap. If the crc is
not the same as define in Incremental osd would request the full
osdmap. I think the logic is quite well.

     But in the client, the objecter do not check the crc. I do not
know why not client do the same logic as osd.

    Could anyone explain this? Thanks.

-- 
Regards,
Xinze Chi

Re: objecter incremental osdmap crc check

[Gregory Farnum]

On Wed, Feb 3, 2016 at 5:54 AM, Xinze Chi (信泽) <xmdxcxz@gmail.com> wrote:
> hi, all:
>
>       When osd update the osdmap, it would apply the
> OSDMap::Incremental and get the crc for the new osdmap. If the crc is
> not the same as define in Incremental osd would request the full
> osdmap. I think the logic is quite well.
>
>      But in the client, the objecter do not check the crc. I do not
> know why not client do the same logic as osd.
>
>     Could anyone explain this? Thanks.

Two things:
1) The client doesn't necessarily have the full OSDMap that the OSDs
do, so the CRC won't match. (You can see this in the encoding family
of functions; there are sections which are server-only.) I suppose the
monitors could include a client-only CRC as well...
2) Nobody bothered since it seemed less important for "ephemeral"
clients versus OSDs that persist their maps and share them with peers.
-Greg
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: objecter incremental osdmap crc check

[Xinze Chi (信泽)]

I wonder when the client receive OSDMap::Incremental and
apply_incremental, why not compare the OSDMap.crc with inc.full_crc,
like this https://github.com/ceph/ceph/blob/master/src/osd/OSD.cc#L6456

The OSDMonitor also check the crc when update OSMap, like this
https://github.com/ceph/ceph/blob/master/src/mon/OSDMonitor.cc#L247

2016-02-04 1:00 GMT+08:00 Gregory Farnum <gfarnum@redhat.com>:
> On Wed, Feb 3, 2016 at 5:54 AM, Xinze Chi (信泽) <xmdxcxz@gmail.com> wrote:
>> hi, all:
>>
>>       When osd update the osdmap, it would apply the
>> OSDMap::Incremental and get the crc for the new osdmap. If the crc is
>> not the same as define in Incremental osd would request the full
>> osdmap. I think the logic is quite well.
>>
>>      But in the client, the objecter do not check the crc. I do not
>> know why not client do the same logic as osd.
>>
>>     Could anyone explain this? Thanks.
>
> Two things:
> 1) The client doesn't necessarily have the full OSDMap that the OSDs
> do, so the CRC won't match. (You can see this in the encoding family
> of functions; there are sections which are server-only.) I suppose the
> monitors could include a client-only CRC as well...
> 2) Nobody bothered since it seemed less important for "ephemeral"
> clients versus OSDs that persist their maps and share them with peers.
> -Greg



-- 
Regards,
Xinze Chi
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Re: objecter incremental osdmap crc check

[Gregory Farnum]

On Thu, Feb 4, 2016 at 3:25 AM, Xinze Chi (信泽) <xmdxcxz@gmail.com> wrote:
> I wonder when the client receive OSDMap::Incremental and
> apply_incremental, why not compare the OSDMap.crc with inc.full_crc,
> like this https://github.com/ceph/ceph/blob/master/src/osd/OSD.cc#L6456
>
> The OSDMonitor also check the crc when update OSMap, like this
> https://github.com/ceph/ceph/blob/master/src/mon/OSDMonitor.cc#L247

The kernel client literally doesn't have the code to apply all of the
changes and check the CRCs. If we don't do it on the userspace client
I guess it was just laziness, but again they might not even get all of
the data required to match CRCs; I forget. (For instance, if the
monitors are generating an OSDMap with new features the clients don't
understand and don't need to, the client won't be able to decode those
parts of the OSDMap correctly, and will get a different crc value on
it.)
-Greg

>
> 2016-02-04 1:00 GMT+08:00 Gregory Farnum <gfarnum@redhat.com>:
>> On Wed, Feb 3, 2016 at 5:54 AM, Xinze Chi (信泽) <xmdxcxz@gmail.com> wrote:
>>> hi, all:
>>>
>>>       When osd update the osdmap, it would apply the
>>> OSDMap::Incremental and get the crc for the new osdmap. If the crc is
>>> not the same as define in Incremental osd would request the full
>>> osdmap. I think the logic is quite well.
>>>
>>>      But in the client, the objecter do not check the crc. I do not
>>> know why not client do the same logic as osd.
>>>
>>>     Could anyone explain this? Thanks.
>>
>> Two things:
>> 1) The client doesn't necessarily have the full OSDMap that the OSDs
>> do, so the CRC won't match. (You can see this in the encoding family
>> of functions; there are sections which are server-only.) I suppose the
>> monitors could include a client-only CRC as well...
>> 2) Nobody bothered since it seemed less important for "ephemeral"
>> clients versus OSDs that persist their maps and share them with peers.
>> -Greg
>
>
>
> --
> Regards,
> Xinze Chi
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html