Print

Print


Hi Andy,

This doesn't seem to be working for our systems?

I included:

cms.space min 15t 15t

in the configuration, and restarted the dataservers yesterday.  I checked one
of the data servers and saw that files were written today on the system.

I take the config line to mean that there must be at least 15TB available before
the dataserver will write.  Do the two values have to be different?

Patrick


On 06/22/2010 11:03 PM, Andrew Hanushevsky wrote:
> Hi Patrick,
>
> Yes, the analysis is correct. You can set it up to allow deletions but
> disallow creation of new files via the redirector.
>
> Andy
>
> ----- Original Message ----- From: "Yang, Wei" <[log in to unmask]>
> To: "Patrick McGuigan" <[log in to unmask]>
> Cc: "xrootd-l" <[log in to unmask]>
> Sent: Tuesday, June 22, 2010 5:07 PM
> Subject: Re: Data Migration issues
>
>
>> I think you only need to change the configuration for the data server.
>> Redirector gets space info from cmsd on data servers. With cms.space,
>> you can still _write_ to the data server if you access it directly.
>> But if you ask the redirector to choose a data server for you for
>> writing, the redirector won't choose this particular data server.
>>
>> regards,
>> Wei Yang | [log in to unmask] | 650-926-3338(O)
>>
>>
>> On Jun 22, 2010, at 4:59 PM, Patrick McGuigan wrote:
>>
>>> Thanks Wei,
>>>
>>> I had thought that something like this method would be available, does
>>> your suggestion require a specific configuration in the redirector? We
>>> are using the defaults with respect to how the redirector chooses
>>> dataservers.
>>>
>>> Regards,
>>> Patrick
>>>
>>> On 06/22/2010 06:54 PM, Yang, Wei wrote:
>>>> Hi Patrick,
>>>>
>>>> If you set all.export path to readonly, you won't be able to delete.
>>>> I never tried any of this because but here is an idea: You may use
>>>> cms.space directive to set the minimum space required for the data
>>>> server to be writeable, and set a high number (above the total space
>>>> of that data server) so that redirector always think the data server
>>>> runs out of space. In that case, you can keep on deleting without
>>>> worrying about new data getting written to it.
>>>>
>>>> regards,
>>>> Wei Yang | [log in to unmask] | 650-926-3338(O)
>>>>
>>>>
>>>> On Jun 22, 2010, at 4:20 PM, Patrick McGuigan wrote:
>>>>
>>>>> Hi,
>>>>>
>>>>>
>>>>> I am preparing to replace 6 smaller dataservers with 2 much larger
>>>>> systems and I am curious if anyone has advice on making the
>>>>> transition.
>>>>> There is about 40TB of data currently on the smaller dataservers.
>>>>>
>>>>> A couple of caveats:
>>>>> 1) The existing data is organized in cache groups on the
>>>>> dataservers and
>>>>> the cgroup information will need to propagate to the new dataservers
>>>>>
>>>>> 2) The smaller dataservers are in a production system and I want to
>>>>> minimize any downtime associated with data on these servers.
>>>>>
>>>>> 3) A sizable portion of the 40TB is no longer relevant to the
>>>>> production
>>>>> system and potentially could be deleted (or at least not migrated)
>>>>>
>>>>>
>>>>> It eases my burden if I can delete data from all the production
>>>>> dataservers at the same time. However, if I delete data from the 6
>>>>> servers that will be migrated, I don't want new data to take its
>>>>> place.
>>>>>
>>>>>
>>>>> I was hoping that I could somehow setup the existing dataservers in a
>>>>> readonly configuration but still be able to delete data via the
>>>>> redirector (actually from a FUSE client talking to a CNS service on
>>>>> the
>>>>> redirector). Is this possible?
>>>>>
>>>>>
>>>>> From my reading, I can make the dataservers readonly by altering the
>>>>> all.export line:
>>>>>
>>>>> all.export /xrd readonly
>>>>>
>>>>> But I am not sure if this still allows deletions.
>>>>>
>>>>> Also, does the use of oss.cache directives affect the readonly status?
>>>>>
>>>>>
>>>>> If anyone has experience with performing this type of migration, or if
>>>>> anyone has suggestions, I would be very thankful to hear from you.
>>>>>
>>>>> Regards,
>>>>>
>>>>> Patrick
>>>>>
>>>>>
>>>>
>>
>>