On Mon, Jun 01, 2015 at 12:31:54PM +0200, Pavel Březina wrote:
On 05/15/2015 11:30 AM, Jakub Hrozek wrote:
>On Fri, May 15, 2015 at 11:24:29AM +0200, Pavel Březina wrote:
>>On 05/15/2015 11:19 AM, Jakub Hrozek wrote:
>>>On Fri, May 15, 2015 at 11:07:05AM +0200, Pavel Březina wrote:
>>>>On 05/15/2015 10:46 AM, Jakub Hrozek wrote:
>>>>>On Fri, May 15, 2015 at 10:37:06AM +0200, Pavel Březina wrote:
>>>>>>On 05/15/2015 10:17 AM, Jakub Hrozek wrote:
>>>>>>>On Thu, May 14, 2015 at 02:44:33PM +0200, Pavel Březina
wrote:
>>>>>>>>https://fedorahosted.org/sssd/wiki/DesignDocs/DBusCachedObjects
>>>>>>>>
>>>>>>>>Original design is located at:
>>>>>>>>https://fedorahosted.org/sssd/wiki/DesignDocs/DBusResponder#Cachedobjects
>>>>>>>
>>>>>>>Question..
>>>>>>>Why did you choose a separate interface and not a Cache()
method of the user
>>>>>>>object?
>>>>>>
>>>>>>it depends on the side you are looking from. The way in the
design page is
>>>>>>easier, but if we want to put it directly on user and group
object paths, we
>>>>>>can do it for example this way:
>>>>>>
>>>>>>Interface:
>>>>>>org.freedesktop.sssd.infopipe.Cache
>>>>>>
>>>>>>It will be implemented on:
>>>>>>/org/freedesktop/sssd/infopipe/Users
>>>>>>/org/freedesktop/sssd/infopipe/Groups
>>>>>>
>>>>>>Methods will be (or other names):
>>>>>>ao List()
>>>>>>ao ListByDomain(s:domain)
>>>>>>
>>>>>>Interface:
>>>>>>org.freedesktop.sssd.infopipe.Cache.Object
>>>>>>
>>>>>>It will be implemented on:
>>>>>>/org/freedesktop/sssd/infopipe/Users/$uid
>>>>>>/org/freedesktop/sssd/infopipe/Groups/$gid
>>>>>>
>>>>>>Methods will be (or other names):
>>>>>>b Store()
>>>>>>b Remove()
>>>>>>
>>>>>>The advantage is the the caller (and also in sssd) can simply
recognize if
>>>>>>the cached object is a user or a group without the need to parse
the object
>>>>>>path.
>>>>>
>>>>>Right..
>>>>>
>>>>>Implementing the interface by the objects would feel more natural to
me
>>>>>API-wise.
>>>>>
>>>>>Is it a lot more complex to implement? Any other pros or cons?
>>>>
>>>>I have updated the design page.
>>>
>>>Thanks, one last question, do you think naming the methods just
"Store"
>>>and "Retrieve" is descriptive enough? Would "Cache" and
"Uncache" be too
>>>confusing given that the interface is called "Cache" ?
>>
>>Store/Remove as to Store in cache and Remove from cache. Let's compare it:
>>org.freedesktop.sssd.infopipe.Cache.Object.Cache
>>org.freedesktop.sssd.infopipe.Cache.Object.Uncache
>>
>>org.freedesktop.sssd.infopipe.Cache.Object.Store
>>org.freedesktop.sssd.infopipe.Cache.Object.Remove
>>
>>What does sound better? I think using Cache and Uncache is quite redundant.
>
>OK, ACK.
You made some concerns about name of the attribute and whether it should be
indexed or not in other thread.
The attribute is currently named 'cached', my suggestions are:
a) ifp_cached
b) ifp_exported
ifp_cached sounds good to me.
What do you say? Or do you have other suggestions?
About the index... In my understanding the amount of objects marked as
cached is supposed to be very, very small in comparison with the total
amount of users and groups, so we could benefit from the index here.
However, how often will we need to do the lookup? That depends on the use
cases and I guess time will tell us. We can add the index anytime later. So
my call is not add it now.
I think this is the right choice. The index is supposed to be used for
attributes that are used very frequently. I don't think the ifp_cached
attribute will be one of them.