[infinispan-dev] Protobuf metadata cache and x-site

Tristan Tarrant ttarrant at redhat.com
Thu Apr 12 16:15:41 EDT 2018


We also need: backup priority for internal caches as well as conflict 
resolution for backups to avoid broken data replicating in the wrong 
direction.

Tristan

On 4/12/18 10:13 PM, Tristan Tarrant wrote:
> I think we can certainly make it additive, especially now that we have 
> configuration templates in place: the user supplies a base template, and 
> the internal cache logic override with what is needed so that broken 
> configs are less probable (but still possible). Alternatively, instead 
> of overriding, we just check that it matches the requirements.
> 
> Tristan
> 
> On 4/12/18 10:10 PM, Adrian Nistor wrote:
>> Backing up caches with protobuf payload to a remote site will not work 
>> if they are indexed, unless the remote site already has the schema for 
>> the types in question, or else indexing will fail. If the cache is not 
>> indexed it matters less.
>>
>> So the replication of protobuf metadata cache has to be arranged 
>> somehow before any other data is replicated. Manual replication is 
>> indeed PITA.
>>
>> I remember in the very early version of remote query the protobuf 
>> metadata cache configuration was created programatically on startup 
>> unless a manually defined configuration with that name was found, 
>> already provided by the user. In that case the user's config was used. 
>> This approach had the benefit of allowing the user to gain control if 
>> needed. But can also lead to gloom and doom. Was that too bad to do it 
>> again :)))?
>>
>> Adrian
>>
>> On 04/12/2018 10:27 PM, Tristan Tarrant wrote:
>>> It is definitely an internal cache. Because of this, automatically
>>> backing it up to a remote site might not be such a good idea.
>>>
>>> Backups are enabled per-cache, and therefore just blindly replicating
>>> the schema cache to the other site is not a good idea.
>>>
>>> I think that we need a cache-manager-level backup setting that does the
>>> right thing.
>>>
>>> Tristan
>>>
>>> On 4/12/18 7:01 PM, Pedro Ruivo wrote:
>>>> Wouldn't be better to assume the protobuf cache doesn't fit the 
>>>> internal
>>>> cache use case? :)
>>>>
>>>> On 12-04-2018 17:21, Galder Zamarreno wrote:
>>>>> Ok, we do need to find a better way to deal with this.
>>>>>
>>>>> JIRA: https://issues.jboss.org/browse/ISPN-9074
>>>>>
>>>>> On Thu, Apr 12, 2018 at 5:56 PM Pedro Ruivo <pedro at infinispan.org
>>>>> <mailto:pedro at infinispan.org>> wrote:
>>>>>
>>>>>
>>>>>
>>>>>       On 12-04-2018 15:49, Galder Zamarreno wrote:
>>>>>        > Hi,
>>>>>        >
>>>>>        > We have an issue with protobuf metadata cache.
>>>>>        >
>>>>>        > If you run in a multi-site scenario, protobuf metadata
>>>>>       information does
>>>>>        > not travel across sites by default.
>>>>>        >
>>>>>        > Being an internal cache, is it possible to somehow
>>>>>       override/reconfigure
>>>>>        > it so that cross-site configuration can be added in 
>>>>> standalone.xml?
>>>>>
>>>>>       No :( since it is an internal cache, its configuration can't 
>>>>> be changed.
>>>>>
>>>>>        >
>>>>>        > We're currently running a periodic job that checks if the
>>>>>       metadata is
>>>>>        > present and if not present add it. So, we have a 
>>>>> workaround for
>>>>>       it, but
>>>>>        > it'd be not very user friendly for end users.
>>>>>        >
>>>>>        > Thoughts?
>>>>>
>>>>>       Unfortunately none... it is the first time an internal cache 
>>>>> needs
>>>>>       to do
>>>>>       some x-site.
>>>>>
>>>>>        >
>>>>>        > Cheers,
>>>>>        > Galder
>>>>>        >
>>>>>        >
>>>>>        > _______________________________________________
>>>>>        > infinispan-dev mailing list
>>>>>        > infinispan-dev at lists.jboss.org
>>>>>       <mailto:infinispan-dev at lists.jboss.org>
>>>>>        > https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>>>>        >
>>>>>       _______________________________________________
>>>>>       infinispan-dev mailing list
>>>>>       infinispan-dev at lists.jboss.org 
>>>>> <mailto:infinispan-dev at lists.jboss.org>
>>>>>       https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> infinispan-dev mailing list
>>>>> infinispan-dev at lists.jboss.org
>>>>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>>>>
>>>> _______________________________________________
>>>> infinispan-dev mailing list
>>>> infinispan-dev at lists.jboss.org
>>>> https://lists.jboss.org/mailman/listinfo/infinispan-dev
>>>>
>>
> 

-- 
Tristan Tarrant
Infinispan Lead and Data Grid Architect
JBoss, a division of Red Hat


More information about the infinispan-dev mailing list