There are moments in time when you focus on small things and you might miss bigger and more important things.
Working on a solution that is distributed in multiple Azure Regions across the world we had to replicated binary content (files) in all data centers where our solution runs.
The system that push this content is an on-premises system. In this context we decided to use a temporary location where the content is copied first and from where we replicate the content in other regions.
As we can see, the Temp location is used to replicate the content all around the glob. The system is smart enough to balance the load and if multiple replicas are needed in the same Azure Regions, only one copy will be done from the Temp location.
Once the replication is done, the binary content from the Temp location is removed automatically.
Nothing special until this, until in the moment when you ask yourself what is happening if one of the Replicas is corrupted or something happens and you lose data (partially or totally).
In such a case, of course you can go to the master - on-premises system. But when you work with hundreds of terabits, the check and the recovery of the lost replica will take a while.
There is no way for our replicas to fetch the data from an Azure location, even if there was a time when we had this content (in the moment when we copied in the Temp folder).
There is a clear mistake on this flow. We don't have a location that can play the role of the master (internally in our system). The on-premises system is an external system that push data. We should still have in our own system a master for this data.
A solution that is complex and requires costs at the development and maintenance is to track all the replicas and when there is a fail, to be able to provide a source from another replica. In this way, we don't need the extra space for storage. On the other hand the solution is complex and might be more buggy.
In the same time, if we take into consideration that from this replicas, there will be a lot of download activity, the real cost is not from storage, it will be from download.
The most simple solution is to have in the same location as the Temp location a 'Master' for our payloads. This master can be used situations when one of our replica is our of sync. The Master can be secured easily for unexpected situations if we activate geo-replication
Working on a solution that is distributed in multiple Azure Regions across the world we had to replicated binary content (files) in all data centers where our solution runs.
The system that push this content is an on-premises system. In this context we decided to use a temporary location where the content is copied first and from where we replicate the content in other regions.
As we can see, the Temp location is used to replicate the content all around the glob. The system is smart enough to balance the load and if multiple replicas are needed in the same Azure Regions, only one copy will be done from the Temp location.
Once the replication is done, the binary content from the Temp location is removed automatically.
Nothing special until this, until in the moment when you ask yourself what is happening if one of the Replicas is corrupted or something happens and you lose data (partially or totally).
In such a case, of course you can go to the master - on-premises system. But when you work with hundreds of terabits, the check and the recovery of the lost replica will take a while.
There is no way for our replicas to fetch the data from an Azure location, even if there was a time when we had this content (in the moment when we copied in the Temp folder).
There is a clear mistake on this flow. We don't have a location that can play the role of the master (internally in our system). The on-premises system is an external system that push data. We should still have in our own system a master for this data.
A solution that is complex and requires costs at the development and maintenance is to track all the replicas and when there is a fail, to be able to provide a source from another replica. In this way, we don't need the extra space for storage. On the other hand the solution is complex and might be more buggy.
In the same time, if we take into consideration that from this replicas, there will be a lot of download activity, the real cost is not from storage, it will be from download.
The most simple solution is to have in the same location as the Temp location a 'Master' for our payloads. This master can be used situations when one of our replica is our of sync. The Master can be secured easily for unexpected situations if we activate geo-replication
As we can see, in this moment we have a Master for our content that can be used in the case of a out of sync situation with one of our replicas.
Comments
Post a Comment