Hi good morning

As far as I understood, we have automatic database replication with PAaaS, but there are not any load balance between these replicas so far. Are there plans for changing that, some enhancement planned? When we create a database replica, all the tasks to balance load are up to users / customer tailor made?

Regards

Márcio

Hi

While i read the documentation I realized that all the work is made by OpenShift, I mean, it is not possible to actively decide which replica will get the job or not, because necessarily there is a leader which deals with all the writes and synchronization.

But what about MDX queries? Are them at least run in parallel or balanced between replicas? Is there any round-robin taking place between replicas in the case of read-only quests?

Regards

Márcio

Hi good morning

As far as I understood, we have automatic database replication with PAaaS, but there are not any load balance between these replicas so far. Are there plans for changing that, some enhancement planned? When we create a database replica, all the tasks to balance load are up to users / customer tailor made?

Regards

Márcio

Hi Marcio,

In TM1 it is actually the daemon service that decides which request goes to which replica of a database. And you are right, any request not deemed idempotent in essence goes to the leader replica, all other requests, like executing MDX queries, already get spread across all replicas that are up to date at the time of the request. This means that in a system with a lot of updates all the time more load ends up on the leader whereas systems with barely any updates, if at all, request load gets spread pretty evenly.

Cheers,

-H

Hi

While i read the documentation I realized that all the work is made by OpenShift, I mean, it is not possible to actively decide which replica will get the job or not, because necessarily there is a leader which deals with all the writes and synchronization.

But what about MDX queries? Are them at least run in parallel or balanced between replicas? Is there any round-robin taking place between replicas in the case of read-only quests?

Regards

Márcio

Hi good morning

As far as I understood, we have automatic database replication with PAaaS, but there are not any load balance between these replicas so far. Are there plans for changing that, some enhancement planned? When we create a database replica, all the tasks to balance load are up to users / customer tailor made?

Regards

Márcio

Hi Marcio,

In TM1 it is actually the daemon service that decides which request goes to which replica of a database. And you are right, any request not deemed idempotent in essence goes to the leader replica, all other requests, like executing MDX queries, already get spread across all replicas that are up to date at the time of the request. This means that in a system with a lot of updates all the time more load ends up on the leader whereas systems with barely any updates, if at all, request load gets spread pretty evenly.

Cheers,

-H

Hi

While i read the documentation I realized that all the work is made by OpenShift, I mean, it is not possible to actively decide which replica will get the job or not, because necessarily there is a leader which deals with all the writes and synchronization.

But what about MDX queries? Are them at least run in parallel or balanced between replicas? Is there any round-robin taking place between replicas in the case of read-only quests?

Regards

Márcio

Hi good morning

As far as I understood, we have automatic database replication with PAaaS, but there are not any load balance between these replicas so far. Are there plans for changing that, some enhancement planned? When we create a database replica, all the tasks to balance load are up to users / customer tailor made?

Regards

Márcio

A final question here - rules and feeders processing can be seen as idempotent, as its final results are the same no matter what replica is processing them, right?

Hi Marcio,

In TM1 it is actually the daemon service that decides which request goes to which replica of a database. And you are right, any request not deemed idempotent in essence goes to the leader replica, all other requests, like executing MDX queries, already get spread across all replicas that are up to date at the time of the request. This means that in a system with a lot of updates all the time more load ends up on the leader whereas systems with barely any updates, if at all, request load gets spread pretty evenly.

Cheers,

-H

Hi

While i read the documentation I realized that all the work is made by OpenShift, I mean, it is not possible to actively decide which replica will get the job or not, because necessarily there is a leader which deals with all the writes and synchronization.

But what about MDX queries? Are them at least run in parallel or balanced between replicas? Is there any round-robin taking place between replicas in the case of read-only quests?

Regards

Márcio

Hi good morning

As far as I understood, we have automatic database replication with PAaaS, but there are not any load balance between these replicas so far. Are there plans for changing that, some enhancement planned? When we create a database replica, all the tasks to balance load are up to users / customer tailor made?

Regards

Márcio

Rules are only processed during query time and as such only influence 'snapshots' of the data, feeders on the other hand are executed as part of update processes but in the case of a v12 replica those end up being executed as updates are made, either on the leader as a result of an update or on a follower as a result of applying the changes made on the leader first.

A final question here - rules and feeders processing can be seen as idempotent, as its final results are the same no matter what replica is processing them, right?

Hi Marcio,

In TM1 it is actually the daemon service that decides which request goes to which replica of a database. And you are right, any request not deemed idempotent in essence goes to the leader replica, all other requests, like executing MDX queries, already get spread across all replicas that are up to date at the time of the request. This means that in a system with a lot of updates all the time more load ends up on the leader whereas systems with barely any updates, if at all, request load gets spread pretty evenly.

Cheers,

-H

Hi

While i read the documentation I realized that all the work is made by OpenShift, I mean, it is not possible to actively decide which replica will get the job or not, because necessarily there is a leader which deals with all the writes and synchronization.

But what about MDX queries? Are them at least run in parallel or balanced between replicas? Is there any round-robin taking place between replicas in the case of read-only quests?

Regards

Márcio

Hi good morning

As far as I understood, we have automatic database replication with PAaaS, but there are not any load balance between these replicas so far. Are there plans for changing that, some enhancement planned? When we create a database replica, all the tasks to balance load are up to users / customer tailor made?

Regards

Márcio

Thanks, understood.

Regards

Rules are only processed during query time and as such only influence 'snapshots' of the data, feeders on the other hand are executed as part of update processes but in the case of a v12 replica those end up being executed as updates are made, either on the leader as a result of an update or on a follower as a result of applying the changes made on the leader first.

A final question here - rules and feeders processing can be seen as idempotent, as its final results are the same no matter what replica is processing them, right?

Hi Marcio,

In TM1 it is actually the daemon service that decides which request goes to which replica of a database. And you are right, any request not deemed idempotent in essence goes to the leader replica, all other requests, like executing MDX queries, already get spread across all replicas that are up to date at the time of the request. This means that in a system with a lot of updates all the time more load ends up on the leader whereas systems with barely any updates, if at all, request load gets spread pretty evenly.

Cheers,

-H

Hi

While i read the documentation I realized that all the work is made by OpenShift, I mean, it is not possible to actively decide which replica will get the job or not, because necessarily there is a leader which deals with all the writes and synchronization.

But what about MDX queries? Are them at least run in parallel or balanced between replicas? Is there any round-robin taking place between replicas in the case of read-only quests?

Regards

Márcio

Hi good morning

As far as I understood, we have automatic database replication with PAaaS, but there are not any load balance between these replicas so far. Are there plans for changing that, some enhancement planned? When we create a database replica, all the tasks to balance load are up to users / customer tailor made?

Regards

Márcio

Cache is managed separately though and does not span replicas: https://www.linkedin.com/pulse/deep-dive-caching-new-planning-analytics-engine-ryan-clapp-oupge/

Thanks, understood.

Regards

Rules are only processed during query time and as such only influence 'snapshots' of the data, feeders on the other hand are executed as part of update processes but in the case of a v12 replica those end up being executed as updates are made, either on the leader as a result of an update or on a follower as a result of applying the changes made on the leader first.

A final question here - rules and feeders processing can be seen as idempotent, as its final results are the same no matter what replica is processing them, right?

Hi Marcio,

In TM1 it is actually the daemon service that decides which request goes to which replica of a database. And you are right, any request not deemed idempotent in essence goes to the leader replica, all other requests, like executing MDX queries, already get spread across all replicas that are up to date at the time of the request. This means that in a system with a lot of updates all the time more load ends up on the leader whereas systems with barely any updates, if at all, request load gets spread pretty evenly.

Cheers,

-H

Hi

While i read the documentation I realized that all the work is made by OpenShift, I mean, it is not possible to actively decide which replica will get the job or not, because necessarily there is a leader which deals with all the writes and synchronization.

But what about MDX queries? Are them at least run in parallel or balanced between replicas? Is there any round-robin taking place between replicas in the case of read-only quests?

Regards

Márcio

Hi good morning

As far as I understood, we have automatic database replication with PAaaS, but there are not any load balance between these replicas so far. Are there plans for changing that, some enhancement planned? When we create a database replica, all the tasks to balance load are up to users / customer tailor made?

Regards

Márcio