ignite-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Mikhail Cherkasov <mcherka...@gridgain.com>
Subject Re: Best practise for setting Ignite.Active to true when using persistence layer in Ignite 2.1
Date Fri, 04 Aug 2017 14:41:32 GMT
Hi Raymond,

1.       Can I set the grid active once all primaries are active, but
before all backups are active? Or do I need to have the entire cluster
running before setting active to true?

*A cluster will work, but it will try to restore backups which are missed
and this will cause rebalancing, so it's better to wait for all nodes is
restored.*

2.       Is there an assumption that the cluster size is static and known
at the time of startup/restart? If so, how are dynamic clusters that grow
and shrink according to demand catered for?

*well, you can add new nodes when a cluster is active or inactive this
doesn't really matter, in case if the cluster isn't active, only after
activation new nodes will be used for data storing. There's no convenient
way to shrink cluster, you can kill nodes wait when rebalancing is done and
kill next one and so on.*

3.       Is it usually the responsibility of one of the cluster nodes to
set active to true, or is it the responsibility of an external client
server node on the grid to detect the point at which the cluster running
correctly and set active to true? If a cluster type node is used, does this
result in a swarm of activation messages as each node inspects the latest
topology changed event and sets active to true?

*You can use both approaches, but in the case of implementing activating
login on server nodes you should surround Ignite.active(true) with try
catch, the exception can be thrown due to concurrent activation. Also, you
can check whether the node is a coordinator and run activation on it only.*

4. This raises a more general coordinate question: You can’t create or
locate a cache in the grid until the grid is active (when using Ignite
persistence). Which means you need to stall any logic in the server nodes
that wants to do this until the grid is marked as active. How does such
logic detect when the grid becomes active? Should it spin and check
periodically, or is there an activation event that can be leveraged?

The Ignite persistence example at
https://github.com/apache/ignite/tree/master/examples/src/main/java/org/apache/ignite/examples/persistentstore
does not address these aspects.

*Unfortunately, there's no event for this now, so you should spin around
Ignite.active()*

Thanks,
Mikhail.


On Fri, Aug 4, 2017 at 5:01 AM, Raymond Wilson <raymond_wilson@trimble.com>
wrote:

> Sorry, adding a question:
>
>
>
> 4. This raises a more general coordinate question: You can’t create or
> locate a cache in the grid until the grid is active (when using Ignite
> persistence). Which means you need to stall any logic in the server nodes
> that wants to do this until the grid is marked as active. How does such
> logic detect when the grid becomes active? Should it spin and check
> periodically, or is there an activation event that can be leveraged?
>
>
>
> The Ignite persistence example at https://github.com/apache/
> ignite/tree/master/examples/src/main/java/org/apache/ignite/examples/
> persistentstore does not address these aspects.
>
>
>
> Thanks,
> Raymond.
>
>
>
> *From:* Raymond Wilson [mailto:raymond_wilson@trimble.com]
> *Sent:* Friday, August 4, 2017 1:56 PM
> *To:* 'user@ignite.apache.org'
> *Subject:* RE: Best practise for setting Ignite.Active to true when using
> persistence layer in Ignite 2.1
>
>
>
> Hi Michael,
>
>
>
> Some more questions:
>
>
>
> 1.       Can I set the grid active once all primaries are active, but
> before all backups are active? Or do I need to have the entire cluster
> running before setting active to true?
>
> 2.       Is there an assumption that the cluster size is static and known
> at the time of startup/restart? If so, how are dynamic clusters that grow
> and shrink according to demand catered for?
>
> 3.       Is it usually the responsibility of one of the cluster nodes to
> set active to true, or is it the responsibility of an external client
> server node on the grid to detect the point at which the cluster running
> correctly and set active to true? If a cluster type node is used, does this
> result in a swarm of activation messages as each node inspects the latest
> topology changed event and sets active to true?
>
>
>
> Thanks,
>
> Raymond.
>
>
>
>
>
> *From:* Michael Cherkasov [mailto:michael.cherkasov@gmail.com
> <michael.cherkasov@gmail.com>]
> *Sent:* Thursday, August 3, 2017 9:27 PM
>
> *To:* user@ignite.apache.org
> *Subject:* Re: Best practise for setting Ignite.Active to true when using
> persistence layer in Ignite 2.1
>
>
>
> Hi Raymond,
>
>
>
> Unfortunately right now there's no auto-activation, restarting cluster is
> like rare event that should be controlled
>
> manually. However you can listen for EVT_NODE_JOINED event, when all nodes
> in place you can activate a cluster.
>
>
>
> And you only need this if you have ignite persistence turn on and you have
> some data on the disk.
>
>
>
> Thanks,
>
> Mikhail.
>
>
>
> 2017-08-03 6:58 GMT+03:00 Raymond Wilson <raymond_wilson@trimble.com>:
>
> Michael,
>
>
>
> Is there a reference implementation in Ignite 2.1 for an agent that
> listens to topology changes to decide when to set active to true?
>
>
>
> Thanks,
>
> Raymond.
>
>
>
> *From:* Michael Cherkasov [mailto:michael.cherkasov@gmail.com]
> *Sent:* Thursday, August 3, 2017 1:25 AM
>
>
> *To:* user@ignite.apache.org
> *Subject:* Re: Best practise for setting Ignite.Active to true when using
> persistence layer in Ignite 2.1
>
>
>
> >Does this mean we have to listen to events of server nodes going up and
> down and activate and deactivate the cluster?
>
>
>
> No, you need to deactivate cluster when you going to shutdown the whole
> cluster. And when you return cluster back to  online, you need to wait when
> all nodes are in place and then activate it.
>
>
>
>
>
> 2017-08-02 16:22 GMT+03:00 Rohan Shetty <rohan.shetty@gmail.com>:
>
> Does this mean we have to listen to events of server nodes going up and
> down and activate and deactivate the cluster?
>
>
>
> On Wed, Aug 2, 2017 at 3:18 PM, Michael Cherkasov <
> michael.cherkasov@gmail.com> wrote:
>
> when all nodes are up, so in latest topology snapshot you can see servers
> count == servers count you run, then cluster can be activated.
>
>
>
> 2017-08-02 0:51 GMT+03:00 Raymond Wilson <raymond_wilson@trimble.com>:
>
> Hi Mikhail,
>
>
>
> Thanks for the clarifications.
>
>
>
> Yes, I knew setting active was only required when using the persistence
> layer, which is the topic of the question J
>
>
>
> I was interested if there were best practices or approaches for
> determining when the grid had fully initialized. I realise this is somewhat
> application specific, but was looking for an established pattern before I
> invented one myself.
>
>
>
> In my case I have an affinity function that responds to topology changes
> which intrinsically would know when it had a ‘quorum’. Is this a typical
> place for setting active to true.
>
>
>
> Thanks,
>
> Raymond.
>
>
>
> *From:* Mikhail Cherkasov [mailto:mcherkasov@gridgain.com]
> *Sent:* Tuesday, August 1, 2017 11:59 PM
> *To:* user@ignite.apache.org
> *Subject:* Re: Best practise for setting Ignite.Active to true when using
> persistence layer in Ignite 2.1
>
>
>
> Hi Raymond,
>
>
>
> Ignite cluster is inactive on startup only if persistence is enabled. This
> is done to avoid unnecessary partition exchanges between nodes,
>
> for example, if you have 3 nodes and 1 backup enabled and you start only 2
> of 3 nodes, then they will treat the third node as dead and start process
>
> of restoring data from backup and rebalance data to spread them among 2
> nodes, when you add the missed third node back the process will be repeated.
>
>
>
> So we start cluster as in active. When all nodes are started and ready, so
> no cluster topology changes aren't expected, you should activate cluster.
>
>
>
> Also when you turn off cluster, some nodes can still accept request for
> data update and other nodes won't see it, so understand which node has the
> latest
>
> data we need to start all nodes first and only then activate cluster.
>
>
>
> Thanks,
>
> Mikhail.
>
>
>
> On Tue, Aug 1, 2017 at 5:05 AM, Raymond Wilson <raymond_wilson@trimble.com>
> wrote:
>
> Hi,
>
>
>
> I am experimenting with a POC looking into using the Ignite persistence
> layer.
>
>
>
> One aspect of this is setting the grid to be ‘Active’ after all cache grid
> nodes have instantiated.
>
>
>
> In practical terms, what is the best practice for ensuring the cluster is
> running and in a good state to be set to active? What is the downside of
> setting active to true before all grid nodes are running?
>
>
>
> Thanks,
>
> Raymond.
>
>
>
>
>
>
>
> --
>
> Thanks,
>
> Mikhail.
>
>
>
>
>
>
>
>
>



-- 
Thanks,
Mikhail.

Mime
View raw message