You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@ignite.apache.org by the_palakkaran <ji...@suntecsbs.com> on 2018/06/07 12:04:50 UTC
Ignite clustering doubts
Hi,
The below is my scenario:
I have a CustomerCache. I have externalizable keys and models. Customers ids
range from 1 to 40000. I have 4 nodes.
I have few doubts on clustering in ignite:
1. How can I restrict 10000 records on each node?
2. How does ignite distribute records between each node? On what basis? Is
there any relevance of AffinityFunction here? Is there a demo of how to use
that ?
3. How would fail over and re-balancing work when a node fails?
4. Do I need to configure number of backups? Is the number of backups
dependent on number of nodes?
5. Will I have any problem to access data from any of these nodes or a
client node since I have used externalization?
6. Is there any possible way to query data on a single node?
7. Leader election is based on age, as per documentation. Can I control this
election?
8. Zero deployment means I do not need to have models deployed on every
class right? Is this the same case for externalized models?
--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/
Re: Ignite clustering doubts
Posted by Mikael <mi...@telia.com>.
The documentation does explain it pretty well:
https://apacheignite.readme.io/docs/affinity-collocation
Den 2018-06-08 kl. 07:41, skrev the_palakkaran:
> Why do I want to map keys to partition ids? Can you tell me more, please?
>
> By keys, it means the keys used for the cache right? If so, particular keys
> will always go to some node, is that so?
>
>
>
> --
> Sent from: http://apache-ignite-users.70518.x6.nabble.com/
>
>
Re: Ignite clustering doubts
Posted by the_palakkaran <ji...@suntecsbs.com>.
Why do I want to map keys to partition ids? Can you tell me more, please?
By keys, it means the keys used for the cache right? If so, particular keys
will always go to some node, is that so?
--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/
Re: Ignite clustering doubts
Posted by Ilya Kasnacheev <il...@gmail.com>.
Hello!
Affinity function determines mapping of keys to partition ids and of
partition ids to nodes. Usually you only override the first part since
changing the second one is a research problem.
Regards,
--
Ilya Kasnacheev
2018-06-07 18:36 GMT+03:00 the_palakkaran <ji...@suntecsbs.com>:
> Wow !!!!! Thanks a ton. If there was an upvote thing here,I would have
> surely
> broke it.
>
> Can you tell me what is the need of AffinityFunction then? Also can I
> control this Rendezvous Hashing?
>
>
>
>
> --
> Sent from: http://apache-ignite-users.70518.x6.nabble.com/
>
Re: Ignite clustering doubts
Posted by the_palakkaran <ji...@suntecsbs.com>.
Wow !!!!! Thanks a ton. If there was an upvote thing here,I would have surely
broke it.
Can you tell me what is the need of AffinityFunction then? Also can I
control this Rendezvous Hashing?
--
Sent from: http://apache-ignite-users.70518.x6.nabble.com/
Re: Ignite clustering doubts
Posted by Ilya Kasnacheev <il...@gmail.com>.
Hello!
This might be helpful: https://apacheignite.readme.io/docs/data-grid
1. It will be done automatically, every node will have around 10000 records.
2. They will be distributed according to Rendezvous hashing of partition
ids.
3. If you have sufficient backups, data will be re-balanced to ensure that
# of backups. If you don't have sufficient backups you will lose data.
4. Yes, you do. It depends on your setup and reliability considerations.
5. As far as I remember you had some pretty non-trivial considerations, so
your mileage may vary. Try it and see.
6. Yes, e.g. by using localPeek().
7. Why would you want to?
8. I'm afraid you might need to have models deployed universally to use
externalized classes.
Regards,
--
Ilya Kasnacheev
2018-06-07 15:04 GMT+03:00 the_palakkaran <ji...@suntecsbs.com>:
> Hi,
>
> The below is my scenario:
>
> I have a CustomerCache. I have externalizable keys and models. Customers
> ids
> range from 1 to 40000. I have 4 nodes.
>
> I have few doubts on clustering in ignite:
>
> 1. How can I restrict 10000 records on each node?
>
> 2. How does ignite distribute records between each node? On what basis? Is
> there any relevance of AffinityFunction here? Is there a demo of how to use
> that ?
>
> 3. How would fail over and re-balancing work when a node fails?
>
> 4. Do I need to configure number of backups? Is the number of backups
> dependent on number of nodes?
>
> 5. Will I have any problem to access data from any of these nodes or a
> client node since I have used externalization?
>
> 6. Is there any possible way to query data on a single node?
>
> 7. Leader election is based on age, as per documentation. Can I control
> this
> election?
>
> 8. Zero deployment means I do not need to have models deployed on every
> class right? Is this the same case for externalized models?
>
>
>
> --
> Sent from: http://apache-ignite-users.70518.x6.nabble.com/
>