Hello!
If we introduce IGNITE_OFFHEAP_MEMORY_PER_NODE, there will be an
expectation that this number will be automatically considered when creating
data region.
I think that default value have to go definitely, but you can surely use
IGNITE_MEMORY_OVERHEAD_PER_NODE *as* IGNITE_OFFHEAP_MEMORY_PER_NODE.
For example, for an Ignite instances of 4G heap and 48G offheap, I would
use IGNITE_MEMORY_PER_NODE of '4G' and IGNITE_MEMORY_OVERHEAD_PER_NODE of
'52G' (offheap + 10%).
We could surely rename both variables, keep old ones for convenience. E.g.
to IGNITE_JVM_MEMORY_PER_NODE and IGNITE_MEMORY_PER_NODE. Or we could just
update the docs.
Are you an active user of Ignite YARN, btw? Does this problem actually bite
you or is your interest more of theoretical? Frankly speaking, we aren't
seeing enough adoption of these features and it puts off when considering
investing more time in them.
Note that this issue probably also affects Ignite Mesos.
Regards,
--
Ilya Kasnacheev
2018-04-25 5:26 GMT+03:00 endianignite <
[hidden email]>:
> I see that our YARN parameters
> <
https://apacheignite.readme.io/docs/yarn-deployment> and default
> values
> <
https://apacheignite.readme.io/docs#section-configuration> do not quite
> align with Ignite 2.x Durable Memory:
>
> IGNITE_MEMORY_OVERHEAD_PER_NODE:
>
> The amount of memory allocated to handle JVM native overheads, interned
> Strings,... but also off-heap memory if you use it. Memory requested to
> YARN
> for containers running Ignite node is the sum of IGNITE_MEMORY_PER_NODE and
> IGNITE_MEMORY_OVERHEAD_PER_NODE
>
> Default value: IGNITE_MEMORY_PER_NODE * 0.10, with a minimum of 384
>
> I think we should add another parameter that is specifically the amount of
> off-heap memory that should be requested: IGNITE_OFFHEAP_MEMORY_PER_NODE.
>
> The total requested would then be IGNITE_MEMORY_PER_NODE +
> IGNITE_MEMORY_OVERHEAD_PER_NODE+IGNITE_OFFHEAP_MEMORY_PER_NODE.
>
>
>
>
>
>
> --
> Sent from:
http://apache-ignite-developers.2346864.n4.nabble.com/>
Locked
