Locking of partition with affinityRun/affinityCall

[dmagda]

Taras,

There is a question in regards to the feature contributed by you recently
https://issues.apache.org/jira/browse/IGNITE-2310 <https://issues.apache.org/jira/browse/IGNITE-2310>

The Java API doc says that the partition will not be migrated while a job is being executed on a target node.

Does it mean that
- the rebalancing will be postponed in general or
- the rebalancing of the partition will be started, moving its content to a new primary, but the partition will not be evicted from the target node while the job is running

In my understanding the second point is correct and I’ve documented the feature saying that the partition is not evicted
http://apacheignite.gridgain.org/v1.7/docs/collocate-compute-and-data#affinity-call-and-run-methods <http://apacheignite.gridgain.org/v1.7/docs/collocate-compute-and-data#affinity-call-and-run-methods>

Please clarify what’s true and update Java API doc if my current understanding is correct.

—
Denis

[Taras Ledkov]

Denis,

The second point is absolutely correct.
The rebalancing isn't blocked. The partition will not be evicted while the
partition is reserved.

I'll update javadoc ASAP.


On Wed, Nov 30, 2016 at 10:23 PM, Denis Magda <[hidden email]> wrote:

> Taras,
>
> There is a question in regards to the feature contributed by you recently
> https://issues.apache.org/jira/browse/IGNITE-2310
>
> The Java API doc says that the partition will not be migrated while a job
> is being executed on a target node.
>
> Does it mean that
> - the rebalancing will be postponed in general or
> - the rebalancing of the partition will be started, moving its content to
> a new primary, but the partition will not be evicted from the target node
> while the job is running
>
> In my understanding the second point is correct and I’ve documented the
> feature saying that the partition is not evicted
> http://apacheignite.gridgain.org/v1.7/docs/collocate-
> compute-and-data#affinity-call-and-run-methods
>
> Please clarify what’s true and update Java API doc if my current
> understanding is correct.
>
> —
> Denis
>

[Taras Ledkov]

Denis,

Please review the javadoc of the affinity jobs.

PR: https://github.com/apache/ignite/pull/1550

On 08.12.2016 12:06, Taras Ledkov wrote:

> Denis,
>
> The second point is absolutely correct.
> The rebalancing isn't blocked. The partition will not be evicted while
> the partition is reserved.
>
> I'll update javadoc ASAP.
>
>
> On Wed, Nov 30, 2016 at 10:23 PM, Denis Magda <[hidden email]
> <mailto:[hidden email]>> wrote:
>
>     Taras,
>
>     There is a question in regards to the feature contributed by you
>     recently
>     https://issues.apache.org/jira/browse/IGNITE-2310
>     <https://issues.apache.org/jira/browse/IGNITE-2310>
>
>     The Java API doc says that the partition will not be migrated
>     while a job is being executed on a target node.
>
>     Does it mean that
>     - the rebalancing will be postponed in general or
>     - the rebalancing of the partition will be started, moving its
>     content to a new primary, but the partition will not be evicted
>     from the target node while the job is running
>
>     In my understanding the second point is correct and I’ve
>     documented the feature saying that the partition is not evicted
>     http://apacheignite.gridgain.org/v1.7/docs/collocate-compute-and-data#affinity-call-and-run-methods
>     <http://apacheignite.gridgain.org/v1.7/docs/collocate-compute-and-data#affinity-call-and-run-methods>
>
>     Please clarify what’s true and update Java API doc if my current
>     understanding is correct.
>
>     —
>     Denis
>
>

--
Taras Ledkov
Mail-To: [hidden email]

[Taras Ledkov]

Denis,

I've fixed javadocs for affinityRun/affinityCall.

Please review the changes at the PR:
https://github.com/apache/ignite/pull/1550


On 08.12.2016 12:06, Taras Ledkov wrote:

> Denis,
>
> The second point is absolutely correct.
> The rebalancing isn't blocked. The partition will not be evicted while
> the partition is reserved.
>
> I'll update javadoc ASAP.
>
>
> On Wed, Nov 30, 2016 at 10:23 PM, Denis Magda <[hidden email]
> <mailto:[hidden email]>> wrote:
>
>     Taras,
>
>     There is a question in regards to the feature contributed by you
>     recently
>     https://issues.apache.org/jira/browse/IGNITE-2310
>     <https://issues.apache.org/jira/browse/IGNITE-2310>
>
>     The Java API doc says that the partition will not be migrated
>     while a job is being executed on a target node.
>
>     Does it mean that
>     - the rebalancing will be postponed in general or
>     - the rebalancing of the partition will be started, moving its
>     content to a new primary, but the partition will not be evicted
>     from the target node while the job is running
>
>     In my understanding the second point is correct and I’ve
>     documented the feature saying that the partition is not evicted
>     http://apacheignite.gridgain.org/v1.7/docs/collocate-compute-and-data#affinity-call-and-run-methods
>     <http://apacheignite.gridgain.org/v1.7/docs/collocate-compute-and-data#affinity-call-and-run-methods>
>
>     Please clarify what’s true and update Java API doc if my current
>     understanding is correct.
>
>     —
>     Denis
>
>

--
Taras Ledkov
Mail-To: [hidden email]

[dmagda]

Taras, thanks. Looks good to merge.

—
Denis

> On Feb 20, 2017, at 4:43 AM, Taras Ledkov <[hidden email]> wrote:
>
> Denis,
>
> I've fixed javadocs for affinityRun/affinityCall.
>
> Please review the changes at the PR: https://github.com/apache/ignite/pull/1550
>
>
> On 08.12.2016 12:06, Taras Ledkov wrote:
>> Denis,
>>
>> The second point is absolutely correct.
>> The rebalancing isn't blocked. The partition will not be evicted while the partition is reserved.
>>
>> I'll update javadoc ASAP.
>>
>>
>> On Wed, Nov 30, 2016 at 10:23 PM, Denis Magda <[hidden email] <mailto:[hidden email]>> wrote:
>>
>>    Taras,
>>
>>    There is a question in regards to the feature contributed by you
>>    recently
>>    https://issues.apache.org/jira/browse/IGNITE-2310
>>    <https://issues.apache.org/jira/browse/IGNITE-2310>
>>
>>    The Java API doc says that the partition will not be migrated
>>    while a job is being executed on a target node.
>>
>>    Does it mean that
>>    - the rebalancing will be postponed in general or
>>    - the rebalancing of the partition will be started, moving its
>>    content to a new primary, but the partition will not be evicted
>>    from the target node while the job is running
>>
>>    In my understanding the second point is correct and I’ve
>>    documented the feature saying that the partition is not evicted
>>    http://apacheignite.gridgain.org/v1.7/docs/collocate-compute-and-data#affinity-call-and-run-methods
>>    <http://apacheignite.gridgain.org/v1.7/docs/collocate-compute-and-data#affinity-call-and-run-methods>
>>
>>    Please clarify what’s true and update Java API doc if my current
>>    understanding is correct.
>>
>>    —
>>    Denis
>>
>>
>
> --
> Taras Ledkov
> Mail-To: [hidden email]
>

[Vladimir Ozerov]

I do not like the change because it is too verbose and adds no useful
information for users. Moreover, normally users have no idea what
"partition eviction" means. Let's simply write "Partition's data is
guaranteed to be available on the node while job is being executed". This
should be enough.

On Wed, Feb 22, 2017 at 2:00 AM, Denis Magda <[hidden email]> wrote:

> Taras, thanks. Looks good to merge.
>
> —
> Denis
>
> > On Feb 20, 2017, at 4:43 AM, Taras Ledkov <[hidden email]> wrote:
> >
> > Denis,
> >
> > I've fixed javadocs for affinityRun/affinityCall.
> >
> > Please review the changes at the PR: https://github.com/apache/
> ignite/pull/1550
> >
> >
> > On 08.12.2016 12:06, Taras Ledkov wrote:
> >> Denis,
> >>
> >> The second point is absolutely correct.
> >> The rebalancing isn't blocked. The partition will not be evicted while
> the partition is reserved.
> >>
> >> I'll update javadoc ASAP.
> >>
> >>
> >> On Wed, Nov 30, 2016 at 10:23 PM, Denis Magda <[hidden email]
> <mailto:[hidden email]>> wrote:
> >>
> >>    Taras,
> >>
> >>    There is a question in regards to the feature contributed by you
> >>    recently
> >>    https://issues.apache.org/jira/browse/IGNITE-2310
> >>    <https://issues.apache.org/jira/browse/IGNITE-2310>
> >>
> >>    The Java API doc says that the partition will not be migrated
> >>    while a job is being executed on a target node.
> >>
> >>    Does it mean that
> >>    - the rebalancing will be postponed in general or
> >>    - the rebalancing of the partition will be started, moving its
> >>    content to a new primary, but the partition will not be evicted
> >>    from the target node while the job is running
> >>
> >>    In my understanding the second point is correct and I’ve
> >>    documented the feature saying that the partition is not evicted
> >>    http://apacheignite.gridgain.org/v1.7/docs/collocate-
> compute-and-data#affinity-call-and-run-methods
> >>    <http://apacheignite.gridgain.org/v1.7/docs/collocate-
> compute-and-data#affinity-call-and-run-methods>
> >>
> >>    Please clarify what’s true and update Java API doc if my current
> >>    understanding is correct.
> >>
> >>    —
> >>    Denis
> >>
> >>
> >
> > --
> > Taras Ledkov
> > Mail-To: [hidden email]
> >
>
>

[dmagda]

I’ve reworked the doc removing “migration” mentioning at all that caused the confusion. Merged my changes:

* It's guaranteed that the data of the whole partition, the affinity key belongs to,
* will present on the destination node throughout the job execution.

Taras, please close your pull-request. No need to merge it.

—
Denis

> On Mar 1, 2017, at 1:13 AM, Vladimir Ozerov <[hidden email]> wrote:
>
> I do not like the change because it is too verbose and adds no useful
> information for users. Moreover, normally users have no idea what
> "partition eviction" means. Let's simply write "Partition's data is
> guaranteed to be available on the node while job is being executed". This
> should be enough.
>
> On Wed, Feb 22, 2017 at 2:00 AM, Denis Magda <[hidden email]> wrote:
>
>> Taras, thanks. Looks good to merge.
>>
>> —
>> Denis
>>
>>> On Feb 20, 2017, at 4:43 AM, Taras Ledkov <[hidden email]> wrote:
>>>
>>> Denis,
>>>
>>> I've fixed javadocs for affinityRun/affinityCall.
>>>
>>> Please review the changes at the PR: https://github.com/apache/
>> ignite/pull/1550
>>>
>>>
>>> On 08.12.2016 12:06, Taras Ledkov wrote:
>>>> Denis,
>>>>
>>>> The second point is absolutely correct.
>>>> The rebalancing isn't blocked. The partition will not be evicted while
>> the partition is reserved.
>>>>
>>>> I'll update javadoc ASAP.
>>>>
>>>>
>>>> On Wed, Nov 30, 2016 at 10:23 PM, Denis Magda <[hidden email]
>> <mailto:[hidden email]>> wrote:
>>>>
>>>>   Taras,
>>>>
>>>>   There is a question in regards to the feature contributed by you
>>>>   recently
>>>>   https://issues.apache.org/jira/browse/IGNITE-2310
>>>>   <https://issues.apache.org/jira/browse/IGNITE-2310>
>>>>
>>>>   The Java API doc says that the partition will not be migrated
>>>>   while a job is being executed on a target node.
>>>>
>>>>   Does it mean that
>>>>   - the rebalancing will be postponed in general or
>>>>   - the rebalancing of the partition will be started, moving its
>>>>   content to a new primary, but the partition will not be evicted
>>>>   from the target node while the job is running
>>>>
>>>>   In my understanding the second point is correct and I’ve
>>>>   documented the feature saying that the partition is not evicted
>>>>   http://apacheignite.gridgain.org/v1.7/docs/collocate-
>> compute-and-data#affinity-call-and-run-methods
>>>>   <http://apacheignite.gridgain.org/v1.7/docs/collocate-
>> compute-and-data#affinity-call-and-run-methods>
>>>>
>>>>   Please clarify what’s true and update Java API doc if my current
>>>>   understanding is correct.
>>>>
>>>>   —
>>>>   Denis
>>>>
>>>>
>>>
>>> --
>>> Taras Ledkov
>>> Mail-To: [hidden email]
>>>
>>
>>