You are viewing a plain text version of this content. The canonical link for it is here.
Posted to user@helix.apache.org by "Leo, Li" <go...@gmail.com> on 2020/11/01 02:19:06 UTC
Re: User-Defined Rebalancing
Hi JunKai and Jiajun,
Thank you for your help. When I tried to use a customized Rebalancer, I saw
the rebalancer is triggered and the idea state is set by it.
3784 18:38:38.268 [GenericHelixController-event_process] INFO
c.u.k.w.c.CustomizedRebalancer - CustomizedRebalancer.computeNewIdealState
invoked!
3785 18:38:38.268 [GenericHelixController-event_process] INFO
c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
localhost_12001 dca1d_0 online
3786 18:38:38.268 [GenericHelixController-event_process] INFO
c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
localhost_12002 dca1d_1 online
3787 18:38:38.269 [GenericHelixController-event_process] WARN
o.a.h.c.s.BestPossibleStateCalcStage - Event b4896f24_DEFAULT : Rebalancer
does not have a mapping calculator, defaulting to SEMI_AUT O, resource:
dca1d
However the Helix UI shows "no date to display". The snapshot is attached.
The EXTERNALVIEW and IDEASTATE are:
>>> get /Watchdog/EXTERNALVIEW/dca1d
{
"id":"dca1d"
,"simpleFields":{
"BUCKET_SIZE":"0"
,"IDEAL_STATE_MODE":"AUTO"
,"NUM_PARTITIONS":"2"
,"REBALANCER_CLASS_NAME":"mypackage.controller.CustomizedRebalancer"
,"REBALANCE_MODE":"USER_DEFINED"
,"REBALANCE_STRATEGY":"DEFAULT"
,"REPLICAS":"1"
,"STATE_MODEL_DEF_REF":"MyStateModel"
,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
}
,"listFields":{
}
,"mapFields":{
"dca1d_0":{
"localhost_12010":"ONLINE"
}
,"dca1d_1":{
"localhost_12000":"ONLINE"
}
}
}
>>> get /Watchdog/IDEALSTATES/dca1d
{
"id" : "dca1d",
"simpleFields" : {
"IDEAL_STATE_MODE" : "AUTO",
"NUM_PARTITIONS" : "2",
"REBALANCER_CLASS_NAME" : "mypackage.controller.CustomizedRebalancer",
"REBALANCE_MODE" : "USER_DEFINED",
"REBALANCE_STRATEGY" : "DEFAULT",
"REPLICAS" : "1",
"STATE_MODEL_DEF_REF" : "MyStateModel",
"STATE_MODEL_FACTORY_NAME" : "DEFAULT"
},
"mapFields" : {
"dca1d_0" : {
},
"dca1d_1" : {
}
},
"listFields" : {
"dca1d_0" : [ ],
"dca1d_1" : [ ]
}
}
Comparing with before, when i use CUSTOMIZED rebalancer and model, The
EXTERNALVIEW and IDEASTATE were working well with Helix UI:
>>> get /Watchdog/IDEALSTATES/dca1d
{
"id":"dca1d"
,"simpleFields":{
"IDEAL_STATE_MODE":"CUSTOMIZED"
,"MAX_PARTITIONS_PER_INSTANCE":"1"
,"NUM_PARTITIONS":"2"
,"REBALANCE_MODE":"CUSTOMIZED"
,"REPLICAS":"1"
,"STATE_MODEL_DEF_REF":"MyStateModel"
,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
}
,"listFields":{
}
,"mapFields":{
"0":{
"localhost_12004":"ONLINE"
}
,"1":{
"localhost_12005":"ONLINE"
}
}
}
>>> get /Watchdog/EXTERNALVIEW/dca1d
{
"id":"dca1d"
,"simpleFields":{
"BUCKET_SIZE":"0"
,"IDEAL_STATE_MODE":"CUSTOMIZED"
,"MAX_PARTITIONS_PER_INSTANCE":"1"
,"NUM_PARTITIONS":"2"
,"REBALANCE_MODE":"CUSTOMIZED"
,"REPLICAS":"1"
,"STATE_MODEL_DEF_REF":"MyStateModel"
,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
}
,"listFields":{
}
,"mapFields":{
"0":{
"localhost_12004":"ONLINE"
}
,"1":{
"localhost_12005":"ONLINE"
}
}
}
I see that the mapFields key changed after using "USER_DEFINED" rebalancer.
Here are the code i used:
public static void addResource(String resource, int num) {
clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num,
STATE_MODEL_NAME);
clusterSetup.addResourceProperty(CLUSTER_NAME, resource,
IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(),
CustomizedRebalancer.class.getName());
clusterSetup.addResourceProperty(CLUSTER_NAME, resource,
IdealState.IdealStateProperty.REBALANCE_MODE.toString(),
IdealState.RebalanceMode.USER_DEFINED.toString());
clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
What is the right way to set a rebalancer?
The rebalancer method has a single resource interface. What is the right
way to rebalance all resources in the cluster for any event change in the
cluster? My concern is the sequence and thread safety.
@Override
public IdealState computeNewIdealState(String resourceName, IdealState
currentIdealState,
CurrentStateOutput currentStateOutput, ClusterDataCache clusterData) {
Regards,
Leo
On Fri, Oct 30, 2020 at 10:21 PM Wang Jiajun <er...@gmail.com> wrote:
> Hi Leo,
>
> Are you looking for guidance on how to configure the IdealState so the
> resource will be rebalanced by your user_defined rebalancer? Maybe you can
> refer to this test case TestCustomizedIdealStateRebalancer.
>
> Best Regards,
> Jiajun
>
>
> On Fri, Oct 30, 2020 at 10:11 PM Xue Junkai <jx...@apache.org> wrote:
>
>> Hi Leo,
>>
>> What kind of examples are you looking for? Participant side? Or how to
>> build IdealState?
>>
>> Best,
>>
>> Junkai
>>
>> On Fri, Oct 30, 2020 at 9:57 PM Leo, Li <go...@gmail.com> wrote:
>>
>>> Hi,
>>> I am a new user of Helix. I wonder if you could give an example code of
>>> using this user_defined model
>>> https://helix.apache.org/1.0.1-docs/tutorial_user_def_rebalancer.html?
>>>
>>> Thank you,
>>> Leo
>>>
>>
Re: User-Defined Rebalancing
Posted by "Leo, Li" <go...@gmail.com>.
Aha, this totally explained. Thank you!
-Leo
On Sat, Oct 31, 2020 at 10:48 PM Xue Junkai <jx...@apache.org> wrote:
> Yeah. That's the reason. In the test, it uses CustomRebalancer which
> extends AbstractRebalancer.
> And here's the definition of that:
> https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/main/java/org/apache/helix/controller/rebalancer/AbstractRebalancer.java
> .
>
> You can follow the implementation of this class and have a try.
>
> best,
>
> Junkai
>
> On Sat, Oct 31, 2020 at 9:53 PM Leo, Li <go...@gmail.com> wrote:
>
>> did not implement MappingCalculator. here is code copied
>> from TestCustomizedIdealStateRebalancer. The interface you mentioned seems
>> related with
>> https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/test/java/org/apache/helix/integration/rebalancer/TestCustomRebalancer.java#L76
>> CustomRebalancer.
>>
>> Thanks.
>> Leo
>>
>> public class LeoDefinedRebalancer implements Rebalancer {
>> private static final Logger LOGGER =
>> LoggerFactory.getLogger(LeoDefinedRebalancer.class);
>>
>> @Override
>> public void init(HelixManager manager){
>> LOGGER.info("LeoDefinedRebalancer.computeNewIdealState setup!");
>> }
>>
>>
>> @Override
>> public IdealState computeNewIdealState(String resourceName, IdealState currentIdealState,
>> CurrentStateOutput currentStateOutput, ClusterDataCache clusterData) {
>> LOGGER.info("LeoDefinedRebalancer.computeNewIdealState invoked!");
>>
>> List<String> liveNodes = Lists.newArrayList(clusterData.getLiveInstances().keySet());
>> int i = 0;
>> for (String partition : currentIdealState.getPartitionSet()) {
>> int index = i++ % liveNodes.size();
>> String instance = liveNodes.get(index);
>> currentIdealState.getPreferenceList(partition).clear();
>> currentIdealState.getPreferenceList(partition).add(instance);
>>
>> currentIdealState.getInstanceStateMap(partition).clear();
>> currentIdealState.getInstanceStateMap(partition).put(instance, "ONLINE");
>> LOGGER.info("set ############### {}: {} {} online", resourceName, instance, partition);
>> }
>> currentIdealState.setReplicas("1");
>> return currentIdealState;
>> }
>> }
>>
>>
>> On Sat, Oct 31, 2020 at 9:39 PM Xue Junkai <jx...@apache.org> wrote:
>>
>>> What does your class look like? Does it implement the public interface
>>> MappingCalculator?This is the interface we casted to get the resource
>>> assignment.
>>>
>>> https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/main/java/org/apache/helix/controller/rebalancer/internal/MappingCalculator.java
>>>
>>>
>>> best,
>>>
>>> Junkai
>>>
>>> On Sat, Oct 31, 2020 at 9:30 PM Leo, Li <go...@gmail.com> wrote:
>>>
>>>> Hi Junkai,
>>>>
>>>> It's my fault to use a misleading class name. I renamed the class name
>>>> as LeoDefinedRebalancer. It is triggered:
>>>>
>>>> 2778 20:15:18.606 [GenericHelixController-event_process] INFO
>>>> c.u.k.w.c.LeoDefinedRebalancer - LeoDefinedRebalancer.computeNewIdealState
>>>> invoked!
>>>>
>>>> 2779 20:15:18.606 [GenericHelixController-event_process] INFO
>>>> c.u.k.w.c.LeoDefinedRebalancer - set ############### dca1e:
>>>> localhost_12001 dca1e_0 online
>>>>
>>>> 2780 20:15:18.606 [GenericHelixController-event_process] INFO
>>>> c.u.k.w.c.LeoDefinedRebalancer - set ############### dca1e:
>>>> localhost_12002 dca1e_1 online
>>>>
>>>> 2781 20:15:18.606 [GenericHelixController-event_process] WARN
>>>> o.a.h.c.s.BestPossibleStateCalcStage - Event 02e76c7b_DEFAULT : Rebalancer
>>>> does not have a mapping calculator, defaulting to SEMI_AUT O,
>>>> resource: dca1d
>>>>
>>>>
>>>> LeoDefinedRebalancer class is the test case code
>>>> <https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/test/java/org/apache/helix/integration/rebalancer/TestCustomizedIdealStateRebalancer.java#L54>
>>>> and runs together with other classes, but the issues remain the same.
>>>>
>>>> HelixManager hMan = startStandAloneController();
>>>> LeoDefinedRebalancer rebalancer = new LeoDefinedRebalancer();
>>>> rebalancer.init(hMan);
>>>>
>>>> clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num, STATE_MODEL_NAME);
>>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(), LeoDefinedRebalancer.class.getName());
>>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCE_MODE.toString(), IdealState.RebalanceMode.USER_DEFINED.toString());
>>>> clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
>>>>
>>>>
>>>> On Sat, Oct 31, 2020 at 7:41 PM Xue Junkai <jx...@apache.org> wrote:
>>>>
>>>>> Thanks for the details Leo.
>>>>>
>>>>> I think there is a misunderstanding between customized IdealState and
>>>>> user defined rebalancer. Actually, they are different levels of concept.
>>>>>
>>>>> For customized IdealState, it still uses Helix defined customized
>>>>> rebalancer, but it takes the map field as the final result. Helix only
>>>>> reacts to the map field change for partition/replica assignment.
>>>>>
>>>>> But user defined rebalancer is different. User has to implement the
>>>>> Rebalancer class and include that class in the same JVM when you start the
>>>>> controller. So Helix will try to use the class loader to load your class to
>>>>> do the computation. So the error could be you dont have the class in the
>>>>> same jar of starting the controller. Either you implement Rebalancer class
>>>>> and include it in controller jar. Or you can switch to use default auto
>>>>> rebalancer but with customized mode for the IdealState.
>>>>>
>>>>> Jiajun provided a test class as a good example of implementing the
>>>>> user defined rebalance. You can take a look if you would like to go that
>>>>> route.
>>>>>
>>>>> best,
>>>>>
>>>>> Junkai
>>>>>
>>>>> On Sat, Oct 31, 2020 at 7:19 PM Leo, Li <go...@gmail.com> wrote:
>>>>>
>>>>>> Hi JunKai and Jiajun,
>>>>>> Thank you for your help. When I tried to use a customized Rebalancer,
>>>>>> I saw the rebalancer is triggered and the idea state is set by it.
>>>>>>
>>>>>> 3784 18:38:38.268 [GenericHelixController-event_process] INFO
>>>>>> c.u.k.w.c.CustomizedRebalancer - CustomizedRebalancer.computeNewIdealState
>>>>>> invoked!
>>>>>>
>>>>>> 3785 18:38:38.268 [GenericHelixController-event_process] INFO
>>>>>> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
>>>>>> localhost_12001 dca1d_0 online
>>>>>>
>>>>>> 3786 18:38:38.268 [GenericHelixController-event_process] INFO
>>>>>> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
>>>>>> localhost_12002 dca1d_1 online
>>>>>>
>>>>>> 3787 18:38:38.269 [GenericHelixController-event_process] WARN
>>>>>> o.a.h.c.s.BestPossibleStateCalcStage - Event b4896f24_DEFAULT : Rebalancer
>>>>>> does not have a mapping calculator, defaulting to SEMI_AUT O,
>>>>>> resource: dca1d
>>>>>>
>>>>>>
>>>>>> However the Helix UI shows "no date to display". The snapshot is
>>>>>> attached. The EXTERNALVIEW and IDEASTATE are:
>>>>>>
>>>>>>
>>>>>> >>> get /Watchdog/EXTERNALVIEW/dca1d
>>>>>>
>>>>>> {
>>>>>>
>>>>>> "id":"dca1d"
>>>>>>
>>>>>> ,"simpleFields":{
>>>>>>
>>>>>> "BUCKET_SIZE":"0"
>>>>>>
>>>>>> ,"IDEAL_STATE_MODE":"AUTO"
>>>>>>
>>>>>> ,"NUM_PARTITIONS":"2"
>>>>>>
>>>>>>
>>>>>> ,"REBALANCER_CLASS_NAME":"mypackage.controller.CustomizedRebalancer"
>>>>>>
>>>>>> ,"REBALANCE_MODE":"USER_DEFINED"
>>>>>>
>>>>>> ,"REBALANCE_STRATEGY":"DEFAULT"
>>>>>>
>>>>>> ,"REPLICAS":"1"
>>>>>>
>>>>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>>>>
>>>>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>>>>
>>>>>> }
>>>>>>
>>>>>> ,"listFields":{
>>>>>>
>>>>>> }
>>>>>>
>>>>>> ,"mapFields":{
>>>>>>
>>>>>> "dca1d_0":{
>>>>>>
>>>>>> "localhost_12010":"ONLINE"
>>>>>>
>>>>>> }
>>>>>>
>>>>>> ,"dca1d_1":{
>>>>>>
>>>>>> "localhost_12000":"ONLINE"
>>>>>>
>>>>>> }
>>>>>>
>>>>>> }
>>>>>>
>>>>>> }
>>>>>>
>>>>>>
>>>>>> >>> get /Watchdog/IDEALSTATES/dca1d
>>>>>>
>>>>>> {
>>>>>>
>>>>>> "id" : "dca1d",
>>>>>>
>>>>>> "simpleFields" : {
>>>>>>
>>>>>> "IDEAL_STATE_MODE" : "AUTO",
>>>>>>
>>>>>> "NUM_PARTITIONS" : "2",
>>>>>>
>>>>>> "REBALANCER_CLASS_NAME" :
>>>>>> "mypackage.controller.CustomizedRebalancer",
>>>>>>
>>>>>> "REBALANCE_MODE" : "USER_DEFINED",
>>>>>>
>>>>>> "REBALANCE_STRATEGY" : "DEFAULT",
>>>>>>
>>>>>> "REPLICAS" : "1",
>>>>>>
>>>>>> "STATE_MODEL_DEF_REF" : "MyStateModel",
>>>>>>
>>>>>> "STATE_MODEL_FACTORY_NAME" : "DEFAULT"
>>>>>>
>>>>>> },
>>>>>>
>>>>>> "mapFields" : {
>>>>>>
>>>>>> "dca1d_0" : {
>>>>>>
>>>>>> },
>>>>>>
>>>>>> "dca1d_1" : {
>>>>>>
>>>>>> }
>>>>>>
>>>>>> },
>>>>>>
>>>>>> "listFields" : {
>>>>>>
>>>>>> "dca1d_0" : [ ],
>>>>>>
>>>>>> "dca1d_1" : [ ]
>>>>>>
>>>>>> }
>>>>>>
>>>>>> }
>>>>>>
>>>>>>
>>>>>> Comparing with before, when i use CUSTOMIZED rebalancer and model, The
>>>>>> EXTERNALVIEW and IDEASTATE were working well with Helix UI:
>>>>>>
>>>>>>
>>>>>> >>> get /Watchdog/IDEALSTATES/dca1d
>>>>>>
>>>>>> {
>>>>>>
>>>>>> "id":"dca1d"
>>>>>>
>>>>>> ,"simpleFields":{
>>>>>>
>>>>>> "IDEAL_STATE_MODE":"CUSTOMIZED"
>>>>>>
>>>>>> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>>>>>>
>>>>>> ,"NUM_PARTITIONS":"2"
>>>>>>
>>>>>> ,"REBALANCE_MODE":"CUSTOMIZED"
>>>>>>
>>>>>> ,"REPLICAS":"1"
>>>>>>
>>>>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>>>>
>>>>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>>>>
>>>>>> }
>>>>>>
>>>>>> ,"listFields":{
>>>>>>
>>>>>> }
>>>>>>
>>>>>> ,"mapFields":{
>>>>>>
>>>>>> "0":{
>>>>>>
>>>>>> "localhost_12004":"ONLINE"
>>>>>>
>>>>>> }
>>>>>>
>>>>>> ,"1":{
>>>>>>
>>>>>> "localhost_12005":"ONLINE"
>>>>>>
>>>>>> }
>>>>>>
>>>>>> }
>>>>>>
>>>>>> }
>>>>>>
>>>>>>
>>>>>> >>> get /Watchdog/EXTERNALVIEW/dca1d
>>>>>>
>>>>>> {
>>>>>>
>>>>>> "id":"dca1d"
>>>>>>
>>>>>> ,"simpleFields":{
>>>>>>
>>>>>> "BUCKET_SIZE":"0"
>>>>>>
>>>>>> ,"IDEAL_STATE_MODE":"CUSTOMIZED"
>>>>>>
>>>>>> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>>>>>>
>>>>>> ,"NUM_PARTITIONS":"2"
>>>>>>
>>>>>> ,"REBALANCE_MODE":"CUSTOMIZED"
>>>>>>
>>>>>> ,"REPLICAS":"1"
>>>>>>
>>>>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>>>>
>>>>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>>>>
>>>>>> }
>>>>>>
>>>>>> ,"listFields":{
>>>>>>
>>>>>> }
>>>>>>
>>>>>> ,"mapFields":{
>>>>>>
>>>>>> "0":{
>>>>>>
>>>>>> "localhost_12004":"ONLINE"
>>>>>>
>>>>>> }
>>>>>>
>>>>>> ,"1":{
>>>>>>
>>>>>> "localhost_12005":"ONLINE"
>>>>>>
>>>>>> }
>>>>>>
>>>>>> }
>>>>>>
>>>>>> }
>>>>>>
>>>>>>
>>>>>> I see that the mapFields key changed after using "USER_DEFINED"
>>>>>> rebalancer.
>>>>>>
>>>>>> Here are the code i used:
>>>>>>
>>>>>> public static void addResource(String resource, int num) {
>>>>>> clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num, STATE_MODEL_NAME);
>>>>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(), CustomizedRebalancer.class.getName());
>>>>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCE_MODE.toString(), IdealState.RebalanceMode.USER_DEFINED.toString());
>>>>>> clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
>>>>>>
>>>>>>
>>>>>> What is the right way to set a rebalancer?
>>>>>>
>>>>>>
>>>>>> The rebalancer method has a single resource interface. What is the
>>>>>> right way to rebalance all resources in the cluster for any event change in
>>>>>> the cluster? My concern is the sequence and thread safety.
>>>>>>
>>>>>> @Override
>>>>>> public IdealState computeNewIdealState(String resourceName, IdealState currentIdealState,
>>>>>> CurrentStateOutput currentStateOutput, ClusterDataCache clusterData) {
>>>>>>
>>>>>>
>>>>>>
>>>>>> Regards,
>>>>>>
>>>>>> Leo
>>>>>>
>>>>>>
>>>>>>
>>>>>> On Fri, Oct 30, 2020 at 10:21 PM Wang Jiajun <er...@gmail.com>
>>>>>> wrote:
>>>>>>
>>>>>>> Hi Leo,
>>>>>>>
>>>>>>> Are you looking for guidance on how to configure the IdealState so
>>>>>>> the resource will be rebalanced by your user_defined rebalancer? Maybe you
>>>>>>> can refer to this test case TestCustomizedIdealStateRebalancer.
>>>>>>>
>>>>>>> Best Regards,
>>>>>>> Jiajun
>>>>>>>
>>>>>>>
>>>>>>> On Fri, Oct 30, 2020 at 10:11 PM Xue Junkai <jx...@apache.org> wrote:
>>>>>>>
>>>>>>>> Hi Leo,
>>>>>>>>
>>>>>>>> What kind of examples are you looking for? Participant side? Or
>>>>>>>> how to build IdealState?
>>>>>>>>
>>>>>>>> Best,
>>>>>>>>
>>>>>>>> Junkai
>>>>>>>>
>>>>>>>> On Fri, Oct 30, 2020 at 9:57 PM Leo, Li <go...@gmail.com>
>>>>>>>> wrote:
>>>>>>>>
>>>>>>>>> Hi,
>>>>>>>>> I am a new user of Helix. I wonder if you could give an example
>>>>>>>>> code of using this user_defined model
>>>>>>>>> https://helix.apache.org/1.0.1-docs/tutorial_user_def_rebalancer.html
>>>>>>>>> ?
>>>>>>>>>
>>>>>>>>> Thank you,
>>>>>>>>> Leo
>>>>>>>>>
>>>>>>>>
Re: User-Defined Rebalancing
Posted by Xue Junkai <jx...@apache.org>.
Yeah. That's the reason. In the test, it uses CustomRebalancer which
extends AbstractRebalancer.
And here's the definition of that:
https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/main/java/org/apache/helix/controller/rebalancer/AbstractRebalancer.java
.
You can follow the implementation of this class and have a try.
best,
Junkai
On Sat, Oct 31, 2020 at 9:53 PM Leo, Li <go...@gmail.com> wrote:
> did not implement MappingCalculator. here is code copied
> from TestCustomizedIdealStateRebalancer. The interface you mentioned seems
> related with
> https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/test/java/org/apache/helix/integration/rebalancer/TestCustomRebalancer.java#L76
> CustomRebalancer.
>
> Thanks.
> Leo
>
> public class LeoDefinedRebalancer implements Rebalancer {
> private static final Logger LOGGER =
> LoggerFactory.getLogger(LeoDefinedRebalancer.class);
>
> @Override
> public void init(HelixManager manager){
> LOGGER.info("LeoDefinedRebalancer.computeNewIdealState setup!");
> }
>
>
> @Override
> public IdealState computeNewIdealState(String resourceName, IdealState currentIdealState,
> CurrentStateOutput currentStateOutput, ClusterDataCache clusterData) {
> LOGGER.info("LeoDefinedRebalancer.computeNewIdealState invoked!");
>
> List<String> liveNodes = Lists.newArrayList(clusterData.getLiveInstances().keySet());
> int i = 0;
> for (String partition : currentIdealState.getPartitionSet()) {
> int index = i++ % liveNodes.size();
> String instance = liveNodes.get(index);
> currentIdealState.getPreferenceList(partition).clear();
> currentIdealState.getPreferenceList(partition).add(instance);
>
> currentIdealState.getInstanceStateMap(partition).clear();
> currentIdealState.getInstanceStateMap(partition).put(instance, "ONLINE");
> LOGGER.info("set ############### {}: {} {} online", resourceName, instance, partition);
> }
> currentIdealState.setReplicas("1");
> return currentIdealState;
> }
> }
>
>
> On Sat, Oct 31, 2020 at 9:39 PM Xue Junkai <jx...@apache.org> wrote:
>
>> What does your class look like? Does it implement the public interface
>> MappingCalculator?This is the interface we casted to get the resource
>> assignment.
>>
>> https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/main/java/org/apache/helix/controller/rebalancer/internal/MappingCalculator.java
>>
>>
>> best,
>>
>> Junkai
>>
>> On Sat, Oct 31, 2020 at 9:30 PM Leo, Li <go...@gmail.com> wrote:
>>
>>> Hi Junkai,
>>>
>>> It's my fault to use a misleading class name. I renamed the class name
>>> as LeoDefinedRebalancer. It is triggered:
>>>
>>> 2778 20:15:18.606 [GenericHelixController-event_process] INFO
>>> c.u.k.w.c.LeoDefinedRebalancer - LeoDefinedRebalancer.computeNewIdealState
>>> invoked!
>>>
>>> 2779 20:15:18.606 [GenericHelixController-event_process] INFO
>>> c.u.k.w.c.LeoDefinedRebalancer - set ############### dca1e:
>>> localhost_12001 dca1e_0 online
>>>
>>> 2780 20:15:18.606 [GenericHelixController-event_process] INFO
>>> c.u.k.w.c.LeoDefinedRebalancer - set ############### dca1e:
>>> localhost_12002 dca1e_1 online
>>>
>>> 2781 20:15:18.606 [GenericHelixController-event_process] WARN
>>> o.a.h.c.s.BestPossibleStateCalcStage - Event 02e76c7b_DEFAULT : Rebalancer
>>> does not have a mapping calculator, defaulting to SEMI_AUT O,
>>> resource: dca1d
>>>
>>>
>>> LeoDefinedRebalancer class is the test case code
>>> <https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/test/java/org/apache/helix/integration/rebalancer/TestCustomizedIdealStateRebalancer.java#L54>
>>> and runs together with other classes, but the issues remain the same.
>>>
>>> HelixManager hMan = startStandAloneController();
>>> LeoDefinedRebalancer rebalancer = new LeoDefinedRebalancer();
>>> rebalancer.init(hMan);
>>>
>>> clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num, STATE_MODEL_NAME);
>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(), LeoDefinedRebalancer.class.getName());
>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCE_MODE.toString(), IdealState.RebalanceMode.USER_DEFINED.toString());
>>> clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
>>>
>>>
>>> On Sat, Oct 31, 2020 at 7:41 PM Xue Junkai <jx...@apache.org> wrote:
>>>
>>>> Thanks for the details Leo.
>>>>
>>>> I think there is a misunderstanding between customized IdealState and
>>>> user defined rebalancer. Actually, they are different levels of concept.
>>>>
>>>> For customized IdealState, it still uses Helix defined customized
>>>> rebalancer, but it takes the map field as the final result. Helix only
>>>> reacts to the map field change for partition/replica assignment.
>>>>
>>>> But user defined rebalancer is different. User has to implement the
>>>> Rebalancer class and include that class in the same JVM when you start the
>>>> controller. So Helix will try to use the class loader to load your class to
>>>> do the computation. So the error could be you dont have the class in the
>>>> same jar of starting the controller. Either you implement Rebalancer class
>>>> and include it in controller jar. Or you can switch to use default auto
>>>> rebalancer but with customized mode for the IdealState.
>>>>
>>>> Jiajun provided a test class as a good example of implementing the user
>>>> defined rebalance. You can take a look if you would like to go that route.
>>>>
>>>> best,
>>>>
>>>> Junkai
>>>>
>>>> On Sat, Oct 31, 2020 at 7:19 PM Leo, Li <go...@gmail.com> wrote:
>>>>
>>>>> Hi JunKai and Jiajun,
>>>>> Thank you for your help. When I tried to use a customized Rebalancer,
>>>>> I saw the rebalancer is triggered and the idea state is set by it.
>>>>>
>>>>> 3784 18:38:38.268 [GenericHelixController-event_process] INFO
>>>>> c.u.k.w.c.CustomizedRebalancer - CustomizedRebalancer.computeNewIdealState
>>>>> invoked!
>>>>>
>>>>> 3785 18:38:38.268 [GenericHelixController-event_process] INFO
>>>>> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
>>>>> localhost_12001 dca1d_0 online
>>>>>
>>>>> 3786 18:38:38.268 [GenericHelixController-event_process] INFO
>>>>> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
>>>>> localhost_12002 dca1d_1 online
>>>>>
>>>>> 3787 18:38:38.269 [GenericHelixController-event_process] WARN
>>>>> o.a.h.c.s.BestPossibleStateCalcStage - Event b4896f24_DEFAULT : Rebalancer
>>>>> does not have a mapping calculator, defaulting to SEMI_AUT O,
>>>>> resource: dca1d
>>>>>
>>>>>
>>>>> However the Helix UI shows "no date to display". The snapshot is
>>>>> attached. The EXTERNALVIEW and IDEASTATE are:
>>>>>
>>>>>
>>>>> >>> get /Watchdog/EXTERNALVIEW/dca1d
>>>>>
>>>>> {
>>>>>
>>>>> "id":"dca1d"
>>>>>
>>>>> ,"simpleFields":{
>>>>>
>>>>> "BUCKET_SIZE":"0"
>>>>>
>>>>> ,"IDEAL_STATE_MODE":"AUTO"
>>>>>
>>>>> ,"NUM_PARTITIONS":"2"
>>>>>
>>>>>
>>>>> ,"REBALANCER_CLASS_NAME":"mypackage.controller.CustomizedRebalancer"
>>>>>
>>>>> ,"REBALANCE_MODE":"USER_DEFINED"
>>>>>
>>>>> ,"REBALANCE_STRATEGY":"DEFAULT"
>>>>>
>>>>> ,"REPLICAS":"1"
>>>>>
>>>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>>>
>>>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>>>
>>>>> }
>>>>>
>>>>> ,"listFields":{
>>>>>
>>>>> }
>>>>>
>>>>> ,"mapFields":{
>>>>>
>>>>> "dca1d_0":{
>>>>>
>>>>> "localhost_12010":"ONLINE"
>>>>>
>>>>> }
>>>>>
>>>>> ,"dca1d_1":{
>>>>>
>>>>> "localhost_12000":"ONLINE"
>>>>>
>>>>> }
>>>>>
>>>>> }
>>>>>
>>>>> }
>>>>>
>>>>>
>>>>> >>> get /Watchdog/IDEALSTATES/dca1d
>>>>>
>>>>> {
>>>>>
>>>>> "id" : "dca1d",
>>>>>
>>>>> "simpleFields" : {
>>>>>
>>>>> "IDEAL_STATE_MODE" : "AUTO",
>>>>>
>>>>> "NUM_PARTITIONS" : "2",
>>>>>
>>>>> "REBALANCER_CLASS_NAME" :
>>>>> "mypackage.controller.CustomizedRebalancer",
>>>>>
>>>>> "REBALANCE_MODE" : "USER_DEFINED",
>>>>>
>>>>> "REBALANCE_STRATEGY" : "DEFAULT",
>>>>>
>>>>> "REPLICAS" : "1",
>>>>>
>>>>> "STATE_MODEL_DEF_REF" : "MyStateModel",
>>>>>
>>>>> "STATE_MODEL_FACTORY_NAME" : "DEFAULT"
>>>>>
>>>>> },
>>>>>
>>>>> "mapFields" : {
>>>>>
>>>>> "dca1d_0" : {
>>>>>
>>>>> },
>>>>>
>>>>> "dca1d_1" : {
>>>>>
>>>>> }
>>>>>
>>>>> },
>>>>>
>>>>> "listFields" : {
>>>>>
>>>>> "dca1d_0" : [ ],
>>>>>
>>>>> "dca1d_1" : [ ]
>>>>>
>>>>> }
>>>>>
>>>>> }
>>>>>
>>>>>
>>>>> Comparing with before, when i use CUSTOMIZED rebalancer and model, The
>>>>> EXTERNALVIEW and IDEASTATE were working well with Helix UI:
>>>>>
>>>>>
>>>>> >>> get /Watchdog/IDEALSTATES/dca1d
>>>>>
>>>>> {
>>>>>
>>>>> "id":"dca1d"
>>>>>
>>>>> ,"simpleFields":{
>>>>>
>>>>> "IDEAL_STATE_MODE":"CUSTOMIZED"
>>>>>
>>>>> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>>>>>
>>>>> ,"NUM_PARTITIONS":"2"
>>>>>
>>>>> ,"REBALANCE_MODE":"CUSTOMIZED"
>>>>>
>>>>> ,"REPLICAS":"1"
>>>>>
>>>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>>>
>>>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>>>
>>>>> }
>>>>>
>>>>> ,"listFields":{
>>>>>
>>>>> }
>>>>>
>>>>> ,"mapFields":{
>>>>>
>>>>> "0":{
>>>>>
>>>>> "localhost_12004":"ONLINE"
>>>>>
>>>>> }
>>>>>
>>>>> ,"1":{
>>>>>
>>>>> "localhost_12005":"ONLINE"
>>>>>
>>>>> }
>>>>>
>>>>> }
>>>>>
>>>>> }
>>>>>
>>>>>
>>>>> >>> get /Watchdog/EXTERNALVIEW/dca1d
>>>>>
>>>>> {
>>>>>
>>>>> "id":"dca1d"
>>>>>
>>>>> ,"simpleFields":{
>>>>>
>>>>> "BUCKET_SIZE":"0"
>>>>>
>>>>> ,"IDEAL_STATE_MODE":"CUSTOMIZED"
>>>>>
>>>>> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>>>>>
>>>>> ,"NUM_PARTITIONS":"2"
>>>>>
>>>>> ,"REBALANCE_MODE":"CUSTOMIZED"
>>>>>
>>>>> ,"REPLICAS":"1"
>>>>>
>>>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>>>
>>>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>>>
>>>>> }
>>>>>
>>>>> ,"listFields":{
>>>>>
>>>>> }
>>>>>
>>>>> ,"mapFields":{
>>>>>
>>>>> "0":{
>>>>>
>>>>> "localhost_12004":"ONLINE"
>>>>>
>>>>> }
>>>>>
>>>>> ,"1":{
>>>>>
>>>>> "localhost_12005":"ONLINE"
>>>>>
>>>>> }
>>>>>
>>>>> }
>>>>>
>>>>> }
>>>>>
>>>>>
>>>>> I see that the mapFields key changed after using "USER_DEFINED"
>>>>> rebalancer.
>>>>>
>>>>> Here are the code i used:
>>>>>
>>>>> public static void addResource(String resource, int num) {
>>>>> clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num, STATE_MODEL_NAME);
>>>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(), CustomizedRebalancer.class.getName());
>>>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCE_MODE.toString(), IdealState.RebalanceMode.USER_DEFINED.toString());
>>>>> clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
>>>>>
>>>>>
>>>>> What is the right way to set a rebalancer?
>>>>>
>>>>>
>>>>> The rebalancer method has a single resource interface. What is the
>>>>> right way to rebalance all resources in the cluster for any event change in
>>>>> the cluster? My concern is the sequence and thread safety.
>>>>>
>>>>> @Override
>>>>> public IdealState computeNewIdealState(String resourceName, IdealState currentIdealState,
>>>>> CurrentStateOutput currentStateOutput, ClusterDataCache clusterData) {
>>>>>
>>>>>
>>>>>
>>>>> Regards,
>>>>>
>>>>> Leo
>>>>>
>>>>>
>>>>>
>>>>> On Fri, Oct 30, 2020 at 10:21 PM Wang Jiajun <er...@gmail.com>
>>>>> wrote:
>>>>>
>>>>>> Hi Leo,
>>>>>>
>>>>>> Are you looking for guidance on how to configure the IdealState so
>>>>>> the resource will be rebalanced by your user_defined rebalancer? Maybe you
>>>>>> can refer to this test case TestCustomizedIdealStateRebalancer.
>>>>>>
>>>>>> Best Regards,
>>>>>> Jiajun
>>>>>>
>>>>>>
>>>>>> On Fri, Oct 30, 2020 at 10:11 PM Xue Junkai <jx...@apache.org> wrote:
>>>>>>
>>>>>>> Hi Leo,
>>>>>>>
>>>>>>> What kind of examples are you looking for? Participant side? Or how
>>>>>>> to build IdealState?
>>>>>>>
>>>>>>> Best,
>>>>>>>
>>>>>>> Junkai
>>>>>>>
>>>>>>> On Fri, Oct 30, 2020 at 9:57 PM Leo, Li <go...@gmail.com> wrote:
>>>>>>>
>>>>>>>> Hi,
>>>>>>>> I am a new user of Helix. I wonder if you could give an example
>>>>>>>> code of using this user_defined model
>>>>>>>> https://helix.apache.org/1.0.1-docs/tutorial_user_def_rebalancer.html
>>>>>>>> ?
>>>>>>>>
>>>>>>>> Thank you,
>>>>>>>> Leo
>>>>>>>>
>>>>>>>
Re: User-Defined Rebalancing
Posted by "Leo, Li" <go...@gmail.com>.
did not implement MappingCalculator. here is code copied
from TestCustomizedIdealStateRebalancer. The interface you mentioned seems
related with
https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/test/java/org/apache/helix/integration/rebalancer/TestCustomRebalancer.java#L76
CustomRebalancer.
Thanks.
Leo
public class LeoDefinedRebalancer implements Rebalancer {
private static final Logger LOGGER =
LoggerFactory.getLogger(LeoDefinedRebalancer.class);
@Override
public void init(HelixManager manager){
LOGGER.info("LeoDefinedRebalancer.computeNewIdealState setup!");
}
@Override
public IdealState computeNewIdealState(String resourceName,
IdealState currentIdealState,
CurrentStateOutput currentStateOutput, ClusterDataCache clusterData) {
LOGGER.info("LeoDefinedRebalancer.computeNewIdealState invoked!");
List<String> liveNodes =
Lists.newArrayList(clusterData.getLiveInstances().keySet());
int i = 0;
for (String partition : currentIdealState.getPartitionSet()) {
int index = i++ % liveNodes.size();
String instance = liveNodes.get(index);
currentIdealState.getPreferenceList(partition).clear();
currentIdealState.getPreferenceList(partition).add(instance);
currentIdealState.getInstanceStateMap(partition).clear();
currentIdealState.getInstanceStateMap(partition).put(instance, "ONLINE");
LOGGER.info("set ############### {}: {} {} online",
resourceName, instance, partition);
}
currentIdealState.setReplicas("1");
return currentIdealState;
}
}
On Sat, Oct 31, 2020 at 9:39 PM Xue Junkai <jx...@apache.org> wrote:
> What does your class look like? Does it implement the public interface
> MappingCalculator?This is the interface we casted to get the resource
> assignment.
>
> https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/main/java/org/apache/helix/controller/rebalancer/internal/MappingCalculator.java
>
>
> best,
>
> Junkai
>
> On Sat, Oct 31, 2020 at 9:30 PM Leo, Li <go...@gmail.com> wrote:
>
>> Hi Junkai,
>>
>> It's my fault to use a misleading class name. I renamed the class name
>> as LeoDefinedRebalancer. It is triggered:
>>
>> 2778 20:15:18.606 [GenericHelixController-event_process] INFO
>> c.u.k.w.c.LeoDefinedRebalancer - LeoDefinedRebalancer.computeNewIdealState
>> invoked!
>>
>> 2779 20:15:18.606 [GenericHelixController-event_process] INFO
>> c.u.k.w.c.LeoDefinedRebalancer - set ############### dca1e:
>> localhost_12001 dca1e_0 online
>>
>> 2780 20:15:18.606 [GenericHelixController-event_process] INFO
>> c.u.k.w.c.LeoDefinedRebalancer - set ############### dca1e:
>> localhost_12002 dca1e_1 online
>>
>> 2781 20:15:18.606 [GenericHelixController-event_process] WARN
>> o.a.h.c.s.BestPossibleStateCalcStage - Event 02e76c7b_DEFAULT : Rebalancer
>> does not have a mapping calculator, defaulting to SEMI_AUT O,
>> resource: dca1d
>>
>>
>> LeoDefinedRebalancer class is the test case code
>> <https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/test/java/org/apache/helix/integration/rebalancer/TestCustomizedIdealStateRebalancer.java#L54>
>> and runs together with other classes, but the issues remain the same.
>>
>> HelixManager hMan = startStandAloneController();
>> LeoDefinedRebalancer rebalancer = new LeoDefinedRebalancer();
>> rebalancer.init(hMan);
>>
>> clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num, STATE_MODEL_NAME);
>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(), LeoDefinedRebalancer.class.getName());
>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCE_MODE.toString(), IdealState.RebalanceMode.USER_DEFINED.toString());
>> clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
>>
>>
>> On Sat, Oct 31, 2020 at 7:41 PM Xue Junkai <jx...@apache.org> wrote:
>>
>>> Thanks for the details Leo.
>>>
>>> I think there is a misunderstanding between customized IdealState and
>>> user defined rebalancer. Actually, they are different levels of concept.
>>>
>>> For customized IdealState, it still uses Helix defined customized
>>> rebalancer, but it takes the map field as the final result. Helix only
>>> reacts to the map field change for partition/replica assignment.
>>>
>>> But user defined rebalancer is different. User has to implement the
>>> Rebalancer class and include that class in the same JVM when you start the
>>> controller. So Helix will try to use the class loader to load your class to
>>> do the computation. So the error could be you dont have the class in the
>>> same jar of starting the controller. Either you implement Rebalancer class
>>> and include it in controller jar. Or you can switch to use default auto
>>> rebalancer but with customized mode for the IdealState.
>>>
>>> Jiajun provided a test class as a good example of implementing the user
>>> defined rebalance. You can take a look if you would like to go that route.
>>>
>>> best,
>>>
>>> Junkai
>>>
>>> On Sat, Oct 31, 2020 at 7:19 PM Leo, Li <go...@gmail.com> wrote:
>>>
>>>> Hi JunKai and Jiajun,
>>>> Thank you for your help. When I tried to use a customized Rebalancer, I
>>>> saw the rebalancer is triggered and the idea state is set by it.
>>>>
>>>> 3784 18:38:38.268 [GenericHelixController-event_process] INFO
>>>> c.u.k.w.c.CustomizedRebalancer - CustomizedRebalancer.computeNewIdealState
>>>> invoked!
>>>>
>>>> 3785 18:38:38.268 [GenericHelixController-event_process] INFO
>>>> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
>>>> localhost_12001 dca1d_0 online
>>>>
>>>> 3786 18:38:38.268 [GenericHelixController-event_process] INFO
>>>> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
>>>> localhost_12002 dca1d_1 online
>>>>
>>>> 3787 18:38:38.269 [GenericHelixController-event_process] WARN
>>>> o.a.h.c.s.BestPossibleStateCalcStage - Event b4896f24_DEFAULT : Rebalancer
>>>> does not have a mapping calculator, defaulting to SEMI_AUT O,
>>>> resource: dca1d
>>>>
>>>>
>>>> However the Helix UI shows "no date to display". The snapshot is
>>>> attached. The EXTERNALVIEW and IDEASTATE are:
>>>>
>>>>
>>>> >>> get /Watchdog/EXTERNALVIEW/dca1d
>>>>
>>>> {
>>>>
>>>> "id":"dca1d"
>>>>
>>>> ,"simpleFields":{
>>>>
>>>> "BUCKET_SIZE":"0"
>>>>
>>>> ,"IDEAL_STATE_MODE":"AUTO"
>>>>
>>>> ,"NUM_PARTITIONS":"2"
>>>>
>>>> ,"REBALANCER_CLASS_NAME":"mypackage.controller.CustomizedRebalancer"
>>>>
>>>> ,"REBALANCE_MODE":"USER_DEFINED"
>>>>
>>>> ,"REBALANCE_STRATEGY":"DEFAULT"
>>>>
>>>> ,"REPLICAS":"1"
>>>>
>>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>>
>>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>>
>>>> }
>>>>
>>>> ,"listFields":{
>>>>
>>>> }
>>>>
>>>> ,"mapFields":{
>>>>
>>>> "dca1d_0":{
>>>>
>>>> "localhost_12010":"ONLINE"
>>>>
>>>> }
>>>>
>>>> ,"dca1d_1":{
>>>>
>>>> "localhost_12000":"ONLINE"
>>>>
>>>> }
>>>>
>>>> }
>>>>
>>>> }
>>>>
>>>>
>>>> >>> get /Watchdog/IDEALSTATES/dca1d
>>>>
>>>> {
>>>>
>>>> "id" : "dca1d",
>>>>
>>>> "simpleFields" : {
>>>>
>>>> "IDEAL_STATE_MODE" : "AUTO",
>>>>
>>>> "NUM_PARTITIONS" : "2",
>>>>
>>>> "REBALANCER_CLASS_NAME" :
>>>> "mypackage.controller.CustomizedRebalancer",
>>>>
>>>> "REBALANCE_MODE" : "USER_DEFINED",
>>>>
>>>> "REBALANCE_STRATEGY" : "DEFAULT",
>>>>
>>>> "REPLICAS" : "1",
>>>>
>>>> "STATE_MODEL_DEF_REF" : "MyStateModel",
>>>>
>>>> "STATE_MODEL_FACTORY_NAME" : "DEFAULT"
>>>>
>>>> },
>>>>
>>>> "mapFields" : {
>>>>
>>>> "dca1d_0" : {
>>>>
>>>> },
>>>>
>>>> "dca1d_1" : {
>>>>
>>>> }
>>>>
>>>> },
>>>>
>>>> "listFields" : {
>>>>
>>>> "dca1d_0" : [ ],
>>>>
>>>> "dca1d_1" : [ ]
>>>>
>>>> }
>>>>
>>>> }
>>>>
>>>>
>>>> Comparing with before, when i use CUSTOMIZED rebalancer and model, The
>>>> EXTERNALVIEW and IDEASTATE were working well with Helix UI:
>>>>
>>>>
>>>> >>> get /Watchdog/IDEALSTATES/dca1d
>>>>
>>>> {
>>>>
>>>> "id":"dca1d"
>>>>
>>>> ,"simpleFields":{
>>>>
>>>> "IDEAL_STATE_MODE":"CUSTOMIZED"
>>>>
>>>> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>>>>
>>>> ,"NUM_PARTITIONS":"2"
>>>>
>>>> ,"REBALANCE_MODE":"CUSTOMIZED"
>>>>
>>>> ,"REPLICAS":"1"
>>>>
>>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>>
>>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>>
>>>> }
>>>>
>>>> ,"listFields":{
>>>>
>>>> }
>>>>
>>>> ,"mapFields":{
>>>>
>>>> "0":{
>>>>
>>>> "localhost_12004":"ONLINE"
>>>>
>>>> }
>>>>
>>>> ,"1":{
>>>>
>>>> "localhost_12005":"ONLINE"
>>>>
>>>> }
>>>>
>>>> }
>>>>
>>>> }
>>>>
>>>>
>>>> >>> get /Watchdog/EXTERNALVIEW/dca1d
>>>>
>>>> {
>>>>
>>>> "id":"dca1d"
>>>>
>>>> ,"simpleFields":{
>>>>
>>>> "BUCKET_SIZE":"0"
>>>>
>>>> ,"IDEAL_STATE_MODE":"CUSTOMIZED"
>>>>
>>>> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>>>>
>>>> ,"NUM_PARTITIONS":"2"
>>>>
>>>> ,"REBALANCE_MODE":"CUSTOMIZED"
>>>>
>>>> ,"REPLICAS":"1"
>>>>
>>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>>
>>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>>
>>>> }
>>>>
>>>> ,"listFields":{
>>>>
>>>> }
>>>>
>>>> ,"mapFields":{
>>>>
>>>> "0":{
>>>>
>>>> "localhost_12004":"ONLINE"
>>>>
>>>> }
>>>>
>>>> ,"1":{
>>>>
>>>> "localhost_12005":"ONLINE"
>>>>
>>>> }
>>>>
>>>> }
>>>>
>>>> }
>>>>
>>>>
>>>> I see that the mapFields key changed after using "USER_DEFINED"
>>>> rebalancer.
>>>>
>>>> Here are the code i used:
>>>>
>>>> public static void addResource(String resource, int num) {
>>>> clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num, STATE_MODEL_NAME);
>>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(), CustomizedRebalancer.class.getName());
>>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCE_MODE.toString(), IdealState.RebalanceMode.USER_DEFINED.toString());
>>>> clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
>>>>
>>>>
>>>> What is the right way to set a rebalancer?
>>>>
>>>>
>>>> The rebalancer method has a single resource interface. What is the
>>>> right way to rebalance all resources in the cluster for any event change in
>>>> the cluster? My concern is the sequence and thread safety.
>>>>
>>>> @Override
>>>> public IdealState computeNewIdealState(String resourceName, IdealState currentIdealState,
>>>> CurrentStateOutput currentStateOutput, ClusterDataCache clusterData) {
>>>>
>>>>
>>>>
>>>> Regards,
>>>>
>>>> Leo
>>>>
>>>>
>>>>
>>>> On Fri, Oct 30, 2020 at 10:21 PM Wang Jiajun <er...@gmail.com>
>>>> wrote:
>>>>
>>>>> Hi Leo,
>>>>>
>>>>> Are you looking for guidance on how to configure the IdealState so the
>>>>> resource will be rebalanced by your user_defined rebalancer? Maybe you can
>>>>> refer to this test case TestCustomizedIdealStateRebalancer.
>>>>>
>>>>> Best Regards,
>>>>> Jiajun
>>>>>
>>>>>
>>>>> On Fri, Oct 30, 2020 at 10:11 PM Xue Junkai <jx...@apache.org> wrote:
>>>>>
>>>>>> Hi Leo,
>>>>>>
>>>>>> What kind of examples are you looking for? Participant side? Or how
>>>>>> to build IdealState?
>>>>>>
>>>>>> Best,
>>>>>>
>>>>>> Junkai
>>>>>>
>>>>>> On Fri, Oct 30, 2020 at 9:57 PM Leo, Li <go...@gmail.com> wrote:
>>>>>>
>>>>>>> Hi,
>>>>>>> I am a new user of Helix. I wonder if you could give an example
>>>>>>> code of using this user_defined model
>>>>>>> https://helix.apache.org/1.0.1-docs/tutorial_user_def_rebalancer.html
>>>>>>> ?
>>>>>>>
>>>>>>> Thank you,
>>>>>>> Leo
>>>>>>>
>>>>>>
Re: User-Defined Rebalancing
Posted by Xue Junkai <jx...@apache.org>.
What does your class look like? Does it implement the public interface
MappingCalculator?This is the interface we casted to get the resource
assignment.
https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/main/java/org/apache/helix/controller/rebalancer/internal/MappingCalculator.java
best,
Junkai
On Sat, Oct 31, 2020 at 9:30 PM Leo, Li <go...@gmail.com> wrote:
> Hi Junkai,
>
> It's my fault to use a misleading class name. I renamed the class name as
> LeoDefinedRebalancer. It is triggered:
>
> 2778 20:15:18.606 [GenericHelixController-event_process] INFO
> c.u.k.w.c.LeoDefinedRebalancer - LeoDefinedRebalancer.computeNewIdealState
> invoked!
>
> 2779 20:15:18.606 [GenericHelixController-event_process] INFO
> c.u.k.w.c.LeoDefinedRebalancer - set ############### dca1e:
> localhost_12001 dca1e_0 online
>
> 2780 20:15:18.606 [GenericHelixController-event_process] INFO
> c.u.k.w.c.LeoDefinedRebalancer - set ############### dca1e:
> localhost_12002 dca1e_1 online
>
> 2781 20:15:18.606 [GenericHelixController-event_process] WARN
> o.a.h.c.s.BestPossibleStateCalcStage - Event 02e76c7b_DEFAULT : Rebalancer
> does not have a mapping calculator, defaulting to SEMI_AUT O,
> resource: dca1d
>
>
> LeoDefinedRebalancer class is the test case code
> <https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/test/java/org/apache/helix/integration/rebalancer/TestCustomizedIdealStateRebalancer.java#L54>
> and runs together with other classes, but the issues remain the same.
>
> HelixManager hMan = startStandAloneController();
> LeoDefinedRebalancer rebalancer = new LeoDefinedRebalancer();
> rebalancer.init(hMan);
>
> clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num, STATE_MODEL_NAME);
> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(), LeoDefinedRebalancer.class.getName());
> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCE_MODE.toString(), IdealState.RebalanceMode.USER_DEFINED.toString());
> clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
>
>
> On Sat, Oct 31, 2020 at 7:41 PM Xue Junkai <jx...@apache.org> wrote:
>
>> Thanks for the details Leo.
>>
>> I think there is a misunderstanding between customized IdealState and
>> user defined rebalancer. Actually, they are different levels of concept.
>>
>> For customized IdealState, it still uses Helix defined customized
>> rebalancer, but it takes the map field as the final result. Helix only
>> reacts to the map field change for partition/replica assignment.
>>
>> But user defined rebalancer is different. User has to implement the
>> Rebalancer class and include that class in the same JVM when you start the
>> controller. So Helix will try to use the class loader to load your class to
>> do the computation. So the error could be you dont have the class in the
>> same jar of starting the controller. Either you implement Rebalancer class
>> and include it in controller jar. Or you can switch to use default auto
>> rebalancer but with customized mode for the IdealState.
>>
>> Jiajun provided a test class as a good example of implementing the user
>> defined rebalance. You can take a look if you would like to go that route.
>>
>> best,
>>
>> Junkai
>>
>> On Sat, Oct 31, 2020 at 7:19 PM Leo, Li <go...@gmail.com> wrote:
>>
>>> Hi JunKai and Jiajun,
>>> Thank you for your help. When I tried to use a customized Rebalancer, I
>>> saw the rebalancer is triggered and the idea state is set by it.
>>>
>>> 3784 18:38:38.268 [GenericHelixController-event_process] INFO
>>> c.u.k.w.c.CustomizedRebalancer - CustomizedRebalancer.computeNewIdealState
>>> invoked!
>>>
>>> 3785 18:38:38.268 [GenericHelixController-event_process] INFO
>>> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
>>> localhost_12001 dca1d_0 online
>>>
>>> 3786 18:38:38.268 [GenericHelixController-event_process] INFO
>>> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
>>> localhost_12002 dca1d_1 online
>>>
>>> 3787 18:38:38.269 [GenericHelixController-event_process] WARN
>>> o.a.h.c.s.BestPossibleStateCalcStage - Event b4896f24_DEFAULT : Rebalancer
>>> does not have a mapping calculator, defaulting to SEMI_AUT O,
>>> resource: dca1d
>>>
>>>
>>> However the Helix UI shows "no date to display". The snapshot is
>>> attached. The EXTERNALVIEW and IDEASTATE are:
>>>
>>>
>>> >>> get /Watchdog/EXTERNALVIEW/dca1d
>>>
>>> {
>>>
>>> "id":"dca1d"
>>>
>>> ,"simpleFields":{
>>>
>>> "BUCKET_SIZE":"0"
>>>
>>> ,"IDEAL_STATE_MODE":"AUTO"
>>>
>>> ,"NUM_PARTITIONS":"2"
>>>
>>> ,"REBALANCER_CLASS_NAME":"mypackage.controller.CustomizedRebalancer"
>>>
>>> ,"REBALANCE_MODE":"USER_DEFINED"
>>>
>>> ,"REBALANCE_STRATEGY":"DEFAULT"
>>>
>>> ,"REPLICAS":"1"
>>>
>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>
>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>
>>> }
>>>
>>> ,"listFields":{
>>>
>>> }
>>>
>>> ,"mapFields":{
>>>
>>> "dca1d_0":{
>>>
>>> "localhost_12010":"ONLINE"
>>>
>>> }
>>>
>>> ,"dca1d_1":{
>>>
>>> "localhost_12000":"ONLINE"
>>>
>>> }
>>>
>>> }
>>>
>>> }
>>>
>>>
>>> >>> get /Watchdog/IDEALSTATES/dca1d
>>>
>>> {
>>>
>>> "id" : "dca1d",
>>>
>>> "simpleFields" : {
>>>
>>> "IDEAL_STATE_MODE" : "AUTO",
>>>
>>> "NUM_PARTITIONS" : "2",
>>>
>>> "REBALANCER_CLASS_NAME" :
>>> "mypackage.controller.CustomizedRebalancer",
>>>
>>> "REBALANCE_MODE" : "USER_DEFINED",
>>>
>>> "REBALANCE_STRATEGY" : "DEFAULT",
>>>
>>> "REPLICAS" : "1",
>>>
>>> "STATE_MODEL_DEF_REF" : "MyStateModel",
>>>
>>> "STATE_MODEL_FACTORY_NAME" : "DEFAULT"
>>>
>>> },
>>>
>>> "mapFields" : {
>>>
>>> "dca1d_0" : {
>>>
>>> },
>>>
>>> "dca1d_1" : {
>>>
>>> }
>>>
>>> },
>>>
>>> "listFields" : {
>>>
>>> "dca1d_0" : [ ],
>>>
>>> "dca1d_1" : [ ]
>>>
>>> }
>>>
>>> }
>>>
>>>
>>> Comparing with before, when i use CUSTOMIZED rebalancer and model, The
>>> EXTERNALVIEW and IDEASTATE were working well with Helix UI:
>>>
>>>
>>> >>> get /Watchdog/IDEALSTATES/dca1d
>>>
>>> {
>>>
>>> "id":"dca1d"
>>>
>>> ,"simpleFields":{
>>>
>>> "IDEAL_STATE_MODE":"CUSTOMIZED"
>>>
>>> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>>>
>>> ,"NUM_PARTITIONS":"2"
>>>
>>> ,"REBALANCE_MODE":"CUSTOMIZED"
>>>
>>> ,"REPLICAS":"1"
>>>
>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>
>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>
>>> }
>>>
>>> ,"listFields":{
>>>
>>> }
>>>
>>> ,"mapFields":{
>>>
>>> "0":{
>>>
>>> "localhost_12004":"ONLINE"
>>>
>>> }
>>>
>>> ,"1":{
>>>
>>> "localhost_12005":"ONLINE"
>>>
>>> }
>>>
>>> }
>>>
>>> }
>>>
>>>
>>> >>> get /Watchdog/EXTERNALVIEW/dca1d
>>>
>>> {
>>>
>>> "id":"dca1d"
>>>
>>> ,"simpleFields":{
>>>
>>> "BUCKET_SIZE":"0"
>>>
>>> ,"IDEAL_STATE_MODE":"CUSTOMIZED"
>>>
>>> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>>>
>>> ,"NUM_PARTITIONS":"2"
>>>
>>> ,"REBALANCE_MODE":"CUSTOMIZED"
>>>
>>> ,"REPLICAS":"1"
>>>
>>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>>
>>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>>
>>> }
>>>
>>> ,"listFields":{
>>>
>>> }
>>>
>>> ,"mapFields":{
>>>
>>> "0":{
>>>
>>> "localhost_12004":"ONLINE"
>>>
>>> }
>>>
>>> ,"1":{
>>>
>>> "localhost_12005":"ONLINE"
>>>
>>> }
>>>
>>> }
>>>
>>> }
>>>
>>>
>>> I see that the mapFields key changed after using "USER_DEFINED"
>>> rebalancer.
>>>
>>> Here are the code i used:
>>>
>>> public static void addResource(String resource, int num) {
>>> clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num, STATE_MODEL_NAME);
>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(), CustomizedRebalancer.class.getName());
>>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCE_MODE.toString(), IdealState.RebalanceMode.USER_DEFINED.toString());
>>> clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
>>>
>>>
>>> What is the right way to set a rebalancer?
>>>
>>>
>>> The rebalancer method has a single resource interface. What is the right
>>> way to rebalance all resources in the cluster for any event change in the
>>> cluster? My concern is the sequence and thread safety.
>>>
>>> @Override
>>> public IdealState computeNewIdealState(String resourceName, IdealState currentIdealState,
>>> CurrentStateOutput currentStateOutput, ClusterDataCache clusterData) {
>>>
>>>
>>>
>>> Regards,
>>>
>>> Leo
>>>
>>>
>>>
>>> On Fri, Oct 30, 2020 at 10:21 PM Wang Jiajun <er...@gmail.com>
>>> wrote:
>>>
>>>> Hi Leo,
>>>>
>>>> Are you looking for guidance on how to configure the IdealState so the
>>>> resource will be rebalanced by your user_defined rebalancer? Maybe you can
>>>> refer to this test case TestCustomizedIdealStateRebalancer.
>>>>
>>>> Best Regards,
>>>> Jiajun
>>>>
>>>>
>>>> On Fri, Oct 30, 2020 at 10:11 PM Xue Junkai <jx...@apache.org> wrote:
>>>>
>>>>> Hi Leo,
>>>>>
>>>>> What kind of examples are you looking for? Participant side? Or how
>>>>> to build IdealState?
>>>>>
>>>>> Best,
>>>>>
>>>>> Junkai
>>>>>
>>>>> On Fri, Oct 30, 2020 at 9:57 PM Leo, Li <go...@gmail.com> wrote:
>>>>>
>>>>>> Hi,
>>>>>> I am a new user of Helix. I wonder if you could give an example code
>>>>>> of using this user_defined model
>>>>>> https://helix.apache.org/1.0.1-docs/tutorial_user_def_rebalancer.html
>>>>>> ?
>>>>>>
>>>>>> Thank you,
>>>>>> Leo
>>>>>>
>>>>>
Re: User-Defined Rebalancing
Posted by "Leo, Li" <go...@gmail.com>.
Hi Junkai,
It's my fault to use a misleading class name. I renamed the class name as
LeoDefinedRebalancer. It is triggered:
2778 20:15:18.606 [GenericHelixController-event_process] INFO
c.u.k.w.c.LeoDefinedRebalancer - LeoDefinedRebalancer.computeNewIdealState
invoked!
2779 20:15:18.606 [GenericHelixController-event_process] INFO
c.u.k.w.c.LeoDefinedRebalancer - set ############### dca1e:
localhost_12001 dca1e_0 online
2780 20:15:18.606 [GenericHelixController-event_process] INFO
c.u.k.w.c.LeoDefinedRebalancer - set ############### dca1e:
localhost_12002 dca1e_1 online
2781 20:15:18.606 [GenericHelixController-event_process] WARN
o.a.h.c.s.BestPossibleStateCalcStage - Event 02e76c7b_DEFAULT : Rebalancer
does not have a mapping calculator, defaulting to SEMI_AUT O, resource:
dca1d
LeoDefinedRebalancer class is the test case code
<https://github.com/apache/helix/blob/a5cddd4b8fa51b1ea53b2f7082daf79361cd81de/helix-core/src/test/java/org/apache/helix/integration/rebalancer/TestCustomizedIdealStateRebalancer.java#L54>
and runs together with other classes, but the issues remain the same.
HelixManager hMan = startStandAloneController();
LeoDefinedRebalancer rebalancer = new LeoDefinedRebalancer();
rebalancer.init(hMan);
clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num,
STATE_MODEL_NAME);
clusterSetup.addResourceProperty(CLUSTER_NAME, resource,
IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(),
LeoDefinedRebalancer.class.getName());
clusterSetup.addResourceProperty(CLUSTER_NAME, resource,
IdealState.IdealStateProperty.REBALANCE_MODE.toString(),
IdealState.RebalanceMode.USER_DEFINED.toString());
clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
On Sat, Oct 31, 2020 at 7:41 PM Xue Junkai <jx...@apache.org> wrote:
> Thanks for the details Leo.
>
> I think there is a misunderstanding between customized IdealState and user
> defined rebalancer. Actually, they are different levels of concept.
>
> For customized IdealState, it still uses Helix defined customized
> rebalancer, but it takes the map field as the final result. Helix only
> reacts to the map field change for partition/replica assignment.
>
> But user defined rebalancer is different. User has to implement the
> Rebalancer class and include that class in the same JVM when you start the
> controller. So Helix will try to use the class loader to load your class to
> do the computation. So the error could be you dont have the class in the
> same jar of starting the controller. Either you implement Rebalancer class
> and include it in controller jar. Or you can switch to use default auto
> rebalancer but with customized mode for the IdealState.
>
> Jiajun provided a test class as a good example of implementing the user
> defined rebalance. You can take a look if you would like to go that route.
>
> best,
>
> Junkai
>
> On Sat, Oct 31, 2020 at 7:19 PM Leo, Li <go...@gmail.com> wrote:
>
>> Hi JunKai and Jiajun,
>> Thank you for your help. When I tried to use a customized Rebalancer, I
>> saw the rebalancer is triggered and the idea state is set by it.
>>
>> 3784 18:38:38.268 [GenericHelixController-event_process] INFO
>> c.u.k.w.c.CustomizedRebalancer - CustomizedRebalancer.computeNewIdealState
>> invoked!
>>
>> 3785 18:38:38.268 [GenericHelixController-event_process] INFO
>> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
>> localhost_12001 dca1d_0 online
>>
>> 3786 18:38:38.268 [GenericHelixController-event_process] INFO
>> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
>> localhost_12002 dca1d_1 online
>>
>> 3787 18:38:38.269 [GenericHelixController-event_process] WARN
>> o.a.h.c.s.BestPossibleStateCalcStage - Event b4896f24_DEFAULT : Rebalancer
>> does not have a mapping calculator, defaulting to SEMI_AUT O,
>> resource: dca1d
>>
>>
>> However the Helix UI shows "no date to display". The snapshot is
>> attached. The EXTERNALVIEW and IDEASTATE are:
>>
>>
>> >>> get /Watchdog/EXTERNALVIEW/dca1d
>>
>> {
>>
>> "id":"dca1d"
>>
>> ,"simpleFields":{
>>
>> "BUCKET_SIZE":"0"
>>
>> ,"IDEAL_STATE_MODE":"AUTO"
>>
>> ,"NUM_PARTITIONS":"2"
>>
>> ,"REBALANCER_CLASS_NAME":"mypackage.controller.CustomizedRebalancer"
>>
>> ,"REBALANCE_MODE":"USER_DEFINED"
>>
>> ,"REBALANCE_STRATEGY":"DEFAULT"
>>
>> ,"REPLICAS":"1"
>>
>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>
>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>
>> }
>>
>> ,"listFields":{
>>
>> }
>>
>> ,"mapFields":{
>>
>> "dca1d_0":{
>>
>> "localhost_12010":"ONLINE"
>>
>> }
>>
>> ,"dca1d_1":{
>>
>> "localhost_12000":"ONLINE"
>>
>> }
>>
>> }
>>
>> }
>>
>>
>> >>> get /Watchdog/IDEALSTATES/dca1d
>>
>> {
>>
>> "id" : "dca1d",
>>
>> "simpleFields" : {
>>
>> "IDEAL_STATE_MODE" : "AUTO",
>>
>> "NUM_PARTITIONS" : "2",
>>
>> "REBALANCER_CLASS_NAME" : "mypackage.controller.CustomizedRebalancer",
>>
>> "REBALANCE_MODE" : "USER_DEFINED",
>>
>> "REBALANCE_STRATEGY" : "DEFAULT",
>>
>> "REPLICAS" : "1",
>>
>> "STATE_MODEL_DEF_REF" : "MyStateModel",
>>
>> "STATE_MODEL_FACTORY_NAME" : "DEFAULT"
>>
>> },
>>
>> "mapFields" : {
>>
>> "dca1d_0" : {
>>
>> },
>>
>> "dca1d_1" : {
>>
>> }
>>
>> },
>>
>> "listFields" : {
>>
>> "dca1d_0" : [ ],
>>
>> "dca1d_1" : [ ]
>>
>> }
>>
>> }
>>
>>
>> Comparing with before, when i use CUSTOMIZED rebalancer and model, The
>> EXTERNALVIEW and IDEASTATE were working well with Helix UI:
>>
>>
>> >>> get /Watchdog/IDEALSTATES/dca1d
>>
>> {
>>
>> "id":"dca1d"
>>
>> ,"simpleFields":{
>>
>> "IDEAL_STATE_MODE":"CUSTOMIZED"
>>
>> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>>
>> ,"NUM_PARTITIONS":"2"
>>
>> ,"REBALANCE_MODE":"CUSTOMIZED"
>>
>> ,"REPLICAS":"1"
>>
>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>
>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>
>> }
>>
>> ,"listFields":{
>>
>> }
>>
>> ,"mapFields":{
>>
>> "0":{
>>
>> "localhost_12004":"ONLINE"
>>
>> }
>>
>> ,"1":{
>>
>> "localhost_12005":"ONLINE"
>>
>> }
>>
>> }
>>
>> }
>>
>>
>> >>> get /Watchdog/EXTERNALVIEW/dca1d
>>
>> {
>>
>> "id":"dca1d"
>>
>> ,"simpleFields":{
>>
>> "BUCKET_SIZE":"0"
>>
>> ,"IDEAL_STATE_MODE":"CUSTOMIZED"
>>
>> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>>
>> ,"NUM_PARTITIONS":"2"
>>
>> ,"REBALANCE_MODE":"CUSTOMIZED"
>>
>> ,"REPLICAS":"1"
>>
>> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>>
>> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>>
>> }
>>
>> ,"listFields":{
>>
>> }
>>
>> ,"mapFields":{
>>
>> "0":{
>>
>> "localhost_12004":"ONLINE"
>>
>> }
>>
>> ,"1":{
>>
>> "localhost_12005":"ONLINE"
>>
>> }
>>
>> }
>>
>> }
>>
>>
>> I see that the mapFields key changed after using "USER_DEFINED"
>> rebalancer.
>>
>> Here are the code i used:
>>
>> public static void addResource(String resource, int num) {
>> clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num, STATE_MODEL_NAME);
>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(), CustomizedRebalancer.class.getName());
>> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCE_MODE.toString(), IdealState.RebalanceMode.USER_DEFINED.toString());
>> clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
>>
>>
>> What is the right way to set a rebalancer?
>>
>>
>> The rebalancer method has a single resource interface. What is the right
>> way to rebalance all resources in the cluster for any event change in the
>> cluster? My concern is the sequence and thread safety.
>>
>> @Override
>> public IdealState computeNewIdealState(String resourceName, IdealState currentIdealState,
>> CurrentStateOutput currentStateOutput, ClusterDataCache clusterData) {
>>
>>
>>
>> Regards,
>>
>> Leo
>>
>>
>>
>> On Fri, Oct 30, 2020 at 10:21 PM Wang Jiajun <er...@gmail.com>
>> wrote:
>>
>>> Hi Leo,
>>>
>>> Are you looking for guidance on how to configure the IdealState so the
>>> resource will be rebalanced by your user_defined rebalancer? Maybe you can
>>> refer to this test case TestCustomizedIdealStateRebalancer.
>>>
>>> Best Regards,
>>> Jiajun
>>>
>>>
>>> On Fri, Oct 30, 2020 at 10:11 PM Xue Junkai <jx...@apache.org> wrote:
>>>
>>>> Hi Leo,
>>>>
>>>> What kind of examples are you looking for? Participant side? Or how to
>>>> build IdealState?
>>>>
>>>> Best,
>>>>
>>>> Junkai
>>>>
>>>> On Fri, Oct 30, 2020 at 9:57 PM Leo, Li <go...@gmail.com> wrote:
>>>>
>>>>> Hi,
>>>>> I am a new user of Helix. I wonder if you could give an example code
>>>>> of using this user_defined model
>>>>> https://helix.apache.org/1.0.1-docs/tutorial_user_def_rebalancer.html
>>>>> ?
>>>>>
>>>>> Thank you,
>>>>> Leo
>>>>>
>>>>
Re: User-Defined Rebalancing
Posted by Xue Junkai <jx...@apache.org>.
Thanks for the details Leo.
I think there is a misunderstanding between customized IdealState and user
defined rebalancer. Actually, they are different levels of concept.
For customized IdealState, it still uses Helix defined customized
rebalancer, but it takes the map field as the final result. Helix only
reacts to the map field change for partition/replica assignment.
But user defined rebalancer is different. User has to implement the
Rebalancer class and include that class in the same JVM when you start the
controller. So Helix will try to use the class loader to load your class to
do the computation. So the error could be you dont have the class in the
same jar of starting the controller. Either you implement Rebalancer class
and include it in controller jar. Or you can switch to use default auto
rebalancer but with customized mode for the IdealState.
Jiajun provided a test class as a good example of implementing the user
defined rebalance. You can take a look if you would like to go that route.
best,
Junkai
On Sat, Oct 31, 2020 at 7:19 PM Leo, Li <go...@gmail.com> wrote:
> Hi JunKai and Jiajun,
> Thank you for your help. When I tried to use a customized Rebalancer, I
> saw the rebalancer is triggered and the idea state is set by it.
>
> 3784 18:38:38.268 [GenericHelixController-event_process] INFO
> c.u.k.w.c.CustomizedRebalancer - CustomizedRebalancer.computeNewIdealState
> invoked!
>
> 3785 18:38:38.268 [GenericHelixController-event_process] INFO
> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
> localhost_12001 dca1d_0 online
>
> 3786 18:38:38.268 [GenericHelixController-event_process] INFO
> c.u.k.w.c.CustomizedRebalancer - set ############### dca1d:
> localhost_12002 dca1d_1 online
>
> 3787 18:38:38.269 [GenericHelixController-event_process] WARN
> o.a.h.c.s.BestPossibleStateCalcStage - Event b4896f24_DEFAULT : Rebalancer
> does not have a mapping calculator, defaulting to SEMI_AUT O,
> resource: dca1d
>
>
> However the Helix UI shows "no date to display". The snapshot is attached.
> The EXTERNALVIEW and IDEASTATE are:
>
>
> >>> get /Watchdog/EXTERNALVIEW/dca1d
>
> {
>
> "id":"dca1d"
>
> ,"simpleFields":{
>
> "BUCKET_SIZE":"0"
>
> ,"IDEAL_STATE_MODE":"AUTO"
>
> ,"NUM_PARTITIONS":"2"
>
> ,"REBALANCER_CLASS_NAME":"mypackage.controller.CustomizedRebalancer"
>
> ,"REBALANCE_MODE":"USER_DEFINED"
>
> ,"REBALANCE_STRATEGY":"DEFAULT"
>
> ,"REPLICAS":"1"
>
> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>
> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>
> }
>
> ,"listFields":{
>
> }
>
> ,"mapFields":{
>
> "dca1d_0":{
>
> "localhost_12010":"ONLINE"
>
> }
>
> ,"dca1d_1":{
>
> "localhost_12000":"ONLINE"
>
> }
>
> }
>
> }
>
>
> >>> get /Watchdog/IDEALSTATES/dca1d
>
> {
>
> "id" : "dca1d",
>
> "simpleFields" : {
>
> "IDEAL_STATE_MODE" : "AUTO",
>
> "NUM_PARTITIONS" : "2",
>
> "REBALANCER_CLASS_NAME" : "mypackage.controller.CustomizedRebalancer",
>
> "REBALANCE_MODE" : "USER_DEFINED",
>
> "REBALANCE_STRATEGY" : "DEFAULT",
>
> "REPLICAS" : "1",
>
> "STATE_MODEL_DEF_REF" : "MyStateModel",
>
> "STATE_MODEL_FACTORY_NAME" : "DEFAULT"
>
> },
>
> "mapFields" : {
>
> "dca1d_0" : {
>
> },
>
> "dca1d_1" : {
>
> }
>
> },
>
> "listFields" : {
>
> "dca1d_0" : [ ],
>
> "dca1d_1" : [ ]
>
> }
>
> }
>
>
> Comparing with before, when i use CUSTOMIZED rebalancer and model, The
> EXTERNALVIEW and IDEASTATE were working well with Helix UI:
>
>
> >>> get /Watchdog/IDEALSTATES/dca1d
>
> {
>
> "id":"dca1d"
>
> ,"simpleFields":{
>
> "IDEAL_STATE_MODE":"CUSTOMIZED"
>
> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>
> ,"NUM_PARTITIONS":"2"
>
> ,"REBALANCE_MODE":"CUSTOMIZED"
>
> ,"REPLICAS":"1"
>
> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>
> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>
> }
>
> ,"listFields":{
>
> }
>
> ,"mapFields":{
>
> "0":{
>
> "localhost_12004":"ONLINE"
>
> }
>
> ,"1":{
>
> "localhost_12005":"ONLINE"
>
> }
>
> }
>
> }
>
>
> >>> get /Watchdog/EXTERNALVIEW/dca1d
>
> {
>
> "id":"dca1d"
>
> ,"simpleFields":{
>
> "BUCKET_SIZE":"0"
>
> ,"IDEAL_STATE_MODE":"CUSTOMIZED"
>
> ,"MAX_PARTITIONS_PER_INSTANCE":"1"
>
> ,"NUM_PARTITIONS":"2"
>
> ,"REBALANCE_MODE":"CUSTOMIZED"
>
> ,"REPLICAS":"1"
>
> ,"STATE_MODEL_DEF_REF":"MyStateModel"
>
> ,"STATE_MODEL_FACTORY_NAME":"DEFAULT"
>
> }
>
> ,"listFields":{
>
> }
>
> ,"mapFields":{
>
> "0":{
>
> "localhost_12004":"ONLINE"
>
> }
>
> ,"1":{
>
> "localhost_12005":"ONLINE"
>
> }
>
> }
>
> }
>
>
> I see that the mapFields key changed after using "USER_DEFINED"
> rebalancer.
>
> Here are the code i used:
>
> public static void addResource(String resource, int num) {
> clusterSetup.addResourceToCluster(CLUSTER_NAME, resource, num, STATE_MODEL_NAME);
> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCER_CLASS_NAME.toString(), CustomizedRebalancer.class.getName());
> clusterSetup.addResourceProperty(CLUSTER_NAME, resource, IdealState.IdealStateProperty.REBALANCE_MODE.toString(), IdealState.RebalanceMode.USER_DEFINED.toString());
> clusterSetup.rebalanceStorageCluster(CLUSTER_NAME, resource, 1, "");
>
>
> What is the right way to set a rebalancer?
>
>
> The rebalancer method has a single resource interface. What is the right
> way to rebalance all resources in the cluster for any event change in the
> cluster? My concern is the sequence and thread safety.
>
> @Override
> public IdealState computeNewIdealState(String resourceName, IdealState currentIdealState,
> CurrentStateOutput currentStateOutput, ClusterDataCache clusterData) {
>
>
>
> Regards,
>
> Leo
>
>
>
> On Fri, Oct 30, 2020 at 10:21 PM Wang Jiajun <er...@gmail.com>
> wrote:
>
>> Hi Leo,
>>
>> Are you looking for guidance on how to configure the IdealState so the
>> resource will be rebalanced by your user_defined rebalancer? Maybe you can
>> refer to this test case TestCustomizedIdealStateRebalancer.
>>
>> Best Regards,
>> Jiajun
>>
>>
>> On Fri, Oct 30, 2020 at 10:11 PM Xue Junkai <jx...@apache.org> wrote:
>>
>>> Hi Leo,
>>>
>>> What kind of examples are you looking for? Participant side? Or how to
>>> build IdealState?
>>>
>>> Best,
>>>
>>> Junkai
>>>
>>> On Fri, Oct 30, 2020 at 9:57 PM Leo, Li <go...@gmail.com> wrote:
>>>
>>>> Hi,
>>>> I am a new user of Helix. I wonder if you could give an example code
>>>> of using this user_defined model
>>>> https://helix.apache.org/1.0.1-docs/tutorial_user_def_rebalancer.html?
>>>>
>>>> Thank you,
>>>> Leo
>>>>
>>>