[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel=16356095#comment-16356095 ] Noble Paul commented on SOLR-5069: -- Streaming API is there way to go > MapReduce for SolrCloud > --- > > Key: SOLR-5069 > URL: https://issues.apache.org/jira/browse/SOLR-5069 > Project: Solr > Issue Type: New Feature > Components: SolrCloud >Reporter: Noble Paul >Assignee: Noble Paul >Priority: Major > > Solr currently does not have a way to run long running computational tasks > across the cluster. We can piggyback on the mapreduce paradigm so that users > have smooth learning curve. > * The mapreduce component will be written as a RequestHandler in Solr > * Works only in SolrCloud mode. (No support for standalone mode) > * Users can write MapReduce programs in Javascript or Java. First cut would > be JS ( ? ) > h1. sample word count program > h2.how to invoke? > http://host:port/solr/collection-x/mapreduce?map===collectionX > h3. params > * map : A javascript implementation of the map program > * reduce : a Javascript implementation of the reduce program > * sink : The collection to which the output is written. If this is not passed > , the request will wait till completion and respond with the output of the > reduce program and will be emitted as a standard solr response. . If no sink > is passed the request will be redirected to the "reduce node" where it will > wait till the process is complete. If the sink param is passed ,the rsponse > will contain an id of the run which can be used to query the status in > another command. > * reduceNode : Node name where the reduce is run . If not passed an arbitrary > node is chosen > The node which received the command would first identify one replica from > each slice where the map program is executed . It will also identify one > another node from the same collection where the reduce program is run. Each > run is given an id and the details of the nodes participating in the run will > be written to ZK (as an ephemeral node). > h4. map script > {code:JavaScript} > var res = $.streamQuery($.param(“q"));//this is not run across the cluster. > //Only on this index > while(res.hasMore()){ > var doc = res.next(); > map(doc); > } > function map(doc) { > var txt = doc.get(“txt”);//the field on which word count is performed > var words = txt.split(" "); >for(i = 0; i < words.length; i++){ > $.emit(words[i],{‘count’:1});// this will send the map over to //the > reduce host > } > } > {code} > Essentially two threads are created in the 'map' hosts . One for running the > program and the other for co-ordinating with the 'reduce' host . The maps > emitted are streamed live over an http connection to the reduce program > h4. reduce script > This script is run in one node . This node accepts http connections from map > nodes and the 'maps' that are sent are collected in a queue which will be > polled and fed into the reduce program. This also keeps the 'reduced' data in > memory till the whole run is complete. It expects a "done" message from all > 'map' nodes before it declares the tasks are complete. After reduce program > is executed for all the input it proceeds to write out the result to the > 'sink' collection or it is written straight out to the response. > {code:JavaScript} > var pair = $.nextMap(); > var reduced = $.getCtx().getReducedMap();// a hashmap > var count = reduced.get(pair.key()); > if(count === null) { > count = {“count”:0}; > reduced.put(pair.key(), count); > } > count.count += pair.val().count ; > {code} > h4.example output > {code:JavaScript} > { > “result”:[ > “wordx”:{ > “count”:15876765 > }, > “wordy” : { >“count”:24657654 > } > > ] > } > {code} > TBD > * The format in which the output is written to the target collection, I > assume the reducedMap will have values mapping to the schema of the collection > -- This message was sent by Atlassian JIRA (v7.6.3#76005) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14552175#comment-14552175 ] Noble Paul commented on SOLR-5069: -- Is there some low hanging fruit that we can achieve easily? MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery($.param(“q));//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); map(doc); } function map(doc) { var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.emit(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
Re: [jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
If you're going to do be shuffling data to multiple worker nodes then data will be crossing the network. Shuffling provides the foundation for certain parallel computing tasks, such as performing large scale parallel relational algebra. For machine learning algorithms we'll likely need a parallel iterative design which leaves the data in place. Joel Bernstein http://joelsolr.blogspot.com/ On Wed, May 20, 2015 at 4:11 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 11:06 AM, Noble Paul noble.p...@gmail.com wrote: The problem with streaming is data locality. Data needs to be transferred across network to do the processing Nothing saying that you can't process data before it's streamed out, right? -Yonik - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
Re: [jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
On Wed, May 20, 2015 at 8:41 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 11:06 AM, Noble Paul noble.p...@gmail.com wrote: The problem with streaming is data locality. Data needs to be transferred across network to do the processing Nothing saying that you can't process data before it's streamed out, right? yes, if our query language is expressive enough . Sometimes you need a little programming language to achieve that -Yonik - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org -- - Noble Paul
Re: [jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
On Wed, May 20, 2015 at 12:04 PM, Noble Paul noble.p...@gmail.com wrote: On Wed, May 20, 2015 at 8:41 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 11:06 AM, Noble Paul noble.p...@gmail.com wrote: The problem with streaming is data locality. Data needs to be transferred across network to do the processing Nothing saying that you can't process data before it's streamed out, right? yes, if our query language is expressive enough . Sometimes you need a little programming language to achieve that Right - and different languages can go on top of the base streaming stuff... either before or after the streaming step. There's no reason we can't stream derived data - it doesn't need to be just documents. -Yonik - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
Re: [jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
On Wed, May 20, 2015 at 10:17 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 12:04 PM, Noble Paul noble.p...@gmail.com wrote: On Wed, May 20, 2015 at 8:41 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 11:06 AM, Noble Paul noble.p...@gmail.com wrote: The problem with streaming is data locality. Data needs to be transferred across network to do the processing Nothing saying that you can't process data before it's streamed out, right? yes, if our query language is expressive enough . Sometimes you need a little programming language to achieve that Right - and different languages can go on top of the base streaming stuff... either before or after the streaming step. There's no reason we can't stream derived data - it doesn't need to be just documents. Yes, but is there away to do it now? If we can have a DSL which can do process docs and emit the processed data , then the streaming API may be able to do without data locality . I guess the streaming API run as a standalone program. can it not be running soemwhere in the Solr cluster itself? -Yonik - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org -- - Noble Paul
Re: [jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
The Streaming Expressions language is a DSL to process docs and emit processed data. The parallel SQL engine will also fit into this category. Both of these languages compile to the Streaming API which is basically a real-time map-reduce framework that runs on SolrCloud worker nodes. The Streaming API has excellent data locality for a Map-Reduce engine because it performs the map stage and sorting and partitioning of result sets inside of Solr before tuples are streamed. Sorted and partitioned tuples are then sent directly to the correct worker nodes to be reduced. The Streaming API doesn't follow a strict map/reduce model though. Streams are merged and manipulated by wrapping decorator streams around each other. So the streaming API is much more flexible then old style map/reduce. But the Streaming API is not designed for parallel iterative algorithms like gradient descent. For the parallel iterative case it's much faster to leave the data in place and run embedded algorithm inside of the Solr. At this point data must cross the network if you have multiple worker nodes. Joel Bernstein http://joelsolr.blogspot.com/ On Wed, May 20, 2015 at 5:57 PM, Noble Paul noble.p...@gmail.com wrote: On Wed, May 20, 2015 at 10:17 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 12:04 PM, Noble Paul noble.p...@gmail.com wrote: On Wed, May 20, 2015 at 8:41 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 11:06 AM, Noble Paul noble.p...@gmail.com wrote: The problem with streaming is data locality. Data needs to be transferred across network to do the processing Nothing saying that you can't process data before it's streamed out, right? yes, if our query language is expressive enough . Sometimes you need a little programming language to achieve that Right - and different languages can go on top of the base streaming stuff... either before or after the streaming step. There's no reason we can't stream derived data - it doesn't need to be just documents. Yes, but is there away to do it now? If we can have a DSL which can do process docs and emit the processed data , then the streaming API may be able to do without data locality . I guess the streaming API run as a standalone program. can it not be running soemwhere in the Solr cluster itself? -Yonik - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org -- - Noble Paul
Re: [jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
Joel, Is this ticket an attempt to solve that ? SOLR-7560 On Wed, May 20, 2015 at 11:08 PM, Joel Bernstein joels...@gmail.com wrote: The Streaming Expressions language is a DSL to process docs and emit processed data. The parallel SQL engine will also fit into this category. Both of these languages compile to the Streaming API which is basically a real-time map-reduce framework that runs on SolrCloud worker nodes. The Streaming API has excellent data locality for a Map-Reduce engine because it performs the map stage and sorting and partitioning of result sets inside of Solr before tuples are streamed. Sorted and partitioned tuples are then sent directly to the correct worker nodes to be reduced. The Streaming API doesn't follow a strict map/reduce model though. Streams are merged and manipulated by wrapping decorator streams around each other. So the streaming API is much more flexible then old style map/reduce. But the Streaming API is not designed for parallel iterative algorithms like gradient descent. For the parallel iterative case it's much faster to leave the data in place and run embedded algorithm inside of the Solr. At this point data must cross the network if you have multiple worker nodes. Joel Bernstein http://joelsolr.blogspot.com/ On Wed, May 20, 2015 at 5:57 PM, Noble Paul noble.p...@gmail.com wrote: On Wed, May 20, 2015 at 10:17 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 12:04 PM, Noble Paul noble.p...@gmail.com wrote: On Wed, May 20, 2015 at 8:41 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 11:06 AM, Noble Paul noble.p...@gmail.com wrote: The problem with streaming is data locality. Data needs to be transferred across network to do the processing Nothing saying that you can't process data before it's streamed out, right? yes, if our query language is expressive enough . Sometimes you need a little programming language to achieve that Right - and different languages can go on top of the base streaming stuff... either before or after the streaming step. There's no reason we can't stream derived data - it doesn't need to be just documents. Yes, but is there away to do it now? If we can have a DSL which can do process docs and emit the processed data , then the streaming API may be able to do without data locality . I guess the streaming API run as a standalone program. can it not be running soemwhere in the Solr cluster itself? -Yonik - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org -- - Noble Paul -- - Noble Paul
Re: [jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
SOLR-7560 will provides a parallel SQL engine for SolrCloud. It's designed to run interactive SQL queries across large clusters of servers. This is one of the core big data use cases. Joel Bernstein http://joelsolr.blogspot.com/ On Wed, May 20, 2015 at 7:07 PM, Noble Paul noble.p...@gmail.com wrote: Joel, Is this ticket an attempt to solve that ? SOLR-7560 On Wed, May 20, 2015 at 11:08 PM, Joel Bernstein joels...@gmail.com wrote: The Streaming Expressions language is a DSL to process docs and emit processed data. The parallel SQL engine will also fit into this category. Both of these languages compile to the Streaming API which is basically a real-time map-reduce framework that runs on SolrCloud worker nodes. The Streaming API has excellent data locality for a Map-Reduce engine because it performs the map stage and sorting and partitioning of result sets inside of Solr before tuples are streamed. Sorted and partitioned tuples are then sent directly to the correct worker nodes to be reduced. The Streaming API doesn't follow a strict map/reduce model though. Streams are merged and manipulated by wrapping decorator streams around each other. So the streaming API is much more flexible then old style map/reduce. But the Streaming API is not designed for parallel iterative algorithms like gradient descent. For the parallel iterative case it's much faster to leave the data in place and run embedded algorithm inside of the Solr. At this point data must cross the network if you have multiple worker nodes. Joel Bernstein http://joelsolr.blogspot.com/ On Wed, May 20, 2015 at 5:57 PM, Noble Paul noble.p...@gmail.com wrote: On Wed, May 20, 2015 at 10:17 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 12:04 PM, Noble Paul noble.p...@gmail.com wrote: On Wed, May 20, 2015 at 8:41 PM, Yonik Seeley ysee...@gmail.com wrote: On Wed, May 20, 2015 at 11:06 AM, Noble Paul noble.p...@gmail.com wrote: The problem with streaming is data locality. Data needs to be transferred across network to do the processing Nothing saying that you can't process data before it's streamed out, right? yes, if our query language is expressive enough . Sometimes you need a little programming language to achieve that Right - and different languages can go on top of the base streaming stuff... either before or after the streaming step. There's no reason we can't stream derived data - it doesn't need to be just documents. Yes, but is there away to do it now? If we can have a DSL which can do process docs and emit the processed data , then the streaming API may be able to do without data locality . I guess the streaming API run as a standalone program. can it not be running soemwhere in the Solr cluster itself? -Yonik - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org -- - Noble Paul -- - Noble Paul
Re: [jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
The problem with streaming is data locality. Data needs to be transferred across network to do the processing On May 20, 2015 8:15 PM, Yonik Seeley (JIRA) j...@apache.org wrote: [ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14552414#comment-14552414 ] Yonik Seeley commented on SOLR-5069: Looks like SOLR-6526 (Solr Streaming API) is pretty much map-reduce? And then on top is SOLR-7377 (Solr Streaming Expressions) and SOLR-7560 (Parallel SQL) MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port /solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery($.param(“q));//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); map(doc); } function map(doc) { var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.emit(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
Re: [jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
On Wed, May 20, 2015 at 11:06 AM, Noble Paul noble.p...@gmail.com wrote: The problem with streaming is data locality. Data needs to be transferred across network to do the processing Nothing saying that you can't process data before it's streamed out, right? -Yonik - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14552414#comment-14552414 ] Yonik Seeley commented on SOLR-5069: Looks like SOLR-6526 (Solr Streaming API) is pretty much map-reduce? And then on top is SOLR-7377 (Solr Streaming Expressions) and SOLR-7560 (Parallel SQL) MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery($.param(“q));//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); map(doc); } function map(doc) { var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.emit(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14549383#comment-14549383 ] Markus Jelsma commented on SOLR-5069: - [~ab] anything new to add to this topic? I am sure interest is still here :) MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery($.param(“q));//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); map(doc); } function map(doc) { var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.emit(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message was sent by Atlassian JIRA (v6.3.4#6332) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13911597#comment-13911597 ] ASF subversion and git services commented on SOLR-5069: --- Commit 1571702 from [~noble.paul] in branch 'dev/trunk' [ https://svn.apache.org/r1571702 ] SOLR-5069 fixing test failure MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message was sent by Atlassian JIRA (v6.1.5#6160) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13911600#comment-13911600 ] ASF subversion and git services commented on SOLR-5069: --- Commit 1571703 from [~noble.paul] in branch 'dev/branches/branch_4x' [ https://svn.apache.org/r1571703 ] SOLR-5069 fixing test failure MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message was sent by Atlassian JIRA (v6.1.5#6160) - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13720869#comment-13720869 ] Andrzej Bialecki commented on SOLR-5069: - See here for an explanation how this works in MongoDB: http://isurues.wordpress.com/2013/05/28/what-is-re-reduce-in-mongodb-map-reduce/ . CouchDB also uses the same reduce function, only it passes a boolean flag to inform the function whether a particular invocation is the first reduce (acting on values straight from mappers) or a re-reduce (acting on results of previous partial reduces). MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13719540#comment-13719540 ] Otis Gospodnetic commented on SOLR-5069: This is great to see - I asked about this in SOLR-1301 - https://issues.apache.org/jira/browse/SOLR-1301?focusedCommentId=13678948page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13678948 :) {quote} The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). {quote} Lukas and Andrzej have already addressed my immediate thought when I read the above, but they talked about using the cost approach, limiting resource use, and such. But I think we should learn from others' mistakes and choices here. Is it good enough to limit resources? Just limiting resources means that any concurrent queries *will* be effected - the question is just how much. Would it be better to mark some nodes as eligible for running analytical/batch/MR jobs + search or eligible for running analytical/batch/MR jobs and NO search - i.e. nodes that are a part of the SolrCloud cluster, but run ONLY these jobs and do NOT handle queries? I think we saw DataStax do this with Cassandra and Brisk and we see that with people using HBase with HBase replication and using one HBase cluster for real-time/interactive access and the other cluster for running jobs. MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count =
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13719550#comment-13719550 ] Noble Paul commented on SOLR-5069: -- bq.Would it be better to mark some nodes as eligible for running analytical/batch/MR jobs + search Instead of marking certain nodes as (eligible for X)how about passing the node names in the request itself ? That way we are not introducing some kind of 'role' in the system but still get all the benefits? MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13719554#comment-13719554 ] Otis Gospodnetic commented on SOLR-5069: bq. Instead of marking certain nodes as (eligible for X)how about passing the node names in the request itself ? That way we are not introducing some kind of 'role' in the system but still get all the benefits? But if searches are running on *all* nodes, then the above doesn't achieve complete separation of search vs. job work. MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13719556#comment-13719556 ] Noble Paul commented on SOLR-5069: -- bq.But if searches are running on all nodes, then the above doesn't achieve complete separation of search vs. job work. makes sense... MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13719563#comment-13719563 ] Otis Gospodnetic commented on SOLR-5069: bq. It should be something we should think of as a feature of Solr. Being a part of a cluster but not taking part in certain roles (leader/search/jobs etc Yeah, perhaps something like that. We already have Overseer and Leader, which are also roles of some sort, though those are completely managed by SolrCloud, meaning SolrCloud/ZK do the node election and node assignment for these particular roles, AFAIK. While for search vs. job (vs. mixed) role the assignment is likely to come from a human+ZK. MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13719593#comment-13719593 ] Yonik Seeley commented on SOLR-5069: bq. It should be something we should think of as a feature of Solr. Right - it's unrelated to this feature. We've already kicked around the idea of roles for nodes for years now (like in SOLR-2765), and they would be useful in many contexts. Someone actually needs to do the work though... patches welcome ;-) MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13719612#comment-13719612 ] Andrzej Bialecki commented on SOLR-5069: - bq. some things will be completely streamable w/o any need for buffering... think of re-implementing the terms component here - we can access terms in sorted order so the reducer would simply need to do a merge sort on the streams and then stream that result back! It could be probably implemented as a special case, because it strongly depends on the map() output being sorted. However, in general case reducer must wait for all mappers to finish because mappers may produce keys out of order and non-unique. +1 on node roles, as a separate issue - it should not hold off this issue. MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail:
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13719654#comment-13719654 ] Andrzej Bialecki commented on SOLR-5069: - An alternative solution for minimizing the amount of data in memory during reduce phase is to use re-reduce, or a reduce-side combiner, using Hadoop terminology. This is an additional function that runs on the reducer and periodically performs intermediate reductions of already accumulated values for a key, and preserves the intermediate results (and discards the accumulated values). This function does not emit anything to the final output. Then the final reduction function operates on a mix of values that arrived since the last intermediate reduction, plus all results of previous intermediate reductions. This works well for simple aggregations (where the additional function may be in fact a copy of the reduce function) but may not be suitable to all classes of problems. MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see:
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13718058#comment-13718058 ] Andrzej Bialecki commented on SOLR-5069: - bq. why can't reduce start as soon as the mappers start producing? Because reducer needs to operate on the complete list of values for a given key. Take for example the wordcount - not waiting for all mappers would cause reducer to emit only partial aggregations. In general mappers should be free to emit arbitrary keys, so new values may appear at any moment until all mappers are finished. MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13718068#comment-13718068 ] Lukas Vlcek commented on SOLR-5069: --- Hello, may be OT but in spite of the fact that having MapReduce in (near) real time [clustered] search server sounds very interesting and indeed useful, is this something that is good to put into the system? I might be naive but as far as I can understand MR tasks can be both RAM and IO (disk,network) intensive. How can one tune the system for fast indexing/search performance if the additional load put on the system from MR is hardly predictable? Not to mention the fact that MR is like a hammer. And if you put hammer into hands of users, then everything starts looking like a you know the story. Regards, Lukas MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands,
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13718067#comment-13718067 ] Noble Paul commented on SOLR-5069: -- I guess , I haven't explained correctly. The reducer output is available only after all the mappers are done. But the reducer is started along with the mappers and working in parallel . MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13718152#comment-13718152 ] Noble Paul commented on SOLR-5069: -- bq.MR tasks can be both RAM and IO (disk,network) intensive You are right. We won't recommend people to use the cluster for MR and search indexing at the same time . They will definitely see degraded performance. But then, that is expected , right? How is it better than setting up another cluster (Hadoop) for MR if if you need it? MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13718434#comment-13718434 ] Andrzej Bialecki commented on SOLR-5069: - bq. The reducer output is available only after all the mappers are done. But the reducer is started along with the mappers and is working in parallel. [~noble.paul]: Sure, you can start the reducer - but for any given key you have to wait anyway with processing until all values for a given key become available - and this practically means that the reducer has to wait until all mappers are done. bq. How can one tune the system for fast indexing/search performance if the additional load put on the system from MR is hardly predictable? [~lukas.vlcek]: that's why I suggested that this framework should have the ability to specify a budget for job execution, at least in terms of RAM or key-value pair count. Still, for occasional analytic jobs or simple aggregations the load should be predictable or bearable, and the performance cost of using this tool would be negligible compared to the cost of integrating and operating a separate analytic platform. MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13718518#comment-13718518 ] Lukas Vlcek commented on SOLR-5069: --- [~porqpine]: Well, I see the point. From the user point of view this sounds very cool and it will be interesting to see how this feature works out. Though, this reminds me the situation that happened in Google couple of year ago (I heard this from one ex-Googler, not sure if there is any official evidence) when they introduced MR platform internally and all summer interns started using it. A lot of non-optimal tasks started eating their resources - because it is so easy to translate a lot of problems into MR (but it does not mean that MR solution to the problem is the optimal one). As for setting up a separate analytical platform, well... the cost of setting it up is one thing, but the benefit of existing tooling and experience is another one. Are you going to reimplement Mahout into Solr? - well may be you are not aiming at this level of complexity. You can throttle the thing on many levels, as a result the task will just run longer, right? Isn't this in fact a big challenge? If I understand Lucene correctly, the costly part is if you need to keep aged IndexReaders around because this leads to higher number of opened segments and consumption of related resources. And what if the data included into the MR calculation changes (reindex/delete) in the meantime? Then you need to be careful in presenting the results to the clients because they may be too used to Hadoop MR where the original data set is still available. Anyway, I am sure you are already aware of all this. I am just curious :-) MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced =
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13718578#comment-13718578 ] Noble Paul commented on SOLR-5069: -- bq.that's why I suggested that this framework should have the ability to specify a budget ... Yoi are right. Even the version 1.0 should have a way to budget the RAM at mapper and reducer for a given task bq.A lot of non-optimal tasks started eating their resources... bq.but the benefit of existing tooling and experience is another one Actually it works both ways. Mahout (or other systems) will have more mature support for certain tasks. There are more people familiar with Solr/Lucene. That will help them to be up and running with little effort. bq.as a result the task will just run longer, right? Well, that is the tradeoff you make. choose expensive h/w or wait longer bq.the costly part is if you need to keep aged IndexReaders around ... Yes, If you have frequent commits and frequent MR tasks running. You will rarely run a long running process on a very frequently changing dataset . Lucene does not delete 'data' because segments are cleaned up. They are just copied over if segment merges happen. If deletes happen in between , Lucene will behave much better because we always operate on the same IndexReader and the results will be consistent with the state of the data at the time the task is fired MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13718762#comment-13718762 ] Yonik Seeley commented on SOLR-5069: Awesome stuff Noble! bq. why can't reduce start as soon as the mappers start producing? whatever is emitted by the mapper is up for reducer to chew on. Right - and some things will be completely streamable w/o any need for buffering... think of re-implementing the terms component here - we can access terms in sorted order so the reducer would simply need to do a merge sort on the streams and then stream that result back! MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13718785#comment-13718785 ] Eks Dev commented on SOLR-5069: --- wow, this is getting pretty close to collection clustering and other candies, somehow to plug-in mahout and it's there Great job and great direction for solr. End-applications not only need to find things, they often want to do something with them as well :) Thanks! MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/collection-x/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13716486#comment-13716486 ] Andrzej Bialecki commented on SOLR-5069: - Exciting idea! Almost as exciting as SolrCloud on MapReduce :) A few comments: # distributed map-reduce in reality is a sequence of: ## split input and assign splits to M nodes ## apply map() on M nodes in parallel ##* for large datasets the emitted data from mappers is spooled to disk ## shuffle - ie. partition and ship emitted data from M mappers into N reducers ##* (wait until all mappers are done, so that each partition's key-space is complete) ## sort by key in each of N reducers, collecting values for each key ##* again, for large datasets this is a disk-based sort ## apply N reducers in parallel and emit final output (in N parts) # if I understand it correctly the model that you presented has some limitations: ## as many input splits as there are shards (and consequently as many mappers) ## single reducer. Theoretically it should be possible to use N nodes to act as reducers if you implement the concept of partitioner - this would cut down the memory load on each reducer node. Of course, streaming back the results would be a challenge, but saving them into a collection should work just fine. ## no shuffling - all data from mappers will go to a single reducer ## no intermediate storage of data, all intermediate values need to fit in memory ## what about the sorting phase? I assume it's an implicit function in the reducedMap (treemap?) # since all fine-grained emitted values from map end up being sent to 1 reducer, which has to collect all this data in memory first before applying the reduce() op, the concept of a map-side combiner seems useful, to be able to quickly minimize the amount of data to be sent to reducer. # it would be very easy to OOM your Solr nodes at the reduce phase. There should be some built-in safety mechanism for this. # what parts of Solr are available in the script's context? Making all Solr API available could lead to unpredictable side-effects, so this set of APIs needs to be curated. E.g. I think it would make sense to make analyzer factories available. And finally, an observation: regular distributed search can be viewed as a special case of map-reduce computation ;) MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13716624#comment-13716624 ] Noble Paul commented on SOLR-5069: -- Thanks Andrzej I started off with a simple model so that the version 1 can be implemented easily. 'N' reducers add to implementation complexity. However , it should be done eventually. bq.no intermediate storage of data, all intermediate values need to fit in memory Yes,in my model, the mappers will be throttled so that we can fix the amount of intermediate data kept in memory. $.map() call would wait if the size threshold is reached bq. what about the sorting phase? I assume it's an implicit function in the reducedMap (treemap?) we should have the choice on how to sort the map .Yes, Some kind of sorted map should be offered .probably sort on some key's value in the map bq.it would be very easy to OOM your Solr nodes at the reduce phase. Sure, here the idea is to do some overflow to disk beyond a threshold. With memory getting abundant , we probably should use some off heap solution , so that the reduce is not I/O bound. bq.what parts of Solr are available in the script's context Good that you asked. We should keep the API's available limited . For instance , anything that can alter the state of the system should not be exposed to the script. Anything that can help processing /manipulating data should be exposed MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} TBD * The format in which the output is written to the target collection, I assume
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13716655#comment-13716655 ] Andrzej Bialecki commented on SOLR-5069: - bq. Sure, here the idea is to do some overflow to disk beyond a threshold. Berkeley DB, db4o, and an Apache-licensed MapDB (mapdb.org), and probably others, all provide persistent Java Collections API. We could use one of these - you could add a provider mechanism to separate the actual implementation from the plain Collections api. bq. $.map() call would wait if the size threshold is reached Throttling the mappers wouldn't help with OOM on the reduce() side - reduce() can start only when all mappers are finished. I think a map-side combiner would be much more helpful, if possible (reductions that are not simple aggregations usually can't be performed in map-side combiners). MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail:
[jira] [Commented] (SOLR-5069) MapReduce for SolrCloud
[ https://issues.apache.org/jira/browse/SOLR-5069?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=13717922#comment-13717922 ] Noble Paul commented on SOLR-5069: -- bq.reduce() can start only when all mappers are finished Why. why can't reduce start as soon as the mappers start producing? whatever is emitted by the mapper is up for reducer to chew on. All said, map side combiner is definitely useful and would reduce memory/network IO MapReduce for SolrCloud --- Key: SOLR-5069 URL: https://issues.apache.org/jira/browse/SOLR-5069 Project: Solr Issue Type: New Feature Components: SolrCloud Reporter: Noble Paul Assignee: Noble Paul Solr currently does not have a way to run long running computational tasks across the cluster. We can piggyback on the mapreduce paradigm so that users have smooth learning curve. * The mapreduce component will be written as a RequestHandler in Solr * Works only in SolrCloud mode. (No support for standalone mode) * Users can write MapReduce programs in Javascript or Java. First cut would be JS ( ? ) h1. sample word count program h2.how to invoke? http://host:port/solr/mapreduce?map=map-scriptreduce=reduce-scriptsink=collectionX h3. params * map : A javascript implementation of the map program * reduce : a Javascript implementation of the reduce program * sink : The collection to which the output is written. If this is not passed , the request will wait till completion and respond with the output of the reduce program and will be emitted as a standard solr response. . If no sink is passed the request will be redirected to the reduce node where it will wait till the process is complete. If the sink param is passed ,the rsponse will contain an id of the run which can be used to query the status in another command. * reduceNode : Node name where the reduce is run . If not passed an arbitrary node is chosen The node which received the command would first identify one replica from each slice where the map program is executed . It will also identify one another node from the same collection where the reduce program is run. Each run is given an id and the details of the nodes participating in the run will be written to ZK (as an ephemeral node). h4. map script {code:JavaScript} var res = $.streamQuery(*:*);//this is not run across the cluster. //Only on this index while(res.hasMore()){ var doc = res.next(); var txt = doc.get(“txt”);//the field on which word count is performed var words = txt.split( ); for(i = 0; i words.length; i++){ $.map(words[i],{‘count’:1});// this will send the map over to //the reduce host } } {code} Essentially two threads are created in the 'map' hosts . One for running the program and the other for co-ordinating with the 'reduce' host . The maps emitted are streamed live over an http connection to the reduce program h4. reduce script This script is run in one node . This node accepts http connections from map nodes and the 'maps' that are sent are collected in a queue which will be polled and fed into the reduce program. This also keeps the 'reduced' data in memory till the whole run is complete. It expects a done message from all 'map' nodes before it declares the tasks are complete. After reduce program is executed for all the input it proceeds to write out the result to the 'sink' collection or it is written straight out to the response. {code:JavaScript} var pair = $.nextMap(); var reduced = $.getCtx().getReducedMap();// a hashmap var count = reduced.get(pair.key()); if(count === null) { count = {“count”:0}; reduced.put(pair.key(), count); } count.count += pair.val().count ; {code} h4.example output {code:JavaScript} { “result”:[ “wordx”:{ “count”:15876765 }, “wordy” : { “count”:24657654 } ] } {code} TBD * The format in which the output is written to the target collection, I assume the reducedMap will have values mapping to the schema of the collection -- This message is automatically generated by JIRA. If you think it was sent incorrectly, please contact your JIRA administrators For more information on JIRA, see: http://www.atlassian.com/software/jira - To unsubscribe, e-mail: dev-unsubscr...@lucene.apache.org For additional commands, e-mail: dev-h...@lucene.apache.org