[PIO 0.10.0][Universtal Recommender 0.5] Exception when using remote Elasticsearch while training

[Marius Rabenarivo]

Hello,

I try to run training from a remote machine.

It fails in later stage due to this exception :

org.elasticsearch.hadoop.rest.EsHadoopNoNodesLeftException: Connection error (check network and/or proxy settings)- all nodes failed; tried [[engine-server:9300]]

According to the stack trace the error is coming from this line :

https://github.com/actionml/universal-recommender/blob/master/src/main/scala/EsClient.scala#L179

It was using a default configuration to access ES.

I tried to replace this line with :


    val elasticSearchConf = Storage.getConfig("ELASTICSEARCH").get

    val conf = Map(
      "es.mapping.id" -> "id",
      "es.nodes" -> elasticSearchConf.properties.get("HOSTS").head,
      "es.port" -> elasticSearchConf.properties.get("PORTS").head
    )

    indexRDD.saveToEs(newIndexURI, conf)

but I still getting the same error.

Regards,

Marius

[Pat Ferrel]

PIO in the past used the “transport client” of Elasticsearch, which is on port 9300 by default, The UR uses the REST client for some things like rapid writes to ES using Spark using a lib supplied by Elasticsearch.

So, Spark needs to have config that the Elasticsearch library can read. This can be done most easily by putting it into a sparkConf section of engine.json

Like this:

  "sparkConf": {

    "spark.serializer": "org.apache.spark.serializer.KryoSerializer",

    "spark.kryo.registrator": "org.apache.mahout.sparkbindings.io.MahoutKryoRegistrator",

    "spark.kryo.referenceTracking": "false",

    "spark.kryoserializer.buffer": "300m",

    "es.index.auto.create": “true”,

    “es.nodes”: “node1,node2”  <========== add this

  },

 

The list is comma separated host ip addresses of DNS names so no spaces. If you are using the default port 9200 you don’t need to specify it. These values are defined by Elasticsearch so see their docs if you have any question about what can be configured in this way. For instance you can also specify password and username in this manner.

--
You received this message because you are subscribed to the Google Groups "actionml-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to actionml-use...@googlegroups.com.
To post to this group, send email to action...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/CAC-ATVES0FM4iFtXGoU9-URikEq%2B2d6NaV10Eh1S4_t1Ub4PcA%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

[Pat Ferrel]

BTW the docs for sparkConf are here: http://actionml.com/docs/ur_config#ur-spark-conf

To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/EA18C433-834A-4917-99C3-A2460C281667%40occamsmachete.com.

[Marius Rabenarivo]

I tried using these configurations in the sparkConf section of engine.json

It has the same effect as the code I mentioned below but I still getting the same exception.

To unsubscribe from this group and stop receiving emails from it, send an email to actionml-user+unsubscribe@googlegroups.com.

To post to this group, send email to action...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/CAC-ATVES0FM4iFtXGoU9-URikEq%2B2d6NaV10Eh1S4_t1Ub4PcA%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

--
You received this message because you are subscribed to the Google Groups "actionml-user" group.

To unsubscribe from this group and stop receiving emails from it, send an email to actionml-user+unsubscribe@googlegroups.com.

To post to this group, send email to action...@googlegroups.com.

[Marius Rabenarivo]

I tried to run the command

curl -X GET http://engine-server:9200

and I'm getting a JSON response but when I'm running the command

pio train

I got

org.elasticsearch.hadoop.rest.EsHadoopTransportException: java.net.NoRouteToHostException: No route to host (Host unreachable)

exception

[Marius Rabenarivo]

It works when using the default port 9200 and by changing the Elasticsearch cluster name to the same as the hostname as suggested here

http://stackoverflow.com/questions/25885411/elasticsearch-error-failed-to-connect-to-master-no-route-to-host

[Pat Ferrel]

We use a clustername different from hostname all the time. In fact if you have more than one node you will have to make them different since there are multiple hostnames and one clustername.

The clustername should be something unique to your application because ES has the sometimes helpful, but often maddeningly bad and hidden, habit of merging nodes into a cluster by clustername. For instance the default for ES is to merge all nodes on a LAN into one cluster.

Anyway can you share your engine.json and the result of curl with exactly the same hostname? You need to add the clustername to pio-env.sh too so can you share that as well?

To unsubscribe from this group and stop receiving emails from it, send an email to actionml-use...@googlegroups.com.

To post to this group, send email to action...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/CAC-ATVGHNYHJ-V78B1knbpqWGn_cpkvYCC_%2BW3Tmdw6kywcSTw%40mail.gmail.com.

[Marius Rabenarivo]

Hello,

Find attached the pio-env.sh and engine.json

that we are using.

2017-04-08 18:46 GMT+04:00 Marius Rabenarivo <mariusra...@gmail.com>:

I'll share them monday.

2017-04-07 22:23 GMT+04:00 Pat Ferrel <p...@occamsmachete.com>:

We use a clustername different from hostname all the time. In fact if you have more than one node you will have to make them different since there are multiple hostnames and one clustername.

The clustername should be something unique to your application because ES has the sometimes helpful, but often maddeningly bad and hidden, habit of merging nodes into a cluster by clustername. For instance the default for ES is to merge all nodes on a LAN into one cluster.

Anyway can you share your engine.json and the result of curl with exactly the same hostname? You need to add the clustername to pio-env.sh too so can you share that as well?

On Apr 7, 2017, at 12:25 AM, Marius Rabenarivo <mariusra...@gmail.com> wrote:

It works when using the default port 9200 and by changing the Elasticsearch cluster name to the same as the hostname as suggested here

http://stackoverflow.com/questions/25885411/elasticsearch-error-failed-to-connect-to-master-no-route-to-host

2017-04-07 9:57 GMT+04:00 Marius Rabenarivo <mariusrabenarivo@gmail.com>:

I tried to run the command

curl -X GET http://engine-server:9200

and I'm getting a JSON response but when I'm running the command

pio train

I got 

org.elasticsearch.hadoop.rest.EsHadoopTransportException: java.net.NoRouteToHostException: No route to host (Host unreachable)

exception

To post to this group, send email to actionml-user@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/CAC-ATVES0FM4iFtXGoU9-URikEq%2B2d6NaV10Eh1S4_t1Ub4PcA%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

-- 
You received this message because you are subscribed to the Google Groups "actionml-user" group.
To unsubscribe from this group and stop receiving emails from it, send an email to actionml-user+unsubscribe@googlegroups.com.

To post to this group, send email to actionml-user@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/EA18C433-834A-4917-99C3-A2460C281667%40occamsmachete.com.
For more options, visit https://groups.google.com/d/optout.

-- 
You received this message because you are subscribed to the Google Groups "actionml-user" group.

To unsubscribe from this group and stop receiving emails from it, send an email to actionml-user+unsubscribe@googlegroups.com.

To post to this group, send email to actionml-user@googlegroups.com.

[Pat Ferrel]

Do you have the clustername set in Elasticsearch config also? It is changed in /usr/local/elasticsearch/config/elasticsearch.yml Setting this in both places will decouple server name from cluster name. Unfortunately changing a cluster name will call for a complete reset of PredictionIO data since it will loose track of engine metadata. I have found it easiest to export/backup then delete all apps and re-install or erase any table in hbase and elasticsearch. You have to get the clustername right before you start because it’s a pain to change.

Also the fields must correspond to arrays of strings, not an array of true or false, an array of one element, a string of “T” or “F” would work for this. For fields the property must be attached to all items by using a $set event, then your visible and sponsored will work more like what you expect. They are not working at all now.

To unsubscribe from this group and stop receiving emails from it, send an email to actionml-use...@googlegroups.com.
To post to this group, send email to action...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/CAC-ATVG8madmx19TqijGra4n6TxDSnxvziB4RNggnvAosB7Etg%40mail.gmail.com.

For more options, visit https://groups.google.com/d/optout.

<engine.json><pio-env.sh>

[Marius Rabenarivo]

Yes, I set the clustername in Elasticsearch config too.

Thank you for your advices about fields.

To unsubscribe from this group and stop receiving emails from it, send an email to actionml-user+unsubscribe@googlegroups.com.

To post to this group, send email to action...@googlegroups.com.

[Pat Ferrel]

RE Elasticsearch, there is no need to have the clustername and hostname the same, in fact I have never done this and we run larger clusters of ES. Are you using /etc/hosts or something other than DNS to resolve the hostname or some other mechanism?

I guess if it’s working, it’s not an immediate issue but just wanted to make that clear.

To unsubscribe from this group and stop receiving emails from it, send an email to actionml-use...@googlegroups.com.

To post to this group, send email to action...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/actionml-user/CAC-ATVESU16R_w_Hs%2B49a7unJmpUdjghASDjXRiYhdZRrEAggQ%40mail.gmail.com.

[Marius Rabenarivo]

Yes, I use /etc/hosts

Unless for HBase's Master and RegionServer which require a DNS for resolving hostname.

I encountered an issue, which is that Zookeeper doesn't recognize the hdfs:// protocol too. It instead create a directory named hdfs:\