openinx / huker Goto Github PK

The config part as following:

    config:
      zoo.cfg:
        - dataDir={{.PkgDataDir}}
        - dataLogDir={{.PkgLogDir}}
        - clientPort=2182
        - tickTime=2000
        - maxClientCnxns=60
        - initLimit=30
        - syncLimit=20
        - maxSessionTimeout=40000
      {{.PkgDataDir}}/myid:

Now we need to put all the properties in xx.yaml, for log config, a better way is to put it into a static file, and the config part can ref to the static file.

For example, we can start a distribute cluster by:

./bin/huker  bootstrap  -f  test-hdfs.yaml 

Cluster.javaHome is actually the main process of the program. after fixing up #32, basically we support all program bootstrapping, but it's a confusing name.
we have two places to change. For configuration,java_home will be changed tomain_process; for the code, javaHome will be mainProcess.

https://github.com/HubSpot/gc_log_visualizer

See https://goreportcard.com/report/github.com/openinx/huker

In method hBootstrapProgram & hCleanupProgram & handleProgram , we can found following code:

if err := handleFunc(prog); err != nil {
	w.Write(renderResp(err))
	return
}

We can abstract a common method to handle the http response (both error case and right case)

Currently, we only render config files by following:

// If skipHostRender is true , will skip to the HostRender
func (c *Cluster) RenderConfigFiles(job *Job, taskId int, skipHostRender bool) (map[string]string, error) {
	//....
}

Need to render all fields in cluster.yaml, such as java_class , extra_args , because when I want to config the following job, it does not work:

  sqlline_thin:
    super_job: job_common
    main_entry:
      java_class: org.apache.phoenix.queryserver.client.SqllineWrapper
      extra_args:
        -d org.apache.phoenix.queryserver.client.Driver
        -u jdbc:phoenix:thin:url=http://%{queryserver.0.host:queryserver.0.base_port};serialization=PROTOBUF
        -n none
        -p none
        --color=true
        --fastConnect=false
        --verbose=true
        --incremental=false
        --isolation=TRANSACTION_READ_COMMITTED

Currently, we do not support the following:

./bin/huker shell hdfs test-hdfs dfs -ls  /hbase 

Aim at deploying Hadoop components efficiently, with built-in monitoring and alerting service and page view.
firstly, we should abstract the deployment procedure, we know a complete requests flow looks like a DAG(directed acyclic graph) , so all the compononts is the point in the graph, and the backend IP will be assumed to be edges.
secondly, we need abstract the specification of single componets, like hdfs node or zookeeper server. docker is the first thing we could use. for monitoring, we install the collecting agent , configuration updating agent in the docker container.

Now, we have a period refresh goroutine to access the program cache of supervisor. and there may be mutiple user-goroutine to access the cache. So need to synchronize between them.
It's a bug and need to fix in release-1.0.

Now, we use the same root directory to store the lib( data/conf etc) of project. A Better way is: different agent can assign different root dir by following:

./bin/huker start-agent  --dir /tmp/rootdir-01 

and their lib/data/conf will store under difference directory. Then, we can start multiple agent processes at a standalone, and it's useful when dev & testing

1. check swap
2. check java
   ...

Now we cann't start a zookeeper cluster in a single node, because current config design can not specific different port for different process with the same job name.

we need to change the host part from:

hosts:
  - 127.0.0.1

to

hosts:
  127.0.0.1/port=9001/id=1:  # host.1
    config:
      zoo.cfg:
        - clientPort=2181
        - server.1=zoo1:2188:3188
  127.0.0.1/port=9001/id=2:  # host.2
    config:
      zoo.cfg:
        - clientPort=2281
        - server.2=zoo1:2288:3288
  127.0.0.1/port=9001/id=3:  # host.3
    config:
      zoo.cfg:
        - clientPort=2381
        - server.3=zoo1:2388:3288

BTW, the base_port part can be removed now.

See https://github.com/openinx/huker/blob/master/prog.go#L187

Currently, we can start a distribute zookeeper cluster under only one host. now, need to test whether we can succeed to build a hdfs cluster based on the existed zookeeper.

BTW, need a common HDFS template file under conf/hdfs/ directory for user to build their own HDFS cluster.

For example

./bin/huker  start-pkg-manager -d 

it will start the pkg-manager process in daemon.

see following:

➜  huker git:(alpha-1.0) ./bin/huker bootstrap  --project zookeeper --cluster test-cluster --job zookeeper 
2018/01/15 07:39:27 [ERROR][github.com/openinx/huker] huker_cli.go:55: Bootstrap job zookeeper at 127.0.0.1 failed, err: {"message":"error: symlink /tmp/huker/agent12/test-cluster/zookeeper/stdout/zookeeper-3.4.11 /tmp/huker/agent12/test-cluster/zookeeper/pkg: file exists","status":1}
➜  huker git:(alpha-1.0) ./bin/huker cleanup  --project zookeeper --cluster test-cluster --job zookeeper 
2018/01/15 07:39:36 [ERROR][github.com/openinx/huker] huker_cli.go:112: Cleanup job zookeeper at 127.0.0.1 failed, err: {"message":"error: name: test-cluster, job: zookeeper not found.","status":1}

Currently, our supervisor will do nothing if call rolling_update. need to implement this.

func (p *Program) rollingUpdate(s *Supervisor) error {
	// TODO
	return nil
}

It'll be useful for unit test.

Will finish this after #11

1. should abstract the zookeeper.znode.parent & hbase.rootdir for differect hbase cluster, otherwise we can not deploy diff hbase cluster ..
2. should update the pkg.yaml configuration now , because the apache official have updated their latest stable binary package .

Currently, we can not render the dependencies's job information to the configuration files. for example, HBase depend on HDFS, and we have to config the namenode address explicitly now ... need to fix this by write a variable, such as

fs.defaultfs:=hdfs://%{dependencies.0.namenode.host}:%{dependencies.0.namenode.port}/hbase/rootdir

Now we have the following in phoenix/common/common.yaml

      hbase-site.xml:
        - hbase.cluster.distributed=true
        - hbase.zookeeper.quorum=%{dependencies.0.zkServer.1.host}:%{dependencies.0.zkServer.1.base_port}
        - zookeeper.znode.parent=/hbase/test-hbase # TODO need to change this to be %{dependencies.0.cluster_name}
      log4j.properties:

Need to fix this.

For hive service, the job name can be hive2 , so the %{hiveserver2.x.base_port} pattern will not be rendered...

In huker, we abstract all services in a single model, so it's easy to design a general dashboard to deploy all services..

Problem encountered:

1. render issue: #31
2. hive depends on hadoop's lib which was not included in release package of hive. so , currently I've to write the constant path in classpath, need to fix this:

classpath:
  - {{.PkgConfDir}}
  - {{.PkgRootDir}}/lib/*
  - {{.PkgRootDir}}/jdbc/*
  - /tmp/huker/.packages/967c24f3c15fcdd058f34923e92ce8ac/hadoop-2.6.5/lib/native/*
  - /tmp/huker/.packages/967c24f3c15fcdd058f34923e92ce8ac/hadoop-2.6.5/share/hadoop/common/*
  - /tmp/huker/.packages/967c24f3c15fcdd058f34923e92ce8ac/hadoop-2.6.5/share/hadoop/common/lib/*
  - /tmp/huker/.packages/967c24f3c15fcdd058f34923e92ce8ac/hadoop-2.6.5/share/hadoop/mapreduce/*
  - /tmp/huker/.packages/967c24f3c15fcdd058f34923e92ce8ac/hadoop-2.6.5/share/hadoop/mapreduce/lib/*
  - /tmp/huker/.packages/967c24f3c15fcdd058f34923e92ce8ac/hadoop-2.6.5/share/hadoop/yarn/*
  - /tmp/huker/.packages/967c24f3c15fcdd058f34923e92ce8ac/hadoop-2.6.5/share/hadoop/yarn/lib/*
  - /tmp/huker/.packages/967c24f3c15fcdd058f34923e92ce8ac/hadoop-2.6.5/share/hadoop/hdfs/*
  - /tmp/huker/.packages/967c24f3c15fcdd058f34923e92ce8ac/hadoop-2.6.5/share/hadoop/hdfs/lib/*

3. in cfg.go , we parsed key value pair by

strings.Split(c.keyValues[i], "=") 

so, for hive case, the hive-site.xml may have the following config:

javax.jdo.option.ConnectionURL=jdbc:derby:;databaseName={{.PkgDataDir}}/metastore_db;create=true

which lead to

if len(parts) != 2 {
	panic(fmt.Sprintf("Invalid key value pair, key or value not found. %s", c.keyValues[i]))
}

4. another problem is , after we start the hiveserver2, and connect to its port by beeline, and create a table, it throws that :

0: jdbc:hive2://localhost:50100> CREATE TABLE u_data (
. . . . . . . . . . . . . . . .>   userid INT,
. . . . . . . . . . . . . . . .>   movieid INT,
. . . . . . . . . . . . . . . .>   rating INT,
. . . . . . . . . . . . . . . .>   unixtime STRING)
. . . . . . . . . . . . . . . .> ROW FORMAT DELIMITED
. . . . . . . . . . . . . . . .> FIELDS TERMINATED BY '\t'
. . . . . . . . . . . . . . . .> STORED AS TEXTFILE;
Error: org.apache.hive.service.cli.HiveSQLException: Error while processing statement: FAILED: Execution Error, return code 1 from org.apache.hadoop.hive.ql.exec.DDLTask. MetaException(message:Got exception: org.apache.hadoop.ipc.RemoteException User: openinx is not allowed to impersonate openinx)
	at org.apache.hive.service.cli.operation.Operation.toSQLException(Operation.java:380)
	at org.apache.hive.service.cli.operation.SQLOperation.runQuery(SQLOperation.java:257)
	at org.apache.hive.service.cli.operation.SQLOperation.access$800(SQLOperation.java:91)
	at org.apache.hive.service.cli.operation.SQLOperation$BackgroundWork$1.run(SQLOperation.java:348)
	at java.security.AccessController.doPrivileged(Native Method)
	at javax.security.auth.Subject.doAs(Subject.java:422)
	at org.apache.hadoop.security.UserGroupInformation.doAs(UserGroupInformation.java:1692)
	at org.apache.hive.service.cli.operation.SQLOperation$BackgroundWork.run(SQLOperation.java:362)
	at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:511)
	at java.util.concurrent.FutureTask.run(FutureTask.java:266)
	at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1142)
	at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:617)
	at java.lang.Thread.run(Thread.java:745)