因为在生产环境中大量使用hive。而hive的权限又较弱，假设可以记录全部hive操作，在增强安全性的同一时候，还可以统计hive表的使用频率；同一时候假设可以记录hql的開始和结束时间，则可以找出系统中花费时间较多的job，针对性的进行优化，因此跟踪hive的使用轨迹，增强安全的同一时候也能方便问题定位。

怎样记录用户操作了？Hive Hook为我们提供的方便的开放接口。

我们对hive的使用主要有两种使用场景，一是平时直接在命令行下运行的hql操作，此时运行hql的实体就是OS的登录用户。第二种是从webapp获取的业务数据需求人员创建定时报表的hql脚本。此时运行hql的真正实体事实上是报表创建者，系统不过代理运行而已，此时记录用户的行为则须要重写hive.security.authenticator.manager。

Hive默认使用HadoopDefaultAuthenticator获取运行hql的用户，使用其返回的用户进行权限验证。

为了使hive可以以代理的模式去运行，我们须要提供自己的authenticator。返回真正的hql运行者。下面配置可设置authenticator：

<property>

<name>hive.security.authenticator.manager</name>

<value>com.pplive.bip.hive.auth.Authenticator</value>

<description>bip user authenticator</description>

</property>

仅仅有管理员能够开启代理模式。能够使用下面方式传递代理用户：

Hive -d bip.user=xxx 或 hive --define bip.user=xxx

重写authenticator代码演示样例：

public
classAuthenticator implements HiveAuthenticationProvider {

private
finalstaticString BIP_USER="bip.user";

privateStringuserName;

privateStringbipUser;

privateList<String>groupNames;

privateConfigurationconf;

@Override

publicList<String> getGroupNames() {

returngroupNames;

}

@Override

publicStringgetUserName() {

this.bipUser = SessionState.get().getHiveVariables().get(BIP_USER);

if(this.bipUser !=null &&!this.bipUser.isEmpty())
{

if( AdminManager.isAdmin(this.userName)) {

returnthis.bipUser;

} else {

thrownewRuntimeException("bip.user is set while youare not admin");

}

} else{

returnthis.userName;

}

}

@Override

publicvoidsetConf(Configuration conf) {

this.conf = conf;

UserGroupInformation ugi = null;

try{

ugi = ShimLoader.getHadoopShims().getUGIForConf(conf);

//     UserGroupInformation.createProxyUser(user, realUser);

} catch(Exception e) {

thrownewRuntimeException(e);

}

if(ugi ==
null){

thrownewRuntimeException(

"Can not initialize PPLive Authenticator.");

}

this.userName = ugi.getUserName();

if(ugi.getGroupNames() !=null) {

this.groupNames = Arrays.asList(ugi.getGroupNames());

}

}

publicString getProxy() {

return 
this.userName;

}

Hive提供的SemanticHook能够方便我们记录hql语义分析前后的状态。Execute Hook能够记录hql翻译成job提交运行前后的状态。 Driver Hook能够记录包含整个编译运行过程前后的状态。

SemanticHook记录语义分析后的行为：

public
voidpostAnalyze(HiveSemanticAnalyzerHookContext context,

List<Task<?

extendsSerializable>> rootTasks)

throws SemanticException {

Hivehive = null;

try {

hive= context.getHive();

}catch(HiveException e) {

e.printStackTrace();

throw
new RuntimeException(e);

}

Set<ReadEntity>inputs = context.getInputs();

Set<WriteEntity>outputs = context.getOutputs();

Set<String>readTables = newHashSet<String>();

for(ReadEntity input :inputs) {

Table table = input.getT();

if(table!=null) {

readTables.add(table.getTableName());

}

}

Set<String>writeTables = newHashSet<String>();

for(WriteEntity output :outputs) {

Table table = output.getT();

if(table!=null) {

writeTables.add(table.getTableName());

}

}

HiveAuthenticationProviderauthenticationProvider = SessionState.get().getAuthenticator();

if(authenticationProviderinstanceof Authenticator) {

Authenticatorauthenticator = (Authenticator)authenticationProvider;       //ip

this.logger.info(String.format("phase=SA&executor=%s&proxy=%s&db=%s&cmd=%s&readTables=%s&writeTables=%s",
authenticator.getUserName(),

authenticator.getProxy(), hive.getCurrentDatabase(),context.getCommand(),readTables.toString(),writeTables.toString()));

}

StringuserName = SessionState.get().getAuthenticator().getUserName();

logger.debug(String.format("%s execute %s, read tables:%s, writetables:%s", userName, context.getCommand(),readTables, writeTables));

}

Execute Hook记录job状态：

public
classExecuteHook implements ExecuteWithHookContext {

Loggerlogger= Logger.getLogger(DriverRunHook.class);

privateHiveAuthenticationProviderauthenticationProvider =
null;

private
static final String JOB_START_TIME="PRE_EXEC_HOOK";

private
static SimpleDateFormat dateFormat =new SimpleDateFormat("yyyy-MM-dd HH:mm:ss");

@Override

public
void run(HookContexthookContext) throwsException {

QueryPlanqueryPlan = hookContext.getQueryPlan();

StringqueryId = queryPlan.getQueryId();

StringqueryStr = queryPlan.getQueryStr();

if(authenticationProvider==null){

authenticationProvider= SessionState.get().getAuthenticator();

}

Stringresult = null;

switch(hookContext.getHookType()){

//hive.exec.pre.hooks

case
PRE_EXEC_HOOK:

hookContext.getConf().setLong(JOB_START_TIME,System.currentTimeMillis());

break;

//hive.exec.post.hooks

case
POST_EXEC_HOOK:

result= "Success";

break;

//hive.exec.failure.hooks

case
ON_FAILURE_HOOK:

result= "Failure";

break;

default:

break;

}

if(hookContext.getHookType()!= HookContext.HookType.PRE_EXEC_HOOK&&authenticationProvider
instanceofAuthenticator) {

long jobEndTime = System.currentTimeMillis();

HiveConfconf = hookContext.getConf();

long jobStartTime =conf.getLong(JOB_START_TIME, jobEndTime);

long timeTaken =(jobEndTime-jobStartTime)/1000;

Authenticatorauthenticator = (Authenticator)authenticationProvider;       //ip

this.logger.info(String.format("phase=EXEC&result=%s&executor=%s&proxy=%s&db=%s&queryId=%s&queryStr=%s&jobName=%s&jobStartTime=%s&jobEndTime=%s&timeTaken=%d",
result,authenticator.getUserName(),authenticator.getProxy(),

Hive.get().getCurrentDatabase(),queryId, queryStr,conf.getVar(HiveConf.ConfVars.HADOOPJOBNAME),dateFormat.format(new
Date(jobStartTime)),

dateFormat.format(newDate(jobEndTime)),timeTaken));

}

}

}

DriverHook记录整个过程运行时间：

public
classDriverRunHook implements HiveDriverRunHook{

Loggerlogger= Logger.getLogger(DriverRunHook.class);

privateHiveAuthenticationProviderauthenticationProvider =
null;

private
static SimpleDateFormat dateFormat =new SimpleDateFormat("yyyy-MM-dd HH:mm:ss");

private
long startTime = 0;

@Override

public
voidpreDriverRun(HiveDriverRunHookContext hookContext)

throws Exception {

if(authenticationProvider==null){

authenticationProvider= SessionState.get().getAuthenticator();

}

startTime = System.currentTimeMillis();

}

@Override

public
voidpostDriverRun(HiveDriverRunHookContext hookContext)

throws Exception {

if(authenticationProviderinstanceofAuthenticator) {

long endTime = System.currentTimeMillis();

long timeTaken = (endTime-startTime)/1000;

Authenticatorauthenticator = (Authenticator)authenticationProvider;       //ip

this.logger.info(String.format("phase=DriverRun&executor=%s&proxy=%s&db=%s&cmd=%s&startTime=%s&endTime=%s&timeTaken=%d",
authenticator.getUserName(),authenticator.getProxy(),

Hive.get().getCurrentDatabase(),hookContext.getCommand(),dateFormat.format(newDate(startTime)),dateFormat.format(new
Date(endTime)),timeTaken));

}

}

}