Hive SQL解析过程
SQL->AST(Abstract Syntax Tree)->Task(MapRedTask,FetchTask)->QueryPlan(Task集合)->Job(Yarn)
SQL解析会在两个地方进行:
- 一个是SQL执行前compile,具体在Driver.compile,为了创建QueryPlan;
- 一个是explain,具体在ExplainSemanticAnalyzer.analyzeInternal,为了创建ExplainTask;
SQL执行过程
1 compile过程(SQL->AST(Abstract Syntax Tree)->QueryPlan)
org.apache.hadoop.hive.ql.Driver
public int compile(String command, boolean resetTaskIds, boolean deferClose) {
...
ParseDriver pd = new ParseDriver();
ASTNode tree = pd.parse(command, ctx);
tree = ParseUtils.findRootNonNullToken(tree);
...
BaseSemanticAnalyzer sem = SemanticAnalyzerFactory.get(queryState, tree);
...
sem.analyze(tree, ctx);
...
// Record any ACID compliant FileSinkOperators we saw so we can add our transaction ID to
// them later.
acidSinks = sem.getAcidFileSinks(); LOG.info("Semantic Analysis Completed"); // validate the plan
sem.validate();
acidInQuery = sem.hasAcidInQuery();
perfLogger.PerfLogEnd(CLASS_NAME, PerfLogger.ANALYZE); if (isInterrupted()) {
return handleInterruption("after analyzing query.");
} // get the output schema
schema = getSchema(sem, conf);
plan = new QueryPlan(queryStr, sem, perfLogger.getStartTime(PerfLogger.DRIVER_RUN), queryId,
queryState.getHiveOperation(), schema);
...
compile过程为先由ParseDriver将SQL转换为ASTNode,然后由BaseSemanticAnalyzer对ASTNode进行分析,最后将BaseSemanticAnalyzer传入QueryPlan构造函数来创建QueryPlan;
1)将SQL转换为ASTNode过程如下(SQL->AST(Abstract Syntax Tree))
org.apache.hadoop.hive.ql.parse.ParseDriver
public ASTNode parse(String command, Context ctx, boolean setTokenRewriteStream)
throws ParseException {
if (LOG.isDebugEnabled()) {
LOG.debug("Parsing command: " + command);
} HiveLexerX lexer = new HiveLexerX(new ANTLRNoCaseStringStream(command));
TokenRewriteStream tokens = new TokenRewriteStream(lexer);
if (ctx != null) {
if ( setTokenRewriteStream) {
ctx.setTokenRewriteStream(tokens);
}
lexer.setHiveConf(ctx.getConf());
}
HiveParser parser = new HiveParser(tokens);
if (ctx != null) {
parser.setHiveConf(ctx.getConf());
}
parser.setTreeAdaptor(adaptor);
HiveParser.statement_return r = null;
try {
r = parser.statement();
} catch (RecognitionException e) {
e.printStackTrace();
throw new ParseException(parser.errors);
} if (lexer.getErrors().size() == 0 && parser.errors.size() == 0) {
LOG.debug("Parse Completed");
} else if (lexer.getErrors().size() != 0) {
throw new ParseException(lexer.getErrors());
} else {
throw new ParseException(parser.errors);
} ASTNode tree = (ASTNode) r.getTree();
tree.setUnknownTokenBoundaries();
return tree;
}
2)analyze过程(AST(Abstract Syntax Tree)->Task)
org.apache.hadoop.hive.ql.parse.BaseSemanticAnalyzer
public void analyze(ASTNode ast, Context ctx) throws SemanticException {
initCtx(ctx);
init(true);
analyzeInternal(ast);
}
其中analyzeInternal是抽象方法,由不同的子类实现,比如DDLSemanticAnalyzer,SemanticAnalyzer,UpdateDeleteSemanticAnalyzer,ExplainSemanticAnalyzer等;
analyzeInternal主要的工作是将ASTNode转化为Task,包括可能的optimize,过程比较复杂,这里不贴代码;
3)创建QueryPlan过程如下(Task->QueryPlan)
org.apache.hadoop.hive.ql.QueryPlan
public QueryPlan(String queryString, BaseSemanticAnalyzer sem, Long startTime, String queryId,
HiveOperation operation, Schema resultSchema) {
this.queryString = queryString; rootTasks = new ArrayList<Task<? extends Serializable>>(sem.getAllRootTasks());
reducerTimeStatsPerJobList = new ArrayList<ReducerTimeStatsPerJob>();
fetchTask = sem.getFetchTask();
// Note that inputs and outputs can be changed when the query gets executed
inputs = sem.getAllInputs();
outputs = sem.getAllOutputs();
linfo = sem.getLineageInfo();
tableAccessInfo = sem.getTableAccessInfo();
columnAccessInfo = sem.getColumnAccessInfo();
idToTableNameMap = new HashMap<String, String>(sem.getIdToTableNameMap()); this.queryId = queryId == null ? makeQueryId() : queryId;
query = new org.apache.hadoop.hive.ql.plan.api.Query();
query.setQueryId(this.queryId);
query.putToQueryAttributes("queryString", this.queryString);
queryProperties = sem.getQueryProperties();
queryStartTime = startTime;
this.operation = operation;
this.autoCommitValue = sem.getAutoCommitValue();
this.resultSchema = resultSchema;
}
可见只是简单的将BaseSemanticAnalyzer中的内容拷贝出来,其中最重要的是sem.getAllRootTasks和sem.getFetchTask;
2 execute过程(QueryPlan->Job)
org.apache.hadoop.hive.ql.Driver
public int execute(boolean deferClose) throws CommandNeedRetryException {
...
// Add root Tasks to runnable
for (Task<? extends Serializable> tsk : plan.getRootTasks()) {
// This should never happen, if it does, it's a bug with the potential to produce
// incorrect results.
assert tsk.getParentTasks() == null || tsk.getParentTasks().isEmpty();
driverCxt.addToRunnable(tsk);
}
...
// Loop while you either have tasks running, or tasks queued up
while (driverCxt.isRunning()) { // Launch upto maxthreads tasks
Task<? extends Serializable> task;
while ((task = driverCxt.getRunnable(maxthreads)) != null) {
TaskRunner runner = launchTask(task, queryId, noName, jobname, jobs, driverCxt);
if (!runner.isRunning()) {
break;
}
}
... private TaskRunner launchTask(Task<? extends Serializable> tsk, String queryId, boolean noName,
String jobname, int jobs, DriverContext cxt) throws HiveException {
...
TaskRunner tskRun = new TaskRunner(tsk, tskRes);
...
tskRun.start();
...
tskRun.runSequential();
...
Driver.run中从QueryPlan中取出Task,并逐个launchTask,launchTask过程为将Task包装为TaskRunner,并最终调用TaskRunner.runSequential,下面看TaskRunner:
org.apache.hadoop.hive.ql.exec.TaskRunner
public void runSequential() {
int exitVal = -101;
try {
exitVal = tsk.executeTask();
...
这里直接调用Task.executeTask
org.apache.hadoop.hive.ql.exec.Task
public int executeTask() {
...
int retval = execute(driverContext);
...
这里execute是抽象方法,由子类实现,比如DDLTask,MapRedTask等,着重看MapRedTask,因为大部分的Task都是MapRedTask:
org.apache.hadoop.hive.ql.exec.mr.MapRedTask
public int execute(DriverContext driverContext) {
...
if (!runningViaChild) {
// we are not running this mapred task via child jvm
// so directly invoke ExecDriver
return super.execute(driverContext);
}
...
这里直接调用父类方法,也就是ExecDriver.execute,下面看:
org.apache.hadoop.hive.ql.exec.mr.ExecDriver
protected transient JobConf job;
...
public int execute(DriverContext driverContext) {
...
JobClient jc = null; MapWork mWork = work.getMapWork();
ReduceWork rWork = work.getReduceWork();
...
if (mWork.getNumMapTasks() != null) {
job.setNumMapTasks(mWork.getNumMapTasks().intValue());
}
...
job.setNumReduceTasks(rWork != null ? rWork.getNumReduceTasks().intValue() : 0);
job.setReducerClass(ExecReducer.class);
...
jc = new JobClient(job);
...
rj = jc.submitJob(job);
this.jobID = rj.getJobID();
...
这里将Task转化为Job提交到Yarn执行;
SQL Explain过程
另外一个SQL解析的过程是explain,在ExplainSemanticAnalyzer中将ASTNode转化为ExplainTask:
org.apache.hadoop.hive.ql.parse.ExplainSemanticAnalyzer
public void analyzeInternal(ASTNode ast) throws SemanticException {
...
ctx.setExplain(true);
ctx.setExplainLogical(logical); // Create a semantic analyzer for the query
ASTNode input = (ASTNode) ast.getChild(0);
BaseSemanticAnalyzer sem = SemanticAnalyzerFactory.get(queryState, input);
sem.analyze(input, ctx);
sem.validate(); ctx.setResFile(ctx.getLocalTmpPath());
List<Task<? extends Serializable>> tasks = sem.getAllRootTasks();
if (tasks == null) {
tasks = Collections.emptyList();
} FetchTask fetchTask = sem.getFetchTask();
if (fetchTask != null) {
// Initialize fetch work such that operator tree will be constructed.
fetchTask.getWork().initializeForFetch(ctx.getOpContext());
} ParseContext pCtx = null;
if (sem instanceof SemanticAnalyzer) {
pCtx = ((SemanticAnalyzer)sem).getParseContext();
} boolean userLevelExplain = !extended
&& !formatted
&& !dependency
&& !logical
&& !authorize
&& (HiveConf.getBoolVar(ctx.getConf(), HiveConf.ConfVars.HIVE_EXPLAIN_USER) && HiveConf
.getVar(conf, HiveConf.ConfVars.HIVE_EXECUTION_ENGINE).equals("tez"));
ExplainWork work = new ExplainWork(ctx.getResFile(),
pCtx,
tasks,
fetchTask,
sem,
extended,
formatted,
dependency,
logical,
authorize,
userLevelExplain,
ctx.getCboInfo()); work.setAppendTaskType(
HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVEEXPLAINDEPENDENCYAPPENDTASKTYPES)); ExplainTask explTask = (ExplainTask) TaskFactory.get(work, conf); fieldList = explTask.getResultSchema();
rootTasks.add(explTask);
}