Impala
Impalaistheopensource,nativeanalyticdatabaseforApacheHadoop.
 All Classes Namespaces Files Functions Variables Typedefs Enumerations Enumerator Friends Macros
impala::HiveUdfCall Class Reference

If the UDF ran into an error, the FE throws an exception. More...

#include <hive-udf-call.h>

Inheritance diagram for impala::HiveUdfCall:
Collaboration diagram for impala::HiveUdfCall:

Public Member Functions

virtual Status Prepare (RuntimeState *state, const RowDescriptor &row_desc, ExprContext *ctx)
 
virtual Status Open (RuntimeState *state, ExprContext *context, FunctionContext::FunctionStateScope scope=FunctionContext::FRAGMENT_LOCAL)
 
virtual void Close (RuntimeState *state, ExprContext *context, FunctionContext::FunctionStateScope scope=FunctionContext::FRAGMENT_LOCAL)
 Subclasses overriding this function should call Expr::Close(). More...
 
virtual BooleanVal GetBooleanVal (ExprContext *ctx, TupleRow *)
 
virtual TinyIntVal GetTinyIntVal (ExprContext *ctx, TupleRow *)
 
virtual SmallIntVal GetSmallIntVal (ExprContext *ctx, TupleRow *)
 
virtual IntVal GetIntVal (ExprContext *ctx, TupleRow *)
 
virtual BigIntVal GetBigIntVal (ExprContext *ctx, TupleRow *)
 
virtual FloatVal GetFloatVal (ExprContext *ctx, TupleRow *)
 
virtual DoubleVal GetDoubleVal (ExprContext *ctx, TupleRow *)
 
virtual StringVal GetStringVal (ExprContext *ctx, TupleRow *)
 
virtual TimestampVal GetTimestampVal (ExprContext *ctx, TupleRow *)
 
virtual DecimalVal GetDecimalVal (ExprContext *ctx, TupleRow *)
 
virtual Status GetCodegendComputeFn (RuntimeState *state, llvm::Function **fn)
 
int output_scale () const
 
void AddChild (Expr *expr)
 
ExprGetChild (int i) const
 
int GetNumChildren () const
 
const ColumnTypetype () const
 
bool is_slotref () const
 
const std::vector< Expr * > & children () const
 
virtual bool IsConstant () const
 
virtual int GetSlotIds (std::vector< SlotId > *slot_ids) const
 
virtual AnyValGetConstVal (ExprContext *context)
 

Static Public Member Functions

static Status CreateExprTree (ObjectPool *pool, const TExpr &texpr, ExprContext **ctx)
 
static Status CreateExprTrees (ObjectPool *pool, const std::vector< TExpr > &texprs, std::vector< ExprContext * > *ctxs)
 
static Status Prepare (const std::vector< ExprContext * > &ctxs, RuntimeState *state, const RowDescriptor &row_desc, MemTracker *tracker)
 
static Status Open (const std::vector< ExprContext * > &ctxs, RuntimeState *state)
 Convenience function for opening multiple expr trees. More...
 
static Status Clone (const std::vector< ExprContext * > &ctxs, RuntimeState *state, std::vector< ExprContext * > *new_ctxs)
 
static void Close (const std::vector< ExprContext * > &ctxs, RuntimeState *state)
 Convenience function for closing multiple expr trees. More...
 
static ExprCreateLiteral (ObjectPool *pool, const ColumnType &type, void *data)
 
static ExprCreateLiteral (ObjectPool *pool, const ColumnType &type, const std::string &)
 
static int ComputeResultsLayout (const std::vector< Expr * > &exprs, std::vector< int > *offsets, int *var_result_begin)
 
static int ComputeResultsLayout (const std::vector< ExprContext * > &ctxs, std::vector< int > *offsets, int *var_result_begin)
 
static std::string DebugString (const std::vector< Expr * > &exprs)
 
static std::string DebugString (const std::vector< ExprContext * > &ctxs)
 
static void InitBuiltinsDummy ()
 

Static Public Attributes

static const char * LLVM_CLASS_NAME = "class.impala::Expr"
 

Protected Member Functions

 HiveUdfCall (const TExprNode &node)
 
virtual std::string DebugString () const
 
std::string DebugString (const std::string &expr_name) const
 Simple debug string that provides no expr subclass-specific information. More...
 
FunctionContextRegisterFunctionContext (ExprContext *ctx, RuntimeState *state, int varargs_buffer_size=0)
 
llvm::Function * CreateIrFunctionPrototype (LlvmCodeGen *codegen, const std::string &name, llvm::Value *(*args)[2])
 
Status GetCodegendComputeFnWrapper (RuntimeState *state, llvm::Function **fn)
 
llvm::Function * GetStaticGetValWrapper (ColumnType type, LlvmCodeGen *codegen)
 

Protected Attributes

LibCache::LibCacheEntrycache_entry_
 Cache entry for the library implementing this function. More...
 
TFunction fn_
 Function description. More...
 
const bool is_slotref_
 recognize if this node is a slotref in order to speed up GetValue() More...
 
const ColumnType type_
 analysis is done, types are fixed at this point More...
 
std::vector< Expr * > children_
 
int output_scale_
 
int context_index_
 
llvm::Function * ir_compute_fn_
 Cached codegened compute function. Exprs should set this in GetCodegendComputeFn(). More...
 
boost::scoped_ptr< AnyValconstant_val_
 

Private Member Functions

AnyValEvaluate (ExprContext *ctx, TupleRow *row)
 

Private Attributes

std::string local_location_
 The path on the local FS to the UDF's jar. More...
 
std::vector< int > input_byte_offsets_
 
int input_buffer_size_
 The size of the buffer for passing in input arguments. More...
 

Friends

class Expr
 
class StringFunctions
 

Detailed Description

If the UDF ran into an error, the FE throws an exception.

Executor for hive udfs using JNI. This works with the UdfExecutor on the java side which calls into the actual UDF. To minimize the JNI overhead, we eliminate as many copies as possible and share memory between the native side and java side. Memory in the native heap can be read with no issues from java but not vice versa (ptrs in the java heap move). Also, JNI calls are cheaper for function calls with no arguments and no return value (void). During Prepare(), we allocate an input buffer that is big enough to store all of the inputs (i.e. the slot size). This buffer is passed to the UdfExecutor in the constructor. During Evaluate(), the input buffer is populated and the UdfExecutor.evaluate() method is called via JNI. For input arguments, strings don't need to be treated any differently. The java side can parse the ptr and length from the StringValue and then read the ptr directly. For return values that are fixed size (i.e. not strings), we allocate an output buffer in Prepare(). This is also passed to the UdfExecutor in the constructor. The UdfExecutor writes to it directly during evaluate(). For strings, we pass a StringValue sized output buffer to the FE. The address of the StringValue does not change. When the FE writes the string result, it populates the StringValue with the buffer it allocated from its native heap. The BE reads the StringValue as normal.

Definition at line 57 of file hive-udf-call.h.

Constructor & Destructor Documentation

impala::HiveUdfCall::HiveUdfCall ( const TExprNode &  node)
protected

Definition at line 67 of file hive-udf-call.cc.

Member Function Documentation

void impala::Expr::AddChild ( Expr expr)
inlineinherited

Definition at line 141 of file expr.h.

Referenced by impala::Expr::CreateTreeFromThrift().

const std::vector<Expr*>& impala::Expr::children ( ) const
inlineinherited
Status Expr::Clone ( const std::vector< ExprContext * > &  ctxs,
RuntimeState state,
std::vector< ExprContext * > *  new_ctxs 
)
staticinherited

Clones each ExprContext for multiple expr trees. 'new_ctxs' should be an empty vector, and a clone of each context in 'ctxs' will be added to it. The new ExprContexts are created in state->obj_pool().

Definition at line 374 of file expr.cc.

References impala::Status::OK, and RETURN_IF_ERROR.

Referenced by impala::HdfsScanNode::GetConjunctCtxs(), and impala::SortExecExprs::Open().

void impala::HiveUdfCall::Close ( RuntimeState state,
ExprContext context,
FunctionContext::FunctionStateScope  scope = FunctionContext::FRAGMENT_LOCAL 
)
virtual
static int impala::Expr::ComputeResultsLayout ( const std::vector< Expr * > &  exprs,
std::vector< int > *  offsets,
int *  var_result_begin 
)
staticinherited

Computes a memory efficient layout for storing the results of evaluating 'exprs' Returns the number of bytes necessary to store all the results and offsets where the result for each expr should be stored. Variable length types are guaranteed to be at the end and 'var_result_begin' will be set the beginning byte offset where variable length results begin. 'var_result_begin' will be set to -1 if there are no variable len types.

Referenced by impala::HashTableCtx::HashTableCtx(), and impala::OldHashTable::OldHashTable().

static int impala::Expr::ComputeResultsLayout ( const std::vector< ExprContext * > &  ctxs,
std::vector< int > *  offsets,
int *  var_result_begin 
)
staticinherited
Status Expr::CreateExprTree ( ObjectPool pool,
const TExpr &  texpr,
ExprContext **  ctx 
)
staticinherited
Status Expr::CreateExprTrees ( ObjectPool pool,
const std::vector< TExpr > &  texprs,
std::vector< ExprContext * > *  ctxs 
)
staticinherited

Creates vector of ExprContexts containing exprs from the given vector of TExprs within 'pool'. Returns an error if any of the individual conversions caused an error, otherwise OK.

Definition at line 149 of file expr.cc.

References impala::Expr::CreateExprTree(), impala::Status::OK, and RETURN_IF_ERROR.

Referenced by impala::DataStreamSender::DataStreamSender(), impala::UnionNode::Init(), impala::HashJoinNode::Init(), impala::AggregationNode::Init(), impala::ExecNode::Init(), impala::PartitionedHashJoinNode::Init(), impala::PartitionedAggregationNode::Init(), impala::HdfsScanNode::Prepare(), and impala::HdfsTableSink::Prepare().

Function * Expr::CreateIrFunctionPrototype ( LlvmCodeGen codegen,
const std::string &  name,
llvm::Value *(*)  args[2] 
)
protectedinherited
static Expr* impala::Expr::CreateLiteral ( ObjectPool pool,
const ColumnType type,
void *  data 
)
staticinherited

Create a new literal expr of 'type' with initial 'data'. data should match the ColumnType (i.e. type == TYPE_INT, data is a int*) The new Expr will be allocated from the pool.

static Expr* impala::Expr::CreateLiteral ( ObjectPool pool,
const ColumnType type,
const std::string &   
)
staticinherited

Create a new literal expr of 'type' by parsing the string. NULL will be returned if the string and type are not compatible. The new Expr will be allocated from the pool.

string impala::HiveUdfCall::DebugString ( ) const
protectedvirtual

Reimplemented from impala::Expr.

Definition at line 275 of file hive-udf-call.cc.

References impala::Expr::DebugString(), and impala::Expr::fn_.

static std::string impala::Expr::DebugString ( const std::vector< Expr * > &  exprs)
staticinherited
static std::string impala::Expr::DebugString ( const std::vector< ExprContext * > &  ctxs)
staticinherited
std::string impala::Expr::DebugString ( const std::string &  expr_name) const
inlineprotectedinherited

Simple debug string that provides no expr subclass-specific information.

Definition at line 332 of file expr.h.

References impala_udf::DebugString().

AnyVal * impala::HiveUdfCall::Evaluate ( ExprContext ctx,
TupleRow row 
)
private
BigIntVal impala::HiveUdfCall::GetBigIntVal ( ExprContext ctx,
TupleRow row 
)
virtual

Reimplemented from impala::Expr.

Definition at line 303 of file hive-udf-call.cc.

References Evaluate(), impala::ColumnType::type, impala::Expr::type_, and impala::TYPE_BIGINT.

BooleanVal impala::HiveUdfCall::GetBooleanVal ( ExprContext context,
TupleRow row 
)
virtual

Virtual compute functions for each *Val type. Each Expr subclass should implement the functions for the return type(s) it supports. For example, a boolean function will only implement GetBooleanVal(). Some Exprs, like Literal, have many possible return types and will implement multiple Get*Val() functions.

Reimplemented from impala::Expr.

Definition at line 283 of file hive-udf-call.cc.

References Evaluate(), impala::ColumnType::type, impala::Expr::type_, and impala::TYPE_BOOLEAN.

Expr* impala::Expr::GetChild ( int  i) const
inlineinherited

Definition at line 142 of file expr.h.

Referenced by Evaluate(), and Prepare().

Status impala::HiveUdfCall::GetCodegendComputeFn ( RuntimeState state,
llvm::Function **  fn 
)
virtual

Returns an llvm::Function* with signature: <subclass of="" anyval>=""> ComputeFn(ExprContext* context, TupleRow* row) The function should evaluate this expr over 'row' and return the result as the appropriate type of AnyVal.

Implements impala::Expr.

Definition at line 271 of file hive-udf-call.cc.

References impala::Expr::GetCodegendComputeFnWrapper().

Status Expr::GetCodegendComputeFnWrapper ( RuntimeState state,
llvm::Function **  fn 
)
protectedinherited

Generates an IR compute function that calls the appropriate interpreted Get*Val() compute function. This is useful for builtins that can't be implemented with the UDF interface (e.g. functions that need short-circuiting) and that don't have custom codegen functions that use the IRBuilder. It doesn't provide any performance benefit over the interpreted path. TODO: this should be replaced with fancier xcompiling infrastructure

Definition at line 546 of file expr.cc.

References impala::LlvmCodeGen::CastPtrToLlvmPtr(), impala::LlvmCodeGen::context(), impala::CodegenAnyVal::CreateCall(), impala::Expr::CreateIrFunctionPrototype(), impala::LlvmCodeGen::FinalizeFunction(), impala::RuntimeState::GetCodegen(), impala::LlvmCodeGen::GetPtrType(), impala::Expr::GetStaticGetValWrapper(), impala::Expr::ir_compute_fn_, impala::Expr::LLVM_CLASS_NAME, impala::Status::OK, RETURN_IF_ERROR, and impala::Expr::type().

Referenced by impala::TupleIsNullPredicate::GetCodegendComputeFn(), and GetCodegendComputeFn().

DecimalVal impala::HiveUdfCall::GetDecimalVal ( ExprContext ctx,
TupleRow row 
)
virtual

Reimplemented from impala::Expr.

Definition at line 328 of file hive-udf-call.cc.

References Evaluate(), impala::ColumnType::type, impala::Expr::type_, and impala::TYPE_DECIMAL.

DoubleVal impala::HiveUdfCall::GetDoubleVal ( ExprContext ctx,
TupleRow row 
)
virtual

Reimplemented from impala::Expr.

Definition at line 313 of file hive-udf-call.cc.

References Evaluate(), impala::ColumnType::type, impala::Expr::type_, and impala::TYPE_DOUBLE.

FloatVal impala::HiveUdfCall::GetFloatVal ( ExprContext ctx,
TupleRow row 
)
virtual

Reimplemented from impala::Expr.

Definition at line 308 of file hive-udf-call.cc.

References Evaluate(), impala::ColumnType::type, impala::Expr::type_, and impala::TYPE_FLOAT.

IntVal impala::HiveUdfCall::GetIntVal ( ExprContext ctx,
TupleRow row 
)
virtual

Reimplemented from impala::Expr.

Definition at line 298 of file hive-udf-call.cc.

References Evaluate(), impala::ColumnType::type, impala::Expr::type_, and impala::TYPE_INT.

int Expr::GetSlotIds ( std::vector< SlotId > *  slot_ids) const
virtualinherited

Returns the slots that are referenced by this expr tree in 'slot_ids'. Returns the number of slots added to the vector

Reimplemented in impala::SlotRef.

Definition at line 418 of file expr.cc.

References impala::Expr::children_.

SmallIntVal impala::HiveUdfCall::GetSmallIntVal ( ExprContext ctx,
TupleRow row 
)
virtual

Reimplemented from impala::Expr.

Definition at line 293 of file hive-udf-call.cc.

References Evaluate(), impala::ColumnType::type, impala::Expr::type_, and impala::TYPE_SMALLINT.

Function * Expr::GetStaticGetValWrapper ( ColumnType  type,
LlvmCodeGen codegen 
)
protectedinherited

Returns the IR version of the static Get*Val() wrapper function corresponding to 'type'. This is used for calling interpreted Get*Val() functions from codegen'd functions (e.g. in ScalarFnCall() when codegen is disabled).

Definition at line 426 of file expr.cc.

References impala::ColumnType::DebugString(), impala::LlvmCodeGen::GetFunction(), impala::ColumnType::type, impala::TYPE_BIGINT, impala::TYPE_BOOLEAN, impala::TYPE_CHAR, impala::TYPE_DECIMAL, impala::TYPE_DOUBLE, impala::TYPE_FLOAT, impala::TYPE_INT, impala::TYPE_SMALLINT, impala::TYPE_STRING, impala::TYPE_TIMESTAMP, impala::TYPE_TINYINT, and impala::TYPE_VARCHAR.

Referenced by impala::ScalarFnCall::GetCodegendComputeFn(), and impala::Expr::GetCodegendComputeFnWrapper().

StringVal impala::HiveUdfCall::GetStringVal ( ExprContext ctx,
TupleRow row 
)
virtual

Reimplemented from impala::Expr.

Definition at line 318 of file hive-udf-call.cc.

References Evaluate(), impala::ColumnType::type, impala::Expr::type_, and impala::TYPE_STRING.

TimestampVal impala::HiveUdfCall::GetTimestampVal ( ExprContext ctx,
TupleRow row 
)
virtual

Reimplemented from impala::Expr.

Definition at line 323 of file hive-udf-call.cc.

References Evaluate(), impala::ColumnType::type, impala::Expr::type_, and impala::TYPE_TIMESTAMP.

TinyIntVal impala::HiveUdfCall::GetTinyIntVal ( ExprContext ctx,
TupleRow row 
)
virtual

Reimplemented from impala::Expr.

Definition at line 288 of file hive-udf-call.cc.

References Evaluate(), impala::ColumnType::type, impala::Expr::type_, and impala::TYPE_TINYINT.

void Expr::InitBuiltinsDummy ( )
staticinherited
bool impala::Expr::is_slotref ( ) const
inlineinherited

Definition at line 146 of file expr.h.

bool Expr::IsConstant ( ) const
virtualinherited

Returns true if GetValue(NULL) can be called on this expr and always returns the same result (e.g., exprs that don't contain slotrefs). The default implementation returns true if all children are constant.

Reimplemented in impala::ScalarFnCall, impala::TupleIsNullPredicate, and impala::SlotRef.

Definition at line 411 of file expr.cc.

References impala::Expr::children_.

Referenced by impala::Expr::GetConstVal(), and impala::ScalarFnCall::IsConstant().

Status impala::HiveUdfCall::Open ( RuntimeState state,
ExprContext context,
FunctionContext::FunctionStateScope  scope = FunctionContext::FRAGMENT_LOCAL 
)
virtual
int impala::Expr::output_scale ( ) const
inlineinherited

Get the number of digits after the decimal that should be displayed for this value. Returns -1 if no scale has been specified (currently the scale is only set for doubles set by RoundUpTo). GetValue() must have already been called. TODO: is this still necessary?

Definition at line 139 of file expr.h.

Status impala::HiveUdfCall::Prepare ( RuntimeState state,
const RowDescriptor row_desc,
ExprContext context 
)
virtual

Initializes this expr instance for execution. This does not include initializing state in the ExprContext; 'context' should only be used to register a FunctionContext via RegisterFunctionContext(). Any IR functions must be generated here. Subclasses overriding this function should call Expr::Prepare() to recursively call Prepare() on the expr tree.

Reimplemented from impala::Expr.

Definition at line 153 of file hive-udf-call.cc.

References impala::Expr::fn_, impala::Expr::GetChild(), impala::Expr::GetNumChildren(), impala::ColumnType::GetSlotSize(), input_buffer_size_, input_byte_offsets_, impala::LibCache::instance(), local_location_, impala::Status::OK, impala::Expr::Prepare(), impala::Expr::RegisterFunctionContext(), RETURN_IF_ERROR, impala::BitUtil::RoundUpNumBytes(), impala::Expr::type(), and impala::LibCache::TYPE_JAR.

FunctionContext * Expr::RegisterFunctionContext ( ExprContext ctx,
RuntimeState state,
int  varargs_buffer_size = 0 
)
protectedinherited

Helper function that calls ctx->Register(), sets context_index_, and returns the registered FunctionContext.

Definition at line 80 of file expr.cc.

References impala::AnyValUtil::ColumnTypeToTypeDesc(), impala::ExprContext::fn_context(), and impala::ExprContext::Register().

Referenced by impala::CaseExpr::Prepare(), and Prepare().

Friends And Related Function Documentation

friend class Expr
friend

Definition at line 80 of file hive-udf-call.h.

friend class StringFunctions
friend

Definition at line 81 of file hive-udf-call.h.

Member Data Documentation

LibCache::LibCacheEntry* impala::Expr::cache_entry_
protectedinherited

Cache entry for the library implementing this function.

Definition at line 281 of file expr.h.

Referenced by impala::Expr::Close(), impala::ScalarFnCall::GetFunction(), impala::ScalarFnCall::GetUdf(), impala::ScalarFnCall::Prepare(), and impala::Expr::~Expr().

boost::scoped_ptr<AnyVal> impala::Expr::constant_val_
protectedinherited

If this expr is constant, this will store and cache the value generated by GetConstVal().

Definition at line 303 of file expr.h.

Referenced by impala::Expr::GetConstVal().

int impala::Expr::context_index_
protectedinherited
int impala::HiveUdfCall::input_buffer_size_
private

The size of the buffer for passing in input arguments.

Definition at line 100 of file hive-udf-call.h.

Referenced by Open(), and Prepare().

std::vector<int> impala::HiveUdfCall::input_byte_offsets_
private

input_byte_offsets_[i] is the byte offset child ith's input argument should be written to.

Definition at line 97 of file hive-udf-call.h.

Referenced by Evaluate(), Open(), and Prepare().

llvm::Function* impala::Expr::ir_compute_fn_
protectedinherited
const bool impala::Expr::is_slotref_
protectedinherited

recognize if this node is a slotref in order to speed up GetValue()

Definition at line 287 of file expr.h.

const char * Expr::LLVM_CLASS_NAME = "class.impala::Expr"
staticinherited
std::string impala::HiveUdfCall::local_location_
private

The path on the local FS to the UDF's jar.

Definition at line 93 of file hive-udf-call.h.

Referenced by Open(), and Prepare().

int impala::Expr::output_scale_
protectedinherited

Definition at line 291 of file expr.h.

Referenced by impala::ExprContext::GetValue(), and impala::ScalarFnCall::Open().

const ColumnType impala::Expr::type_
protectedinherited

analysis is done, types are fixed at this point

Definition at line 289 of file expr.h.

Referenced by impala::Literal::DebugString(), impala::Expr::DebugString(), impala::SlotRef::GetBigIntVal(), impala::Literal::GetBigIntVal(), impala::ScalarFnCall::GetBigIntVal(), GetBigIntVal(), impala::SlotRef::GetBooleanVal(), impala::Literal::GetBooleanVal(), GetBooleanVal(), impala::ScalarFnCall::GetBooleanVal(), impala::SlotRef::GetCodegendComputeFn(), impala::Literal::GetCodegendComputeFn(), impala::Expr::GetConstVal(), impala::SlotRef::GetDecimalVal(), impala::Literal::GetDecimalVal(), impala::ScalarFnCall::GetDecimalVal(), GetDecimalVal(), impala::SlotRef::GetDoubleVal(), impala::Literal::GetDoubleVal(), impala::ScalarFnCall::GetDoubleVal(), GetDoubleVal(), impala::SlotRef::GetFloatVal(), impala::Literal::GetFloatVal(), impala::ScalarFnCall::GetFloatVal(), GetFloatVal(), impala::SlotRef::GetIntVal(), impala::Literal::GetIntVal(), impala::ScalarFnCall::GetIntVal(), GetIntVal(), impala::SlotRef::GetSmallIntVal(), impala::Literal::GetSmallIntVal(), impala::ScalarFnCall::GetSmallIntVal(), GetSmallIntVal(), impala::SlotRef::GetStringVal(), impala::Literal::GetStringVal(), GetStringVal(), impala::ScalarFnCall::GetStringVal(), impala::SlotRef::GetTimestampVal(), GetTimestampVal(), impala::ScalarFnCall::GetTimestampVal(), impala::SlotRef::GetTinyIntVal(), impala::Literal::GetTinyIntVal(), impala::ScalarFnCall::GetTinyIntVal(), GetTinyIntVal(), impala::ExprContext::GetValue(), impala::Literal::Literal(), impala::ScalarFnCall::Open(), Open(), impala::ScalarFnCall::Prepare(), and impala::Expr::Prepare().


The documentation for this class was generated from the following files: