wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-06-30 20:18:33 +08:00
Code Issues Packages Projects Releases Wiki Activity

renaming project, halo -> ray (#95)

Browse Source
This commit is contained in:
Robert Nishihara
2016-06-10 14:12:15 -07:00
committed by Philipp Moritz
parent 44ae1788ee
commit 4cc024ae36
41 changed files with 453 additions and 481 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/computation_graph.cc
+3 -3
View File
@@ -4,7 +4,7 @@ OperationId ComputationGraph::add_operation(std::unique_ptr<Operation> operation
OperationId operationid = operations_.size();
OperationId creator_operationid = operation->creator_operationid();
if (spawned_operations_.size() != operationid) {
HALO_LOG(HALO_FATAL, "ComputationGraph is attempting to call add_operation, but spawned_operations_.size() != operationid.");
RAY_LOG(RAY_FATAL, "ComputationGraph is attempting to call add_operation, but spawned_operations_.size() != operationid.");
}
operations_.emplace_back(std::move(operation));
if (creator_operationid != NO_OPERATION && creator_operationid != ROOT_OPERATION) {
@@ -16,10 +16,10 @@ OperationId ComputationGraph::add_operation(std::unique_ptr<Operation> operation
const Task& ComputationGraph::get_task(OperationId operationid) {
if (operationid >= operations_.size()) {
HALO_LOG(HALO_FATAL, "ComputationGraph attempting to get_task with operationid " << operationid << ", but operationid >= operations_.size().");
RAY_LOG(RAY_FATAL, "ComputationGraph attempting to get_task with operationid " << operationid << ", but operationid >= operations_.size().");
}
if (!operations_[operationid]->has_task()) {
HALO_LOG(HALO_FATAL, "Calling get_task with operationid " << operationid << ", but this corresponds to a push not a task.");
RAY_LOG(RAY_FATAL, "Calling get_task with operationid " << operationid << ", but this corresponds to a push not a task.");
}
return operations_[operationid]->task();
}
src/computation_graph.h
+4 -4
View File
@@ -1,11 +1,11 @@
#ifndef HALO_COMPUTATIONGRAPH_H
#define HALO_COMPUTATIONGRAPH_H
#ifndef RAY_COMPUTATIONGRAPH_H
#define RAY_COMPUTATIONGRAPH_H
#include <iostream>
#include <limits>
#include "halo/halo.h"
#include "halo.grpc.pb.h"
#include "ray/ray.h"
#include "ray.grpc.pb.h"
#include "types.pb.h"
// used to represent the root operation (that is, the driver code)
src/ipc.cc
+6 -6
View File
@@ -31,9 +31,9 @@ MemorySegmentPool::MemorySegmentPool(ObjStoreId objstoreid, bool create) : objst
// creates a memory segment if it is not already there; if the pool is in create mode,
// space is allocated, if it is in open mode, the shared memory is mapped into the process
void MemorySegmentPool::open_segment(SegmentId segmentid, size_t size) {
HALO_LOG(HALO_DEBUG, "Opening segmentid " << segmentid << " on object store " << objstoreid_ << " with create_mode_ = " << create_mode_);
RAY_LOG(RAY_DEBUG, "Opening segmentid " << segmentid << " on object store " << objstoreid_ << " with create_mode_ = " << create_mode_);
if (segmentid != segments_.size() && create_mode_) {
HALO_LOG(HALO_FATAL, "Object store " << objstoreid_ << " is attempting to open segmentid " << segmentid << " on the object store, but segments_.size() = " << segments_.size());
RAY_LOG(RAY_FATAL, "Object store " << objstoreid_ << " is attempting to open segmentid " << segmentid << " on the object store, but segments_.size() = " << segments_.size());
}
if (segmentid >= segments_.size()) { // resize and initialize segments_
int current_size = segments_.size();
@@ -47,7 +47,7 @@ void MemorySegmentPool::open_segment(SegmentId segmentid, size_t size) {
return;
}
if (segments_[segmentid].second == SegmentStatusType::CLOSED) {
HALO_LOG(HALO_FATAL, "Attempting to open segmentid " << segmentid << ", but segments_[segmentid].second == SegmentStatusType::CLOSED.");
RAY_LOG(RAY_FATAL, "Attempting to open segmentid " << segmentid << ", but segments_[segmentid].second == SegmentStatusType::CLOSED.");
}
std::string segment_name = get_segment_name(segmentid);
if (create_mode_) {
@@ -61,7 +61,7 @@ void MemorySegmentPool::open_segment(SegmentId segmentid, size_t size) {
}
void MemorySegmentPool::close_segment(SegmentId segmentid) {
HALO_LOG(HALO_DEBUG, "closing segmentid " << segmentid);
RAY_LOG(RAY_DEBUG, "closing segmentid " << segmentid);
std::string segment_name = get_segment_name(segmentid);
shared_memory_object::remove(segment_name.c_str());
segments_[segmentid].first.reset();
@@ -70,7 +70,7 @@ void MemorySegmentPool::close_segment(SegmentId segmentid) {
ObjHandle MemorySegmentPool::allocate(size_t size) {
if (!create_mode_) { // allocate is called only by the object store
HALO_LOG(HALO_FATAL, "Attempting to call allocate, but create_mode_ is false");
RAY_LOG(RAY_FATAL, "Attempting to call allocate, but create_mode_ is false");
}
// TODO(pcm): at the moment, this always creates a new segment, this will be changed
SegmentId segmentid = segments_.size();
@@ -91,7 +91,7 @@ void MemorySegmentPool::deallocate(ObjHandle pointer) {
// the process that will use the address
uint8_t* MemorySegmentPool::get_address(ObjHandle pointer) {
if (create_mode_ && segments_[pointer.segmentid()].second != SegmentStatusType::OPENED) {
HALO_LOG(HALO_FATAL, "Object store " << objstoreid_ << " is attempting to call get_address on segmentid " << pointer.segmentid() << ", which has not been opened yet.");
RAY_LOG(RAY_FATAL, "Object store " << objstoreid_ << " is attempting to call get_address on segmentid " << pointer.segmentid() << ", which has not been opened yet.");
}
if (!create_mode_) {
open_segment(pointer.segmentid());
src/ipc.h
+6 -6
View File
@@ -1,5 +1,5 @@
#ifndef HALO_IPC_H
#define HALO_IPC_H
#ifndef RAY_IPC_H
#define RAY_IPC_H
#include <iostream>
#include <limits>
@@ -10,7 +10,7 @@
#include <arrow/api.h>
#include <arrow/ipc/memory.h>
#include "halo/halo.h"
#include "ray/ray.h"
using namespace boost::interprocess;
@@ -42,7 +42,7 @@ public:
queue_ = std::unique_ptr<message_queue>(new message_queue(open_only, name.c_str()));
}
} catch(interprocess_exception &ex) {
HALO_LOG(HALO_FATAL, "boost::interprocess exception: " << ex.what());
RAY_LOG(RAY_FATAL, "boost::interprocess exception: " << ex.what());
}
return true;
};
@@ -55,7 +55,7 @@ public:
try {
queue_->send(object, sizeof(T), 0);
} catch(interprocess_exception &ex) {
HALO_LOG(HALO_FATAL, "boost::interprocess exception: " << ex.what());
RAY_LOG(RAY_FATAL, "boost::interprocess exception: " << ex.what());
}
return true;
};
@@ -66,7 +66,7 @@ public:
try {
queue_->receive(object, sizeof(T), recvd_size, priority);
} catch(interprocess_exception &ex) {
HALO_LOG(HALO_FATAL, "boost::interprocess exception: " << ex.what());
RAY_LOG(RAY_FATAL, "boost::interprocess exception: " << ex.what());
}
return true;
}
src/objstore.cc
+32 -32
View File
@@ -8,7 +8,7 @@ const size_t ObjStoreService::CHUNK_SIZE = 8 * 1024;
// this method needs to be protected by a objstore_lock_
// TODO(rkn): Make sure that we do not in fact need the objstore_lock_. We want multiple deliveries to be able to happen simultaneously.
void ObjStoreService::pull_data_from(ObjRef objref, ObjStore::Stub& stub) {
HALO_LOG(HALO_DEBUG, "Objstore " << objstoreid_ << " is beginning to pull objref " << objref);
RAY_LOG(RAY_DEBUG, "Objstore " << objstoreid_ << " is beginning to pull objref " << objref);
ObjChunk chunk;
ClientContext context;
StreamObjToRequest stream_request;
@@ -27,7 +27,7 @@ void ObjStoreService::pull_data_from(ObjRef objref, ObjStore::Stub& stub) {
segmentpool_lock_.unlock();
do {
if (num_bytes + chunk.data().size() > total_size) {
HALO_LOG(HALO_FATAL, "The reader attempted to stream too many bytes.");
RAY_LOG(RAY_FATAL, "The reader attempted to stream too many bytes.");
}
std::memcpy(data, chunk.data().c_str(), chunk.data().size());
data += chunk.data().size();
@@ -37,10 +37,10 @@ void ObjStoreService::pull_data_from(ObjRef objref, ObjStore::Stub& stub) {
// finalize object
if (num_bytes != total_size) {
HALO_LOG(HALO_FATAL, "Streamed objref " << objref << ", but num_bytes != total_size");
RAY_LOG(RAY_FATAL, "Streamed objref " << objref << ", but num_bytes != total_size");
}
object_ready(objref, chunk.metadata_offset());
HALO_LOG(HALO_DEBUG, "finished streaming data, objref was " << objref << " and size was " << num_bytes);
RAY_LOG(RAY_DEBUG, "finished streaming data, objref was " << objref << " and size was " << num_bytes);
}
ObjStoreService::ObjStoreService(const std::string& objstore_address, std::shared_ptr<Channel> scheduler_channel)
@@ -79,10 +79,10 @@ Status ObjStoreService::StartDelivery(ServerContext* context, const StartDeliver
if (memory_[objref].second == MemoryStatusType::NOT_PRESENT) {
}
else if (memory_[objref].second == MemoryStatusType::DEALLOCATED) {
HALO_LOG(HALO_FATAL, "Objstore " << objstoreid_ << " is attempting to get objref " << objref << ", but memory_[objref] == DEALLOCATED.");
RAY_LOG(RAY_FATAL, "Objstore " << objstoreid_ << " is attempting to get objref " << objref << ", but memory_[objref] == DEALLOCATED.");
}
else {
HALO_LOG(HALO_DEBUG, "Objstore " << objstoreid_ << " already has objref " << objref << " or it is already being shipped, so no need to pull it again.");
RAY_LOG(RAY_DEBUG, "Objstore " << objstoreid_ << " already has objref " << objref << " or it is already being shipped, so no need to pull it again.");
return Status::OK;
}
memory_[objref].second = MemoryStatusType::PRE_ALLOCED;
@@ -115,15 +115,15 @@ Status ObjStoreService::ObjStoreInfo(ServerContext* context, const ObjStoreInfoR
}
Status ObjStoreService::StreamObjTo(ServerContext* context, const StreamObjToRequest* request, ServerWriter<ObjChunk>* writer) {
HALO_LOG(HALO_DEBUG, "begin to stream data from object store " << objstoreid_);
RAY_LOG(RAY_DEBUG, "begin to stream data from object store " << objstoreid_);
ObjChunk chunk;
ObjRef objref = request->objref();
memory_lock_.lock();
if (objref >= memory_.size()) {
HALO_LOG(HALO_FATAL, "Objstore " << objstoreid_ << " is attempting to use objref " << objref << " in StreamObjTo, but this objref is not present in the object store.");
RAY_LOG(RAY_FATAL, "Objstore " << objstoreid_ << " is attempting to use objref " << objref << " in StreamObjTo, but this objref is not present in the object store.");
}
if (memory_[objref].second != MemoryStatusType::READY) {
HALO_LOG(HALO_FATAL, "Objstore " << objstoreid_ << " is attempting to stream objref " << objref << ", but memory_[objref].second != MemoryStatusType::READY.");
RAY_LOG(RAY_FATAL, "Objstore " << objstoreid_ << " is attempting to stream objref " << objref << ", but memory_[objref].second != MemoryStatusType::READY.");
}
ObjHandle handle = memory_[objref].first;
memory_lock_.unlock(); // TODO(rkn): Make sure we don't still need to hold on to this lock.
@@ -136,7 +136,7 @@ Status ObjStoreService::StreamObjTo(ServerContext* context, const StreamObjToReq
chunk.set_total_size(size);
chunk.set_data(head + i, std::min(CHUNK_SIZE, size - i));
if (!writer->Write(chunk)) {
HALO_LOG(HALO_FATAL, "stream connection prematurely closed")
RAY_LOG(RAY_FATAL, "stream connection prematurely closed")
}
}
return Status::OK;
@@ -146,20 +146,20 @@ Status ObjStoreService::NotifyAlias(ServerContext* context, const NotifyAliasReq
// NotifyAlias assumes that the objstore already holds canonical_objref
ObjRef alias_objref = request->alias_objref();
ObjRef canonical_objref = request->canonical_objref();
HALO_LOG(HALO_DEBUG, "Aliasing objref " << alias_objref << " with objref " << canonical_objref);
RAY_LOG(RAY_DEBUG, "Aliasing objref " << alias_objref << " with objref " << canonical_objref);
{
std::lock_guard<std::mutex> memory_lock(memory_lock_);
if (canonical_objref >= memory_.size()) {
HALO_LOG(HALO_FATAL, "Attempting to alias objref " << alias_objref << " with objref " << canonical_objref << ", but objref " << canonical_objref << " is not in the objstore.")
RAY_LOG(RAY_FATAL, "Attempting to alias objref " << alias_objref << " with objref " << canonical_objref << ", but objref " << canonical_objref << " is not in the objstore.")
}
if (memory_[canonical_objref].second == MemoryStatusType::NOT_READY) {
HALO_LOG(HALO_FATAL, "Attempting to alias objref " << alias_objref << " with objref " << canonical_objref << ", but objref " << canonical_objref << " is not ready yet in the objstore.")
RAY_LOG(RAY_FATAL, "Attempting to alias objref " << alias_objref << " with objref " << canonical_objref << ", but objref " << canonical_objref << " is not ready yet in the objstore.")
}
if (memory_[canonical_objref].second == MemoryStatusType::NOT_PRESENT) {
HALO_LOG(HALO_FATAL, "Attempting to alias objref " << alias_objref << " with objref " << canonical_objref << ", but objref " << canonical_objref << " is not present in the objstore.")
RAY_LOG(RAY_FATAL, "Attempting to alias objref " << alias_objref << " with objref " << canonical_objref << ", but objref " << canonical_objref << " is not present in the objstore.")
}
if (memory_[canonical_objref].second == MemoryStatusType::DEALLOCATED) {
HALO_LOG(HALO_FATAL, "Attempting to alias objref " << alias_objref << " with objref " << canonical_objref << ", but objref " << canonical_objref << " has already been deallocated.")
RAY_LOG(RAY_FATAL, "Attempting to alias objref " << alias_objref << " with objref " << canonical_objref << ", but objref " << canonical_objref << " has already been deallocated.")
}
if (alias_objref >= memory_.size()) {
memory_.resize(alias_objref + 1, std::make_pair(ObjHandle(), MemoryStatusType::NOT_PRESENT));
@@ -176,13 +176,13 @@ Status ObjStoreService::NotifyAlias(ServerContext* context, const NotifyAliasReq
Status ObjStoreService::DeallocateObject(ServerContext* context, const DeallocateObjectRequest* request, AckReply* reply) {
ObjRef canonical_objref = request->canonical_objref();
HALO_LOG(HALO_REFCOUNT, "Deallocating canonical_objref " << canonical_objref);
RAY_LOG(RAY_REFCOUNT, "Deallocating canonical_objref " << canonical_objref);
std::lock_guard<std::mutex> memory_lock(memory_lock_);
if (memory_[canonical_objref].second != MemoryStatusType::READY) {
HALO_LOG(HALO_FATAL, "Attempting to deallocate canonical_objref " << canonical_objref << ", but memory_[canonical_objref].second = " << memory_[canonical_objref].second);
RAY_LOG(RAY_FATAL, "Attempting to deallocate canonical_objref " << canonical_objref << ", but memory_[canonical_objref].second = " << memory_[canonical_objref].second);
}
if (canonical_objref >= memory_.size()) {
HALO_LOG(HALO_FATAL, "Attempting to deallocate canonical_objref " << canonical_objref << ", but it is not in the objstore.");
RAY_LOG(RAY_FATAL, "Attempting to deallocate canonical_objref " << canonical_objref << ", but it is not in the objstore.");
}
segmentpool_lock_.lock();
segmentpool_->deallocate(memory_[canonical_objref].first);
@@ -208,7 +208,7 @@ void ObjStoreService::process_objstore_request(const ObjRequest request) {
}
break;
default: {
HALO_LOG(HALO_FATAL, "Attempting to process request of type " << request.type << ". This code should be unreachable.");
RAY_LOG(RAY_FATAL, "Attempting to process request of type " << request.type << ". This code should be unreachable.");
}
}
}
@@ -237,13 +237,13 @@ void ObjStoreService::process_worker_request(const ObjRequest request) {
std::lock_guard<std::mutex> memory_lock(memory_lock_);
std::pair<ObjHandle, MemoryStatusType>& item = memory_[request.objref];
if (item.second == MemoryStatusType::READY) {
HALO_LOG(HALO_DEBUG, "Responding to GET request: returning objref " << request.objref);
RAY_LOG(RAY_DEBUG, "Responding to GET request: returning objref " << request.objref);
send_queues_[request.workerid].send(&item.first);
} else if (item.second == MemoryStatusType::NOT_READY || item.second == MemoryStatusType::NOT_PRESENT || item.second == MemoryStatusType::PRE_ALLOCED) {
std::lock_guard<std::mutex> lock(pull_queue_lock_);
pull_queue_.push_back(std::make_pair(request.workerid, request.objref));
} else {
HALO_LOG(HALO_FATAL, "A worker requested objref " << request.objref << ", but memory_[objref].second = " << memory_[request.objref].second);
RAY_LOG(RAY_FATAL, "A worker requested objref " << request.objref << ", but memory_[objref].second = " << memory_[request.objref].second);
}
}
break;
@@ -252,7 +252,7 @@ void ObjStoreService::process_worker_request(const ObjRequest request) {
}
break;
default: {
HALO_LOG(HALO_FATAL, "Attempting to process request of type " << request.type << ". This code should be unreachable.");
RAY_LOG(RAY_FATAL, "Attempting to process request of type " << request.type << ". This code should be unreachable.");
}
}
}
@@ -264,17 +264,17 @@ void ObjStoreService::process_requests() {
recv_queue_.receive(&request);
switch (request.type) {
case ObjRequestType::ALLOC: {
HALO_LOG(HALO_VERBOSE, "Request (worker " << request.workerid << " to objstore " << objstoreid_ << "): Allocate object with objref " << request.objref << " and size " << request.size);
RAY_LOG(RAY_VERBOSE, "Request (worker " << request.workerid << " to objstore " << objstoreid_ << "): Allocate object with objref " << request.objref << " and size " << request.size);
process_worker_request(request);
}
break;
case ObjRequestType::GET: {
HALO_LOG(HALO_VERBOSE, "Request (worker " << request.workerid << " to objstore " << objstoreid_ << "): Get object with objref " << request.objref);
RAY_LOG(RAY_VERBOSE, "Request (worker " << request.workerid << " to objstore " << objstoreid_ << "): Get object with objref " << request.objref);
process_worker_request(request);
}
break;
case ObjRequestType::WORKER_DONE: {
HALO_LOG(HALO_VERBOSE, "Request (worker " << request.workerid << " to objstore " << objstoreid_ << "): Finalize object with objref " << request.objref);
RAY_LOG(RAY_VERBOSE, "Request (worker " << request.workerid << " to objstore " << objstoreid_ << "): Finalize object with objref " << request.objref);
process_worker_request(request);
}
break;
@@ -283,7 +283,7 @@ void ObjStoreService::process_requests() {
}
break;
default: {
HALO_LOG(HALO_FATAL, "Attempting to process request of type " << request.type << ". This code should be unreachable.");
RAY_LOG(RAY_FATAL, "Attempting to process request of type " << request.type << ". This code should be unreachable.");
}
}
}
@@ -309,9 +309,9 @@ ObjHandle ObjStoreService::alloc(ObjRef objref, size_t size) {
ObjHandle handle = segmentpool_->allocate(size);
segmentpool_lock_.unlock();
std::lock_guard<std::mutex> memory_lock(memory_lock_);
HALO_LOG(HALO_VERBOSE, "Allocating space for objref " << objref << " on object store " << objstoreid_);
RAY_LOG(RAY_VERBOSE, "Allocating space for objref " << objref << " on object store " << objstoreid_);
if (memory_[objref].second != MemoryStatusType::NOT_PRESENT && memory_[objref].second != MemoryStatusType::PRE_ALLOCED) {
HALO_LOG(HALO_FATAL, "Attempting to allocate space for objref " << objref << ", but memory_[objref].second = " << memory_[objref].second);
RAY_LOG(RAY_FATAL, "Attempting to allocate space for objref " << objref << ", but memory_[objref].second = " << memory_[objref].second);
}
memory_[objref].first = handle;
memory_[objref].second = MemoryStatusType::NOT_READY;
@@ -323,7 +323,7 @@ void ObjStoreService::object_ready(ObjRef objref, size_t metadata_offset) {
std::lock_guard<std::mutex> memory_lock(memory_lock_);
std::pair<ObjHandle, MemoryStatusType>& item = memory_[objref];
if (item.second != MemoryStatusType::NOT_READY) {
HALO_LOG(HALO_FATAL, "A worker notified the object store that objref " << objref << " has been written to the object store, but memory_[objref].second != NOT_READY.");
RAY_LOG(RAY_FATAL, "A worker notified the object store that objref " << objref << " has been written to the object store, but memory_[objref].second != NOT_READY.");
}
item.first.set_metadata_offset(metadata_offset);
item.second = MemoryStatusType::READY;
@@ -341,14 +341,14 @@ void ObjStoreService::object_ready(ObjRef objref, size_t metadata_offset) {
void ObjStoreService::start_objstore_service() {
communicator_thread_ = std::thread([this]() {
HALO_LOG(HALO_INFO, "started object store communicator server");
RAY_LOG(RAY_INFO, "started object store communicator server");
process_requests();
});
}
void start_objstore(const char* scheduler_addr, const char* objstore_addr) {
auto scheduler_channel = grpc::CreateChannel(scheduler_addr, grpc::InsecureChannelCredentials());
HALO_LOG(HALO_INFO, "object store " << objstore_addr << " connected to scheduler " << scheduler_addr);
RAY_LOG(RAY_INFO, "object store " << objstore_addr << " connected to scheduler " << scheduler_addr);
std::string objstore_address(objstore_addr);
ObjStoreService service(objstore_address, scheduler_channel);
service.start_objstore_service();
@@ -365,7 +365,7 @@ void start_objstore(const char* scheduler_addr, const char* objstore_addr) {
int main(int argc, char** argv) {
if (argc != 3) {
HALO_LOG(HALO_FATAL, "object store: expected two arguments (scheduler ip address and object store ip address)");
RAY_LOG(RAY_FATAL, "object store: expected two arguments (scheduler ip address and object store ip address)");
return 1;
}
src/objstore.h
+4 -4
View File
@@ -1,5 +1,5 @@
#ifndef HALO_OBJSTORE_H
#define HALO_OBJSTORE_H
#ifndef RAY_OBJSTORE_H
#define RAY_OBJSTORE_H
#include <unordered_map>
#include <memory>
@@ -7,8 +7,8 @@
#include <iostream>
#include <grpc++/grpc++.h>
#include "halo/halo.h"
#include "halo.grpc.pb.h"
#include "ray/ray.h"
#include "ray.grpc.pb.h"
#include "types.pb.h"
#include "ipc.h"
src/halolib.cc → src/raylib.cc
+20 -20
View File
@@ -45,7 +45,7 @@ static int PyObjRef_init(PyObjRef *self, PyObject *args, PyObject *kwds) {
}
std::vector<ObjRef> objrefs;
objrefs.push_back(self->val);
HALO_LOG(HALO_REFCOUNT, "In PyObjRef_init, calling increment_reference_count for objref " << objrefs[0]);
RAY_LOG(RAY_REFCOUNT, "In PyObjRef_init, calling increment_reference_count for objref " << objrefs[0]);
self->worker->increment_reference_count(objrefs);
return 0;
};
@@ -70,7 +70,7 @@ static PyMemberDef PyObjRef_members[] = {
static PyTypeObject PyObjRefType = {
PyObject_HEAD_INIT(NULL)
0, /* ob_size */
"halo.ObjRef", /* tp_name */
"ray.ObjRef", /* tp_name */
sizeof(PyObjRef), /* tp_basicsize */
0, /* tp_itemsize */
(destructor)PyObjRef_dealloc, /* tp_dealloc */
@@ -89,7 +89,7 @@ static PyTypeObject PyObjRefType = {
0, /* tp_setattro */
0, /* tp_as_buffer */
Py_TPFLAGS_DEFAULT, /* tp_flags */
"Halo objects", /* tp_doc */
"Ray objects", /* tp_doc */
0, /* tp_traverse */
0, /* tp_clear */
0, /* tp_richcompare */
@@ -119,7 +119,7 @@ PyObject* make_pyobjref(PyObject* worker_capsule, ObjRef objref) {
// Error handling
static PyObject *HaloError;
static PyObject *RayError;
// Pass arguments from Python to C++
@@ -312,12 +312,12 @@ int serialize(PyObject* worker_capsule, PyObject* val, Obj* obj, std::vector<Obj
}
break;
default:
PyErr_SetString(HaloError, "serialization: numpy datatype not know");
PyErr_SetString(RayError, "serialization: numpy datatype not know");
return -1;
}
Py_DECREF(array); // TODO(rkn): is this right?
} else {
PyErr_SetString(HaloError, "serialization: type not know");
PyErr_SetString(RayError, "serialization: type not know");
return -1;
}
return 0;
@@ -402,7 +402,7 @@ PyObject* deserialize(PyObject* worker_capsule, const Obj& obj, std::vector<ObjR
}
break;
default:
PyErr_SetString(HaloError, "deserialization: internal error (array type not implemented)");
PyErr_SetString(RayError, "deserialization: internal error (array type not implemented)");
return NULL;
}
} else if (array.uint_data_size() > 0) {
@@ -423,7 +423,7 @@ PyObject* deserialize(PyObject* worker_capsule, const Obj& obj, std::vector<ObjR
}
break;
default:
PyErr_SetString(HaloError, "deserialization: internal error (array type not implemented)");
PyErr_SetString(RayError, "deserialization: internal error (array type not implemented)");
return NULL;
}
} else if (array.objref_data_size() > 0) {
@@ -434,12 +434,12 @@ PyObject* deserialize(PyObject* worker_capsule, const Obj& obj, std::vector<ObjR
objrefs.push_back(array.objref_data(i));
}
} else {
PyErr_SetString(HaloError, "deserialization: internal error (array type not implemented)");
PyErr_SetString(RayError, "deserialization: internal error (array type not implemented)");
return NULL;
}
return (PyObject*) pyarray;
} else {
PyErr_SetString(HaloError, "deserialization: internal error (type not implemented)");
PyErr_SetString(RayError, "deserialization: internal error (type not implemented)");
return NULL;
}
}
@@ -535,13 +535,13 @@ PyObject* serialize_task(PyObject* self, PyObject* args) {
}
}
} else {
PyErr_SetString(HaloError, "serialize_task: second argument needs to be a list");
PyErr_SetString(RayError, "serialize_task: second argument needs to be a list");
return NULL;
}
Worker* worker;
PyObjectToWorker(worker_capsule, &worker);
if (objrefs.size() > 0) {
HALO_LOG(HALO_REFCOUNT, "In serialize_task, calling increment_reference_count for contained objrefs");
RAY_LOG(RAY_REFCOUNT, "In serialize_task, calling increment_reference_count for contained objrefs");
worker->increment_reference_count(objrefs);
}
return PyCapsule_New(static_cast<void*>(task), "task", &TaskCapsule_Destructor);
@@ -584,7 +584,7 @@ PyObject* deserialize_task(PyObject* self, PyObject* args) {
return t;
}
// Halo Python API
// Ray Python API
PyObject* create_worker(PyObject* self, PyObject* args) {
const char* scheduler_addr;
@@ -692,7 +692,7 @@ PyObject* put_object(PyObject* self, PyObject* args) {
return NULL;
}
if (!PyList_Check(contained_objrefs)) {
HALO_LOG(HALO_FATAL, "The contained_objrefs argument must be a list.")
RAY_LOG(RAY_FATAL, "The contained_objrefs argument must be a list.")
}
std::vector<ObjRef> vec_contained_objrefs;
size_t size = PyList_Size(contained_objrefs);
@@ -773,7 +773,7 @@ PyObject* scheduler_info(PyObject* self, PyObject* args) {
return dict;
}
static PyMethodDef HaloLibMethods[] = {
static PyMethodDef RayLibMethods[] = {
{ "serialize_object", serialize_object, METH_VARARGS, "serialize an object to protocol buffers" },
{ "deserialize_object", deserialize_object, METH_VARARGS, "deserialize an object from protocol buffers" },
{ "put_arrow", put_arrow, METH_VARARGS, "put an arrow array on the local object store"},
@@ -798,18 +798,18 @@ static PyMethodDef HaloLibMethods[] = {
{ NULL, NULL, 0, NULL }
};
PyMODINIT_FUNC initlibhalolib(void) {
PyMODINIT_FUNC initlibraylib(void) {
PyObject* m;
PyObjRefType.tp_new = PyType_GenericNew;
if (PyType_Ready(&PyObjRefType) < 0) {
return;
}
m = Py_InitModule3("libhalolib", HaloLibMethods, "Python C Extension for Halo");
m = Py_InitModule3("libraylib", RayLibMethods, "Python C Extension for Ray");
Py_INCREF(&PyObjRefType);
PyModule_AddObject(m, "ObjRef", (PyObject *)&PyObjRefType);
HaloError = PyErr_NewException("halo.error", NULL, NULL);
Py_INCREF(HaloError);
PyModule_AddObject(m, "error", HaloError);
RayError = PyErr_NewException("ray.error", NULL, NULL);
Py_INCREF(RayError);
PyModule_AddObject(m, "error", RayError);
import_array();
}
src/scheduler.cc
+51 -51
View File
@@ -14,7 +14,7 @@ Status SchedulerService::SubmitTask(ServerContext* context, const SubmitTaskRequ
if (fntable_.find(task->name()) == fntable_.end()) {
// TODO(rkn): In the future, this should probably not be fatal. Instead, propagate the error back to the worker.
HALO_LOG(HALO_FATAL, "The function " << task->name() << " has not been registered by any worker.");
RAY_LOG(RAY_FATAL, "The function " << task->name() << " has not been registered by any worker.");
}
size_t num_return_vals = fntable_[task->name()].num_return_vals();
@@ -29,8 +29,8 @@ Status SchedulerService::SubmitTask(ServerContext* context, const SubmitTaskRequ
}
{
std::lock_guard<std::mutex> reference_counts_lock(reference_counts_lock_); // we grab this lock because increment_ref_count assumes it has been acquired
increment_ref_count(result_objrefs); // We increment once so the objrefs don't go out of scope before we reply to the worker that called SubmitTask. The corresponding decrement will happen in submit_task in halolib.
increment_ref_count(result_objrefs); // We increment once so the objrefs don't go out of scope before the task is scheduled on the worker. The corresponding decrement will happen in deserialize_task in halolib.
increment_ref_count(result_objrefs); // We increment once so the objrefs don't go out of scope before we reply to the worker that called SubmitTask. The corresponding decrement will happen in submit_task in raylib.
increment_ref_count(result_objrefs); // We increment once so the objrefs don't go out of scope before the task is scheduled on the worker. The corresponding decrement will happen in deserialize_task in raylib.
}
auto operation = std::unique_ptr<Operation>(new Operation());
@@ -64,7 +64,7 @@ Status SchedulerService::RequestObj(ServerContext* context, const RequestObjRequ
ObjRef objref = request->objref();
if (objref >= size) {
HALO_LOG(HALO_FATAL, "internal error: no object with objref " << objref << " exists");
RAY_LOG(RAY_FATAL, "internal error: no object with objref " << objref << " exists");
}
pull_queue_lock_.lock();
@@ -77,23 +77,23 @@ Status SchedulerService::RequestObj(ServerContext* context, const RequestObjRequ
Status SchedulerService::AliasObjRefs(ServerContext* context, const AliasObjRefsRequest* request, AckReply* reply) {
ObjRef alias_objref = request->alias_objref();
ObjRef target_objref = request->target_objref();
HALO_LOG(HALO_ALIAS, "Aliasing objref " << alias_objref << " with objref " << target_objref);
RAY_LOG(RAY_ALIAS, "Aliasing objref " << alias_objref << " with objref " << target_objref);
if (alias_objref == target_objref) {
HALO_LOG(HALO_FATAL, "internal error: attempting to alias objref " << alias_objref << " with itself.");
RAY_LOG(RAY_FATAL, "internal error: attempting to alias objref " << alias_objref << " with itself.");
}
objtable_lock_.lock();
size_t size = objtable_.size();
objtable_lock_.unlock();
if (alias_objref >= size) {
HALO_LOG(HALO_FATAL, "internal error: no object with objref " << alias_objref << " exists");
RAY_LOG(RAY_FATAL, "internal error: no object with objref " << alias_objref << " exists");
}
if (target_objref >= size) {
HALO_LOG(HALO_FATAL, "internal error: no object with objref " << target_objref << " exists");
RAY_LOG(RAY_FATAL, "internal error: no object with objref " << target_objref << " exists");
}
{
std::lock_guard<std::mutex> target_objrefs_lock(target_objrefs_lock_);
if (target_objrefs_[alias_objref] != UNITIALIZED_ALIAS) {
HALO_LOG(HALO_FATAL, "internal error: attempting to alias objref " << alias_objref << " with objref " << target_objref << ", but objref " << alias_objref << " has already been aliased with objref " << target_objrefs_[alias_objref]);
RAY_LOG(RAY_FATAL, "internal error: attempting to alias objref " << alias_objref << " with objref " << target_objref << ", but objref " << alias_objref << " has already been aliased with objref " << target_objrefs_[alias_objref]);
}
target_objrefs_[alias_objref] = target_objref;
}
@@ -121,7 +121,7 @@ Status SchedulerService::RegisterWorker(ServerContext* context, const RegisterWo
std::pair<WorkerId, ObjStoreId> info = register_worker(request->worker_address(), request->objstore_address());
WorkerId workerid = info.first;
ObjStoreId objstoreid = info.second;
HALO_LOG(HALO_INFO, "registered worker with workerid " << workerid);
RAY_LOG(RAY_INFO, "registered worker with workerid " << workerid);
reply->set_workerid(workerid);
reply->set_objstoreid(objstoreid);
schedule();
@@ -129,7 +129,7 @@ Status SchedulerService::RegisterWorker(ServerContext* context, const RegisterWo
}
Status SchedulerService::RegisterFunction(ServerContext* context, const RegisterFunctionRequest* request, AckReply* reply) {
HALO_LOG(HALO_INFO, "register function " << request->fnname() << " from workerid " << request->workerid());
RAY_LOG(RAY_INFO, "register function " << request->fnname() << " from workerid " << request->workerid());
register_function(request->fnname(), request->workerid(), request->num_return_vals());
schedule();
return Status::OK;
@@ -137,7 +137,7 @@ Status SchedulerService::RegisterFunction(ServerContext* context, const Register
Status SchedulerService::ObjReady(ServerContext* context, const ObjReadyRequest* request, AckReply* reply) {
ObjRef objref = request->objref();
HALO_LOG(HALO_DEBUG, "object " << objref << " ready on store " << request->objstoreid());
RAY_LOG(RAY_DEBUG, "object " << objref << " ready on store " << request->objstoreid());
add_canonical_objref(objref);
add_location(objref, request->objstoreid());
schedule();
@@ -145,7 +145,7 @@ Status SchedulerService::ObjReady(ServerContext* context, const ObjReadyRequest*
}
Status SchedulerService::WorkerReady(ServerContext* context, const WorkerReadyRequest* request, AckReply* reply) {
HALO_LOG(HALO_INFO, "worker " << request->workerid() << " reported back");
RAY_LOG(RAY_INFO, "worker " << request->workerid() << " reported back");
{
std::lock_guard<std::mutex> lock(avail_workers_lock_);
avail_workers_.push_back(request->workerid());
@@ -157,7 +157,7 @@ Status SchedulerService::WorkerReady(ServerContext* context, const WorkerReadyRe
Status SchedulerService::IncrementRefCount(ServerContext* context, const IncrementRefCountRequest* request, AckReply* reply) {
int num_objrefs = request->objref_size();
if (num_objrefs == 0) {
HALO_LOG(HALO_FATAL, "Scheduler received IncrementRefCountRequest with 0 objrefs.");
RAY_LOG(RAY_FATAL, "Scheduler received IncrementRefCountRequest with 0 objrefs.");
}
std::vector<ObjRef> objrefs;
for (int i = 0; i < num_objrefs; ++i) {
@@ -171,7 +171,7 @@ Status SchedulerService::IncrementRefCount(ServerContext* context, const Increme
Status SchedulerService::DecrementRefCount(ServerContext* context, const DecrementRefCountRequest* request, AckReply* reply) {
int num_objrefs = request->objref_size();
if (num_objrefs == 0) {
HALO_LOG(HALO_FATAL, "Scheduler received DecrementRefCountRequest with 0 objrefs.");
RAY_LOG(RAY_FATAL, "Scheduler received DecrementRefCountRequest with 0 objrefs.");
}
std::vector<ObjRef> objrefs;
for (int i = 0; i < num_objrefs; ++i) {
@@ -186,11 +186,11 @@ Status SchedulerService::AddContainedObjRefs(ServerContext* context, const AddCo
ObjRef objref = request->objref();
// if (!is_canonical(objref)) {
// TODO(rkn): Perhaps we don't need this check. It won't work because the objstore may not have called ObjReady yet.
// HALO_LOG(HALO_FATAL, "Attempting to add contained objrefs for non-canonical objref " << objref);
// RAY_LOG(RAY_FATAL, "Attempting to add contained objrefs for non-canonical objref " << objref);
// }
std::lock_guard<std::mutex> contained_objrefs_lock(contained_objrefs_lock_);
if (contained_objrefs_[objref].size() != 0) {
HALO_LOG(HALO_FATAL, "Attempting to add contained objrefs for objref " << objref << ", but contained_objrefs_[objref].size() != 0.");
RAY_LOG(RAY_FATAL, "Attempting to add contained objrefs for objref " << objref << ", but contained_objrefs_[objref].size() != 0.");
}
for (int i = 0; i < request->contained_objref_size(); ++i) {
contained_objrefs_[objref].push_back(request->contained_objref(i));
@@ -212,10 +212,10 @@ Status SchedulerService::SchedulerInfo(ServerContext* context, const SchedulerIn
// deliver_object assumes that the aliasing for objref has already been completed. That is, has_canonical_objref(objref) == true
void SchedulerService::deliver_object(ObjRef objref, ObjStoreId from, ObjStoreId to) {
if (from == to) {
HALO_LOG(HALO_FATAL, "attempting to deliver objref " << objref << " from objstore " << from << " to itself.");
RAY_LOG(RAY_FATAL, "attempting to deliver objref " << objref << " from objstore " << from << " to itself.");
}
if (!has_canonical_objref(objref)) {
HALO_LOG(HALO_FATAL, "attempting to deliver objref " << objref << ", but this objref does not yet have a canonical objref.");
RAY_LOG(RAY_FATAL, "attempting to deliver objref " << objref << ", but this objref does not yet have a canonical objref.");
}
ClientContext context;
AckReply reply;
@@ -235,7 +235,7 @@ void SchedulerService::schedule() {
} else if (scheduling_algorithm_ == SCHEDULING_ALGORITHM_LOCALITY_AWARE) {
schedule_tasks_location_aware(); // See what we can do in task_queue_
} else {
HALO_LOG(HALO_FATAL, "scheduling algorithm not known");
RAY_LOG(RAY_FATAL, "scheduling algorithm not known");
}
perform_notify_aliases(); // See what we can do in alias_notification_queue_
}
@@ -247,7 +247,7 @@ void SchedulerService::assign_task(OperationId operationid, WorkerId workerid) {
ClientContext context;
ExecuteTaskRequest request;
ExecuteTaskReply reply;
HALO_LOG(HALO_INFO, "starting to send arguments");
RAY_LOG(RAY_INFO, "starting to send arguments");
for (size_t i = 0; i < task.arg_size(); ++i) {
if (!task.arg(i).has_obj()) {
ObjRef objref = task.arg(i).ref();
@@ -259,7 +259,7 @@ void SchedulerService::assign_task(OperationId operationid, WorkerId workerid) {
}
attempt_notify_alias(get_store(workerid), objref, canonical_objref);
HALO_LOG(HALO_DEBUG, "task contains object ref " << canonical_objref);
RAY_LOG(RAY_DEBUG, "task contains object ref " << canonical_objref);
std::lock_guard<std::mutex> objtable_lock(objtable_lock_);
auto &objstores = objtable_[canonical_objref];
std::lock_guard<std::mutex> workers_lock(workers_lock_);
@@ -290,7 +290,7 @@ bool SchedulerService::can_run(const Task& task) {
}
std::pair<WorkerId, ObjStoreId> SchedulerService::register_worker(const std::string& worker_address, const std::string& objstore_address) {
HALO_LOG(HALO_INFO, "registering worker " << worker_address << " connected to object store " << objstore_address);
RAY_LOG(RAY_INFO, "registering worker " << worker_address << " connected to object store " << objstore_address);
ObjStoreId objstoreid = std::numeric_limits<size_t>::max();
for (int num_attempts = 0; num_attempts < 5; ++num_attempts) {
std::lock_guard<std::mutex> lock(objstores_lock_);
@@ -304,7 +304,7 @@ std::pair<WorkerId, ObjStoreId> SchedulerService::register_worker(const std::str
}
}
if (objstoreid == std::numeric_limits<size_t>::max()) {
HALO_LOG(HALO_FATAL, "object store with address " << objstore_address << " not yet registered");
RAY_LOG(RAY_FATAL, "object store with address " << objstore_address << " not yet registered");
}
workers_lock_.lock();
WorkerId workerid = workers_.size();
@@ -334,16 +334,16 @@ ObjRef SchedulerService::register_new_object() {
ObjRef reference_counts_size = reference_counts_.size();
ObjRef contained_objrefs_size = contained_objrefs_.size();
if (objtable_size != target_objrefs_size) {
HALO_LOG(HALO_FATAL, "objtable_ and target_objrefs_ should have the same size, but objtable_.size() = " << objtable_size << " and target_objrefs_.size() = " << target_objrefs_size);
RAY_LOG(RAY_FATAL, "objtable_ and target_objrefs_ should have the same size, but objtable_.size() = " << objtable_size << " and target_objrefs_.size() = " << target_objrefs_size);
}
if (objtable_size != reverse_target_objrefs_size) {
HALO_LOG(HALO_FATAL, "objtable_ and reverse_target_objrefs_ should have the same size, but objtable_.size() = " << objtable_size << " and reverse_target_objrefs_.size() = " << reverse_target_objrefs_size);
RAY_LOG(RAY_FATAL, "objtable_ and reverse_target_objrefs_ should have the same size, but objtable_.size() = " << objtable_size << " and reverse_target_objrefs_.size() = " << reverse_target_objrefs_size);
}
if (objtable_size != reference_counts_size) {
HALO_LOG(HALO_FATAL, "objtable_ and reference_counts_ should have the same size, but objtable_.size() = " << objtable_size << " and reference_counts_.size() = " << reference_counts_size);
RAY_LOG(RAY_FATAL, "objtable_ and reference_counts_ should have the same size, but objtable_.size() = " << objtable_size << " and reference_counts_.size() = " << reference_counts_size);
}
if (objtable_size != contained_objrefs_size) {
HALO_LOG(HALO_FATAL, "objtable_ and contained_objrefs_ should have the same size, but objtable_.size() = " << objtable_size << " and contained_objrefs_.size() = " << contained_objrefs_size);
RAY_LOG(RAY_FATAL, "objtable_ and contained_objrefs_ should have the same size, but objtable_.size() = " << objtable_size << " and contained_objrefs_.size() = " << contained_objrefs_size);
}
objtable_.push_back(std::vector<ObjStoreId>());
target_objrefs_.push_back(UNITIALIZED_ALIAS);
@@ -356,11 +356,11 @@ ObjRef SchedulerService::register_new_object() {
void SchedulerService::add_location(ObjRef canonical_objref, ObjStoreId objstoreid) {
// add_location must be called with a canonical objref
if (!is_canonical(canonical_objref)) {
HALO_LOG(HALO_FATAL, "Attempting to call add_location with a non-canonical objref (objref " << canonical_objref << ")");
RAY_LOG(RAY_FATAL, "Attempting to call add_location with a non-canonical objref (objref " << canonical_objref << ")");
}
std::lock_guard<std::mutex> objtable_lock(objtable_lock_);
if (canonical_objref >= objtable_.size()) {
HALO_LOG(HALO_FATAL, "trying to put an object in the object store that was not registered with the scheduler (objref " << canonical_objref << ")");
RAY_LOG(RAY_FATAL, "trying to put an object in the object store that was not registered with the scheduler (objref " << canonical_objref << ")");
}
// do a binary search
auto pos = std::lower_bound(objtable_[canonical_objref].begin(), objtable_[canonical_objref].end(), objstoreid);
@@ -372,10 +372,10 @@ void SchedulerService::add_location(ObjRef canonical_objref, ObjStoreId objstore
void SchedulerService::add_canonical_objref(ObjRef objref) {
std::lock_guard<std::mutex> lock(target_objrefs_lock_);
if (objref >= target_objrefs_.size()) {
HALO_LOG(HALO_FATAL, "internal error: attempting to insert objref " << objref << " in target_objrefs_, but target_objrefs_.size() is " << target_objrefs_.size());
RAY_LOG(RAY_FATAL, "internal error: attempting to insert objref " << objref << " in target_objrefs_, but target_objrefs_.size() is " << target_objrefs_.size());
}
if (target_objrefs_[objref] != UNITIALIZED_ALIAS && target_objrefs_[objref] != objref) {
HALO_LOG(HALO_FATAL, "internal error: attempting to declare objref " << objref << " as a canonical objref, but target_objrefs_[objref] is already aliased with objref " << target_objrefs_[objref]);
RAY_LOG(RAY_FATAL, "internal error: attempting to declare objref " << objref << " as a canonical objref, but target_objrefs_[objref] is already aliased with objref " << target_objrefs_[objref]);
}
target_objrefs_[objref] = objref;
}
@@ -433,7 +433,7 @@ void SchedulerService::get_info(const SchedulerInfoRequest& request, SchedulerIn
ObjStoreId SchedulerService::pick_objstore(ObjRef canonical_objref) {
std::mt19937 rng;
if (!is_canonical(canonical_objref)) {
HALO_LOG(HALO_FATAL, "Attempting to call pick_objstore with a non-canonical objref, (objref " << canonical_objref << ")");
RAY_LOG(RAY_FATAL, "Attempting to call pick_objstore with a non-canonical objref, (objref " << canonical_objref << ")");
}
std::uniform_int_distribution<int> uni(0, objtable_[canonical_objref].size() - 1);
ObjStoreId objstoreid = objtable_[canonical_objref][uni(rng)];
@@ -443,7 +443,7 @@ ObjStoreId SchedulerService::pick_objstore(ObjRef canonical_objref) {
bool SchedulerService::is_canonical(ObjRef objref) {
std::lock_guard<std::mutex> lock(target_objrefs_lock_);
if (target_objrefs_[objref] == UNITIALIZED_ALIAS) {
HALO_LOG(HALO_FATAL, "Attempting to call is_canonical on an objref for which aliasing is not complete or the object is not ready, target_objrefs_[objref] == UNITIALIZED_ALIAS for objref " << objref << ".");
RAY_LOG(RAY_FATAL, "Attempting to call is_canonical on an objref for which aliasing is not complete or the object is not ready, target_objrefs_[objref] == UNITIALIZED_ALIAS for objref " << objref << ".");
}
return objref == target_objrefs_[objref];
}
@@ -456,11 +456,11 @@ void SchedulerService::perform_pulls() {
ObjRef objref = pull.second;
WorkerId workerid = pull.first;
if (!has_canonical_objref(objref)) {
HALO_LOG(HALO_ALIAS, "objref " << objref << " does not have a canonical_objref, so continuing");
RAY_LOG(RAY_ALIAS, "objref " << objref << " does not have a canonical_objref, so continuing");
continue;
}
ObjRef canonical_objref = get_canonical_objref(objref);
HALO_LOG(HALO_DEBUG, "attempting to pull objref " << pull.second << " with canonical objref " << canonical_objref << " to objstore " << get_store(workerid));
RAY_LOG(RAY_DEBUG, "attempting to pull objref " << pull.second << " with canonical objref " << canonical_objref << " to objstore " << get_store(workerid));
objtable_lock_.lock();
int num_stores = objtable_[canonical_objref].size();
@@ -538,7 +538,7 @@ void SchedulerService::schedule_tasks_location_aware() {
if (!task.arg(j).has_obj()) {
ObjRef objref = task.arg(j).ref();
if (!has_canonical_objref(objref)) {
HALO_LOG(HALO_FATAL, "no canonical object ref found even though task is ready; that should not be possible!");
RAY_LOG(RAY_FATAL, "no canonical object ref found even though task is ready; that should not be possible!");
}
ObjRef canonical_objref = get_canonical_objref(objref);
// check if the object is already in the local object store
@@ -585,7 +585,7 @@ bool SchedulerService::has_canonical_objref(ObjRef objref) {
ObjRef objref_temp = objref;
while (true) {
if (objref_temp >= target_objrefs_.size()) {
HALO_LOG(HALO_FATAL, "Attempting to index target_objrefs_ with objref " << objref_temp << ", but target_objrefs_.size() = " << target_objrefs_.size());
RAY_LOG(RAY_FATAL, "Attempting to index target_objrefs_ with objref " << objref_temp << ", but target_objrefs_.size() = " << target_objrefs_.size());
}
if (target_objrefs_[objref_temp] == UNITIALIZED_ALIAS) {
return false;
@@ -603,16 +603,16 @@ ObjRef SchedulerService::get_canonical_objref(ObjRef objref) {
ObjRef objref_temp = objref;
while (true) {
if (objref_temp >= target_objrefs_.size()) {
HALO_LOG(HALO_FATAL, "Attempting to index target_objrefs_ with objref " << objref_temp << ", but target_objrefs_.size() = " << target_objrefs_.size());
RAY_LOG(RAY_FATAL, "Attempting to index target_objrefs_ with objref " << objref_temp << ", but target_objrefs_.size() = " << target_objrefs_.size());
}
if (target_objrefs_[objref_temp] == UNITIALIZED_ALIAS) {
HALO_LOG(HALO_FATAL, "Attempting to get canonical objref for objref " << objref << ", which aliases, objref " << objref_temp << ", but target_objrefs_[objref_temp] == UNITIALIZED_ALIAS for objref_temp = " << objref_temp << ".");
RAY_LOG(RAY_FATAL, "Attempting to get canonical objref for objref " << objref << ", which aliases, objref " << objref_temp << ", but target_objrefs_[objref_temp] == UNITIALIZED_ALIAS for objref_temp = " << objref_temp << ".");
}
if (target_objrefs_[objref_temp] == objref_temp) {
return objref_temp;
}
objref_temp = target_objrefs_[objref_temp];
HALO_LOG(HALO_ALIAS, "Looping in get_canonical_objref.");
RAY_LOG(RAY_ALIAS, "Looping in get_canonical_objref.");
}
}
@@ -646,7 +646,7 @@ void SchedulerService::deallocate_object(ObjRef canonical_objref) {
// these methods require reference_counts_lock_ to have been acquired, and
// so the lock must before outside of these methods (it is acquired in
// DecrementRefCount).
HALO_LOG(HALO_REFCOUNT, "Deallocating canonical_objref " << canonical_objref << ".");
RAY_LOG(RAY_REFCOUNT, "Deallocating canonical_objref " << canonical_objref << ".");
{
std::lock_guard<std::mutex> objtable_lock(objtable_lock_);
auto &objstores = objtable_[canonical_objref];
@@ -657,7 +657,7 @@ void SchedulerService::deallocate_object(ObjRef canonical_objref) {
DeallocateObjectRequest request;
request.set_canonical_objref(canonical_objref);
ObjStoreId objstoreid = objstores[i];
HALO_LOG(HALO_REFCOUNT, "Attempting to deallocate canonical_objref " << canonical_objref << " from objstore " << objstoreid);
RAY_LOG(RAY_REFCOUNT, "Attempting to deallocate canonical_objref " << canonical_objref << " from objstore " << objstoreid);
objstores_[objstoreid].objstore_stub->DeallocateObject(&context, request, &reply);
}
objtable_[canonical_objref].clear();
@@ -670,10 +670,10 @@ void SchedulerService::increment_ref_count(std::vector<ObjRef> &objrefs) {
for (int i = 0; i < objrefs.size(); ++i) {
ObjRef objref = objrefs[i];
if (reference_counts_[objref] == DEALLOCATED) {
HALO_LOG(HALO_FATAL, "Attempting to increment the reference count for objref " << objref << ", but this object appears to have been deallocated already.");
RAY_LOG(RAY_FATAL, "Attempting to increment the reference count for objref " << objref << ", but this object appears to have been deallocated already.");
}
reference_counts_[objref] += 1;
HALO_LOG(HALO_REFCOUNT, "Incremented ref count for objref " << objref <<". New reference count is " << reference_counts_[objref]);
RAY_LOG(RAY_REFCOUNT, "Incremented ref count for objref " << objref <<". New reference count is " << reference_counts_[objref]);
}
}
@@ -682,13 +682,13 @@ void SchedulerService::decrement_ref_count(std::vector<ObjRef> &objrefs) {
for (int i = 0; i < objrefs.size(); ++i) {
ObjRef objref = objrefs[i];
if (reference_counts_[objref] == DEALLOCATED) {
HALO_LOG(HALO_FATAL, "Attempting to decrement the reference count for objref " << objref << ", but this object appears to have been deallocated already.");
RAY_LOG(RAY_FATAL, "Attempting to decrement the reference count for objref " << objref << ", but this object appears to have been deallocated already.");
}
if (reference_counts_[objref] == 0) {
HALO_LOG(HALO_FATAL, "Attempting to decrement the reference count for objref " << objref << ", but the reference count for this object is already 0.");
RAY_LOG(RAY_FATAL, "Attempting to decrement the reference count for objref " << objref << ", but the reference count for this object is already 0.");
}
reference_counts_[objref] -= 1;
HALO_LOG(HALO_REFCOUNT, "Decremented ref count for objref " << objref << ". New reference count is " << reference_counts_[objref]);
RAY_LOG(RAY_REFCOUNT, "Decremented ref count for objref " << objref << ". New reference count is " << reference_counts_[objref]);
// See if we can deallocate the object
std::vector<ObjRef> equivalent_objrefs;
get_equivalent_objrefs(objref, equivalent_objrefs);
@@ -702,7 +702,7 @@ void SchedulerService::decrement_ref_count(std::vector<ObjRef> &objrefs) {
if (can_deallocate) {
ObjRef canonical_objref = equivalent_objrefs[0];
if (!is_canonical(canonical_objref)) {
HALO_LOG(HALO_FATAL, "canonical_objref is not canonical.");
RAY_LOG(RAY_FATAL, "canonical_objref is not canonical.");
}
deallocate_object(canonical_objref);
for (int j = 0; j < equivalent_objrefs.size(); ++j) {
@@ -724,7 +724,7 @@ void SchedulerService::get_equivalent_objrefs(ObjRef objref, std::vector<ObjRef>
std::lock_guard<std::mutex> target_objrefs_lock(target_objrefs_lock_);
ObjRef downstream_objref = objref;
while (target_objrefs_[downstream_objref] != downstream_objref && target_objrefs_[downstream_objref] != UNITIALIZED_ALIAS) {
HALO_LOG(HALO_ALIAS, "Looping in get_equivalent_objrefs");
RAY_LOG(RAY_ALIAS, "Looping in get_equivalent_objrefs");
downstream_objref = target_objrefs_[downstream_objref];
}
std::lock_guard<std::mutex> reverse_target_objrefs_lock(reverse_target_objrefs_lock_);
@@ -755,7 +755,7 @@ char* get_cmd_option(char** begin, char** end, const std::string& option) {
int main(int argc, char** argv) {
SchedulingAlgorithmType scheduling_algorithm = SCHEDULING_ALGORITHM_LOCALITY_AWARE;
if (argc < 2) {
HALO_LOG(HALO_FATAL, "scheduler: expected at least one argument (scheduler ip address)");
RAY_LOG(RAY_FATAL, "scheduler: expected at least one argument (scheduler ip address)");
return 1;
}
if (argc > 2) {
src/scheduler.h
+4 -4
View File
@@ -1,5 +1,5 @@
#ifndef HALO_SCHEDULER_H
#define HALO_SCHEDULER_H
#ifndef RAY_SCHEDULER_H
#define RAY_SCHEDULER_H
#include <deque>
@@ -10,8 +10,8 @@
#include <grpc++/grpc++.h>
#include "halo/halo.h"
#include "halo.grpc.pb.h"
#include "ray/ray.h"
#include "ray.grpc.pb.h"
#include "types.pb.h"
#include "computation_graph.h"
src/utils.h
+4 -4
View File
@@ -1,5 +1,5 @@
#ifndef HALO_UTILS_H
#define HALO_UTILS_H
#ifndef RAY_UTILS_H
#define RAY_UTILS_H
inline std::string::iterator split_ip_address(std::string& ip_address) {
if (ip_address[0] == '[') { // IPv6
@@ -10,11 +10,11 @@ inline std::string::iterator split_ip_address(std::string& ip_address) {
if(split_end != ip_address.end() && *split_end == ':') {
return split_end;
}
HALO_LOG(HALO_FATAL, "ip address should contain a port number");
RAY_LOG(RAY_FATAL, "ip address should contain a port number");
} else { // IPv4
auto split_point = std::find(ip_address.rbegin(), ip_address.rend(), ':').base();
if (split_point == ip_address.begin()) {
HALO_LOG(HALO_FATAL, "ip address should contain a port number");
RAY_LOG(RAY_FATAL, "ip address should contain a port number");
} else {
return split_point;
}
src/worker.cc
+21 -21
View File
@@ -5,12 +5,12 @@
#include <pynumbuf/serialize.h>
extern "C" {
static PyObject *HaloError;
static PyObject *RayError;
}
Status WorkerServiceImpl::ExecuteTask(ServerContext* context, const ExecuteTaskRequest* request, ExecuteTaskReply* reply) {
task_ = request->task(); // Copy task
HALO_LOG(HALO_INFO, "invoked task " << request->task().name());
RAY_LOG(RAY_INFO, "invoked task " << request->task().name());
Task* taskptr = &task_;
send_queue_.send(&taskptr);
return Status::OK;
@@ -26,7 +26,7 @@ Worker::Worker(const std::string& worker_address, std::shared_ptr<Channel> sched
SubmitTaskReply Worker::submit_task(SubmitTaskRequest* request) {
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform submit_task, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform submit_task, but connected_ = " << connected_ << ".");
}
SubmitTaskReply reply;
ClientContext context;
@@ -52,7 +52,7 @@ void Worker::register_worker(const std::string& worker_address, const std::strin
void Worker::request_object(ObjRef objref) {
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform request_object, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform request_object, but connected_ = " << connected_ << ".");
}
RequestObjRequest request;
request.set_workerid(workerid_);
@@ -66,7 +66,7 @@ void Worker::request_object(ObjRef objref) {
ObjRef Worker::get_objref() {
// first get objref for the new object
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform get_objref, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform get_objref, but connected_ = " << connected_ << ".");
}
PushObjRequest push_request;
PushObjReply push_reply;
@@ -78,7 +78,7 @@ ObjRef Worker::get_objref() {
slice Worker::get_object(ObjRef objref) {
// get_object assumes that objref is a canonical objref
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform get_object, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform get_object, but connected_ = " << connected_ << ".");
}
ObjRequest request;
request.workerid = workerid_;
@@ -97,7 +97,7 @@ slice Worker::get_object(ObjRef objref) {
// contained_objrefs is a vector of all the objrefs contained in obj
void Worker::put_object(ObjRef objref, const Obj* obj, std::vector<ObjRef> &contained_objrefs) {
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform put_object, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform put_object, but connected_ = " << connected_ << ".");
}
std::string data;
obj->SerializeToString(&data); // TODO(pcm): get rid of this serialization
@@ -108,7 +108,7 @@ void Worker::put_object(ObjRef objref, const Obj* obj, std::vector<ObjRef> &cont
request.size = data.size();
request_obj_queue_.send(&request);
if (contained_objrefs.size() > 0) {
HALO_LOG(HALO_REFCOUNT, "In put_object, calling increment_reference_count for contained objrefs");
RAY_LOG(RAY_REFCOUNT, "In put_object, calling increment_reference_count for contained objrefs");
increment_reference_count(contained_objrefs); // Notify the scheduler that some object references are serialized in the objstore.
}
ObjHandle result;
@@ -135,14 +135,14 @@ void Worker::put_object(ObjRef objref, const Obj* obj, std::vector<ObjRef> &cont
arrow::Status _s = (s); \
if (!_s.ok()) { \
std::string _errmsg = std::string(msg) + _s.ToString(); \
PyErr_SetString(HaloError, _errmsg.c_str()); \
PyErr_SetString(RayError, _errmsg.c_str()); \
return NULL; \
} \
} while (0);
PyObject* Worker::put_arrow(ObjRef objref, PyObject* value) {
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform put_arrow, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform put_arrow, but connected_ = " << connected_ << ".");
}
ObjRequest request;
pynumbuf::PythonObjectWriter writer;
@@ -168,7 +168,7 @@ PyObject* Worker::put_arrow(ObjRef objref, PyObject* value) {
PyObject* Worker::get_arrow(ObjRef objref) {
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform get_arrow, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform get_arrow, but connected_ = " << connected_ << ".");
}
ObjRequest request;
request.workerid = workerid_;
@@ -186,7 +186,7 @@ PyObject* Worker::get_arrow(ObjRef objref) {
bool Worker::is_arrow(ObjRef objref) {
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform is_arrow, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform is_arrow, but connected_ = " << connected_ << ".");
}
ObjRequest request;
request.workerid = workerid_;
@@ -200,7 +200,7 @@ bool Worker::is_arrow(ObjRef objref) {
void Worker::alias_objrefs(ObjRef alias_objref, ObjRef target_objref) {
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform alias_objrefs, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform alias_objrefs, but connected_ = " << connected_ << ".");
}
ClientContext context;
AliasObjRefsRequest request;
@@ -212,14 +212,14 @@ void Worker::alias_objrefs(ObjRef alias_objref, ObjRef target_objref) {
void Worker::increment_reference_count(std::vector<ObjRef> &objrefs) {
if (!connected_) {
HALO_LOG(HALO_DEBUG, "Attempting to increment_reference_count for objrefs, but connected_ = " << connected_ << " so returning instead.");
RAY_LOG(RAY_DEBUG, "Attempting to increment_reference_count for objrefs, but connected_ = " << connected_ << " so returning instead.");
return;
}
if (objrefs.size() > 0) {
ClientContext context;
IncrementRefCountRequest request;
for (int i = 0; i < objrefs.size(); ++i) {
HALO_LOG(HALO_REFCOUNT, "Incrementing reference count for objref " << objrefs[i]);
RAY_LOG(RAY_REFCOUNT, "Incrementing reference count for objref " << objrefs[i]);
request.add_objref(objrefs[i]);
}
AckReply reply;
@@ -229,14 +229,14 @@ void Worker::increment_reference_count(std::vector<ObjRef> &objrefs) {
void Worker::decrement_reference_count(std::vector<ObjRef> &objrefs) {
if (!connected_) {
HALO_LOG(HALO_DEBUG, "Attempting to decrement_reference_count, but connected_ = " << connected_ << " so returning instead.");
RAY_LOG(RAY_DEBUG, "Attempting to decrement_reference_count, but connected_ = " << connected_ << " so returning instead.");
return;
}
if (objrefs.size() > 0) {
ClientContext context;
DecrementRefCountRequest request;
for (int i = 0; i < objrefs.size(); ++i) {
HALO_LOG(HALO_REFCOUNT, "Decrementing reference count for objref " << objrefs[i]);
RAY_LOG(RAY_REFCOUNT, "Decrementing reference count for objref " << objrefs[i]);
request.add_objref(objrefs[i]);
}
AckReply reply;
@@ -246,7 +246,7 @@ void Worker::decrement_reference_count(std::vector<ObjRef> &objrefs) {
void Worker::register_function(const std::string& name, size_t num_return_vals) {
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform register_function, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform register_function, but connected_ = " << connected_ << ".");
}
ClientContext context;
RegisterFunctionRequest request;
@@ -265,7 +265,7 @@ Task* Worker::receive_next_task() {
void Worker::notify_task_completed() {
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to perform notify_task_completed, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to perform notify_task_completed, but connected_ = " << connected_ << ".");
}
ClientContext context;
WorkerReadyRequest request;
@@ -285,7 +285,7 @@ bool Worker::connected() {
// TODO(rkn): Should we be using pointers or references? And should they be const?
void Worker::scheduler_info(ClientContext &context, SchedulerInfoRequest &request, SchedulerInfoReply &reply) {
if (!connected_) {
HALO_LOG(HALO_FATAL, "Attempting to get scheduler info, but connected_ = " << connected_ << ".");
RAY_LOG(RAY_FATAL, "Attempting to get scheduler info, but connected_ = " << connected_ << ".");
}
scheduler_stub_->SchedulerInfo(&context, request, &reply);
}
@@ -306,7 +306,7 @@ void Worker::start_worker_service() {
builder.AddListeningPort(std::string("0.0.0.0:") + port, grpc::InsecureServerCredentials());
builder.RegisterService(&service);
std::unique_ptr<Server> server(builder.BuildAndStart());
HALO_LOG(HALO_INFO, "worker server listening on " << service_address);
RAY_LOG(RAY_INFO, "worker server listening on " << service_address);
server->Wait();
});
}
src/worker.h
+4 -4
View File
@@ -1,5 +1,5 @@
#ifndef HALO_WORKER_H
#define HALO_WORKER_H
#ifndef RAY_WORKER_H
#define RAY_WORKER_H
#include <iostream>
#include <memory>
@@ -15,8 +15,8 @@ using grpc::ServerBuilder;
using grpc::ServerContext;
using grpc::Status;
#include "halo.grpc.pb.h"
#include "halo/halo.h"
#include "ray.grpc.pb.h"
#include "ray/ray.h"
#include "ipc.h"
using grpc::Channel;