All Classes Namespaces Functions Variables Typedefs Enumerations Friends
worker.h
1 
5 #pragma once
6 
7 #include <functional>
8 #include <memory>
9 #include <mutex>
10 #include <queue>
11 #include <string>
12 #include <thread>
13 
14 #include <ucp/api/ucp.h>
15 
16 #include <ucxx/component.h>
17 #include <ucxx/constructors.h>
18 #include <ucxx/context.h>
19 #include <ucxx/delayed_submission.h>
20 #include <ucxx/future.h>
21 #include <ucxx/inflight_requests.h>
22 #include <ucxx/notifier.h>
23 #include <ucxx/typedefs.h>
24 #include <ucxx/worker_progress_thread.h>
25 
26 namespace ucxx {
27 
28 class Address;
29 class Buffer;
30 class Endpoint;
31 class Listener;
32 class RequestAm;
33 
34 namespace internal {
35 class AmData;
36 } // namespace internal
37 
44 class Worker : public Component {
45  private:
46  ucp_worker_h _handle{nullptr};
47  int _epollFileDescriptor{-1};
48  int _workerFileDescriptor{-1};
49  std::mutex _inflightRequestsMutex{};
50  std::unique_ptr<InflightRequests> _inflightRequests{
51  std::make_unique<InflightRequests>()};
52  std::mutex
53  _inflightRequestsToCancelMutex{};
54  std::unique_ptr<InflightRequests> _inflightRequestsToCancel{
55  std::make_unique<InflightRequests>()};
56  WorkerProgressThread _progressThread{};
57  std::thread::id _progressThreadId{};
58  std::function<void(void*)> _progressThreadStartCallback{
59  nullptr};
60  void* _progressThreadStartCallbackArg{
61  nullptr};
62  std::shared_ptr<DelayedSubmissionCollection> _delayedSubmissionCollection{
63  nullptr};
64 
65  friend std::shared_ptr<RequestAm> createRequestAm(
66  std::shared_ptr<Endpoint> endpoint,
67  const std::variant<data::AmSend, data::AmReceive> requestData,
68  const bool enablePythonFuture,
69  RequestCallbackUserFunction callbackFunction,
70  RequestCallbackUserData callbackData);
71 
72  protected:
74  false};
75  std::mutex _futuresPoolMutex{};
76  std::queue<std::shared_ptr<Future>>
78  std::shared_ptr<Notifier> _notifier{nullptr};
79  std::shared_ptr<internal::AmData>
81 
82  private:
89  void drainWorkerTagRecv();
90 
105  [[nodiscard]] std::shared_ptr<RequestAm> getAmRecv(
106  ucp_ep_h ep, std::function<std::shared_ptr<RequestAm>()> createAmRecvRequestFunction);
107 
114  void stopProgressThreadNoWarn();
115 
126  [[nodiscard]] std::shared_ptr<Request> registerInflightRequest(std::shared_ptr<Request> request);
127 
135  bool progressPending();
136 
137  protected:
155  explicit Worker(std::shared_ptr<Context> context,
156  const bool enableDelayedSubmission = false,
157  const bool enableFuture = false);
158 
159  public:
160  Worker() = delete;
161  Worker(const Worker&) = delete;
162  Worker& operator=(Worker const&) = delete;
163  Worker(Worker&& o) = delete;
164  Worker& operator=(Worker&& o) = delete;
165 
190  [[nodiscard]] friend std::shared_ptr<Worker> createWorker(std::shared_ptr<Context> context,
191  const bool enableDelayedSubmission,
192  const bool enableFuture);
193 
197  virtual ~Worker();
198 
214  [[nodiscard]] ucp_worker_h getHandle();
215 
224  [[nodiscard]] std::string getInfo();
225 
255 
272 
283  bool arm();
284 
312  bool progressWorkerEvent(const int epollTimeout = -1);
313 
348  void signal();
349 
371  bool waitProgress();
372 
387  bool progressOnce();
388 
404  bool progress();
405 
422  void registerDelayedSubmission(std::shared_ptr<Request> request,
424 
453  uint64_t period = 0);
454 
482  uint64_t period = 0);
483 
491  [[nodiscard]] bool isDelayedRequestSubmissionEnabled() const;
492 
500  [[nodiscard]] bool isFutureEnabled() const;
501 
513  virtual void populateFuturesPool();
514 
524  virtual void clearFuturesPool();
525 
537  [[nodiscard]] virtual std::shared_ptr<Future> getFuture();
538 
553  [[nodiscard]] virtual RequestNotifierWaitState waitRequestNotifier(uint64_t periodNs);
554 
567  virtual void runRequestNotifier();
568 
577 
588  void setProgressThreadStartCallback(std::function<void(void*)> callback, void* callbackArg);
589 
602  void startProgressThread(const bool pollingMode = false, const int epollTimeout = 1);
603 
613 
621  [[nodiscard]] bool isProgressThreadRunning();
622 
630  [[nodiscard]] std::thread::id getProgressThreadId();
631 
652  size_t cancelInflightRequests(uint64_t period = 0, uint64_t maxAttempts = 1);
653 
667 
680  void removeInflightRequest(const Request* const request);
681 
701  [[nodiscard]] bool tagProbe(const Tag tag);
702 
727  [[nodiscard]] std::shared_ptr<Request> tagRecv(
728  void* buffer,
729  size_t length,
730  Tag tag,
731  TagMask tagMask,
732  const bool enableFuture = false,
733  RequestCallbackUserFunction callbackFunction = nullptr,
734  RequestCallbackUserData callbackData = nullptr);
735 
747  [[nodiscard]] std::shared_ptr<Address> getAddress();
748 
773  [[nodiscard]] std::shared_ptr<Endpoint> createEndpointFromHostname(
774  std::string ipAddress, uint16_t port, bool endpointErrorHandling = true);
775 
804  [[nodiscard]] std::shared_ptr<Endpoint> createEndpointFromWorkerAddress(
805  std::shared_ptr<Address> address, bool endpointErrorHandling = true);
806 
824  [[nodiscard]] std::shared_ptr<Listener> createListener(uint16_t port,
825  ucp_listener_conn_callback_t callback,
826  void* callbackArgs);
827 
855  void registerAmAllocator(ucs_memory_type_t memoryType, AmAllocatorType allocator);
856 
895 
915  [[nodiscard]] bool amProbe(const ucp_ep_h endpointHandle) const;
916 
938  [[nodiscard]] std::shared_ptr<Request> flush(
939  const bool enablePythonFuture = false,
940  RequestCallbackUserFunction callbackFunction = nullptr,
941  RequestCallbackUserData callbackData = nullptr);
942 };
943 
944 } // namespace ucxx
Information of an Active Message receiver callback.
Definition: typedefs.h:154
A UCXX component class to prevent early destruction of parent object.
Definition: component.h:17
Base type for a UCXX transfer request.
Definition: request.h:38
A thread to progress a ucxx::Worker.
Definition: worker_progress_thread.h:48
Component encapsulating a UCP worker.
Definition: worker.h:44
bool amProbe(const ucp_ep_h endpointHandle) const
Check for uncaught active messages.
int getEpollFileDescriptor()
Get the epoll file descriptor associated with the worker.
bool progress()
Progress the worker until all communication events are completed.
void registerAmReceiverCallback(AmReceiverCallbackInfo info, AmReceiverCallbackType callback)
Register receiver callback for active messages.
void signal()
Signal the worker that an event happened.
virtual void populateFuturesPool()
Populate the futures pool.
bool registerGenericPre(DelayedSubmissionCallbackType callback, uint64_t period=0)
Register callback to be executed in progress thread before progressing.
void setProgressThreadStartCallback(std::function< void(void *)> callback, void *callbackArg)
Set callback to be executed at the progress thread start.
bool progressOnce()
Progress the worker only once.
void startProgressThread(const bool pollingMode=false, const int epollTimeout=1)
Start the progress thread.
bool isProgressThreadRunning()
Inquire if worker has a progress thread running.
friend std::shared_ptr< RequestAm > createRequestAm(std::shared_ptr< Endpoint > endpoint, const std::variant< data::AmSend, data::AmReceive > requestData, const bool enablePythonFuture, RequestCallbackUserFunction callbackFunction, RequestCallbackUserData callbackData)
bool tagProbe(const Tag tag)
Check for uncaught tag messages.
void registerAmAllocator(ucs_memory_type_t memoryType, AmAllocatorType allocator)
Register allocator for active messages.
void stopProgressThread()
Stop the progress thread.
Worker(std::shared_ptr< Context > context, const bool enableDelayedSubmission=false, const bool enableFuture=false)
Protected constructor of ucxx::Worker.
std::shared_ptr< Endpoint > createEndpointFromHostname(std::string ipAddress, uint16_t port, bool endpointErrorHandling=true)
Create endpoint to worker listening on specific IP and port.
std::shared_ptr< Listener > createListener(uint16_t port, ucp_listener_conn_callback_t callback, void *callbackArgs)
Listen for remote connections on given port.
std::queue< std::shared_ptr< Future > > _futuresPool
Futures pool to prevent running out of fresh futures.
Definition: worker.h:77
bool isDelayedRequestSubmissionEnabled() const
Inquire if worker has been created with delayed submission enabled.
virtual ~Worker()
ucxx::Worker destructor.
virtual void clearFuturesPool()
Clear the futures pool.
ucp_worker_h getHandle()
Get the underlying ucp_worker_h handle.
std::mutex _futuresPoolMutex
Mutex to access the futures pool.
Definition: worker.h:75
virtual RequestNotifierWaitState waitRequestNotifier(uint64_t periodNs)
Block until a request event.
void scheduleRequestCancel(TrackedRequestsPtr trackedRequests)
Schedule cancelation of inflight requests.
bool arm()
Arm the UCP worker.
std::shared_ptr< Request > flush(const bool enablePythonFuture=false, RequestCallbackUserFunction callbackFunction=nullptr, RequestCallbackUserData callbackData=nullptr)
Enqueue a flush operation.
std::shared_ptr< Endpoint > createEndpointFromWorkerAddress(std::shared_ptr< Address > address, bool endpointErrorHandling=true)
Create endpoint to worker located at UCX address.
bool waitProgress()
Block until an event has happened, then progresses.
virtual void runRequestNotifier()
Notify futures of each completed communication request.
bool registerGenericPost(DelayedSubmissionCallbackType callback, uint64_t period=0)
Register callback to be executed in progress thread before progressing.
void removeInflightRequest(const Request *const request)
Remove reference to request from internal container.
std::shared_ptr< Notifier > _notifier
Notifier object.
Definition: worker.h:78
void registerDelayedSubmission(std::shared_ptr< Request > request, DelayedSubmissionCallbackType callback)
Register delayed request submission.
std::thread::id getProgressThreadId()
Get the progress thread ID.
size_t cancelInflightRequests(uint64_t period=0, uint64_t maxAttempts=1)
Cancel inflight requests.
bool _enableFuture
Boolean identifying whether the worker was created with future capability.
Definition: worker.h:73
bool isFutureEnabled() const
Inquire if worker has been created with future support.
virtual std::shared_ptr< Future > getFuture()
Get a future from the pool.
virtual void stopRequestNotifierThread()
Signal the notifier to terminate.
std::shared_ptr< Address > getAddress()
Get the address of the UCX worker object.
friend std::shared_ptr< Worker > createWorker(std::shared_ptr< Context > context, const bool enableDelayedSubmission, const bool enableFuture)
Constructor of shared_ptr<ucxx::Worker>.
bool progressWorkerEvent(const int epollTimeout=-1)
Progress worker event while in blocking progress mode.
std::shared_ptr< internal::AmData > _amData
Worker data made available to Active Messages callback.
Definition: worker.h:80
std::shared_ptr< Request > tagRecv(void *buffer, size_t length, Tag tag, TagMask tagMask, const bool enableFuture=false, RequestCallbackUserFunction callbackFunction=nullptr, RequestCallbackUserData callbackData=nullptr)
Enqueue a tag receive operation.
void initBlockingProgressMode()
Initialize blocking progress mode.
std::string getInfo()
Get information about the underlying ucp_worker_h object.
Definition: address.h:15
std::function< void(ucs_status_t, std::shared_ptr< void >)> RequestCallbackUserFunction
A user-defined function to execute as part of a ucxx::Request callback.
Definition: typedefs.h:89
std::shared_ptr< void > RequestCallbackUserData
Data for the user-defined function provided to the ucxx::Request callback.
Definition: typedefs.h:97
std::function< void()> DelayedSubmissionCallbackType
A user-defined function to execute as part of delayed submission callback.
Definition: delayed_submission.h:32
std::function< std::shared_ptr< Buffer >size_t)> AmAllocatorType
Custom Active Message allocator type.
Definition: typedefs.h:121
std::unique_ptr< TrackedRequests > TrackedRequestsPtr
Pre-defined type for a pointer to a container of tracked requests.
Definition: inflight_requests.h:44
std::function< void(std::shared_ptr< Request >)> AmReceiverCallbackType
Active Message receiver callback.
Definition: typedefs.h:129
RequestNotifierWaitState
The state with which a wait operation completed.
Definition: notifier.h:26
TagMask
Strong type for a UCP tag mask.
Definition: typedefs.h:66
Tag
Strong type for a UCP tag.
Definition: typedefs.h:58