artdaq  v3_04_00
SharedMemoryEventManager.hh
1 #ifndef ARTDAQ_DAQRATE_SHAREDMEMORYEVENTMANAGER_HH
2 #define ARTDAQ_DAQRATE_SHAREDMEMORYEVENTMANAGER_HH
3 
4 #include "artdaq/DAQdata/Globals.hh" // Before trace.h gets included in ConcurrentQueue (from GlobalQueue)
5 
6 #include <sys/stat.h>
7 #include <deque>
8 #include <fstream>
9 #include <iomanip>
10 #include <set>
11 #include "artdaq-core/Core/SharedMemoryManager.hh"
12 #include "artdaq-core/Data/RawEvent.hh"
13 #include "artdaq/Application/StatisticsHelper.hh"
14 #include "artdaq/DAQrate/RequestSender.hh"
15 #include "artdaq/DAQrate/detail/ArtConfig.hh"
16 #include "fhiclcpp/fwd.h"
17 #define ART_SUPPORTS_DUPLICATE_EVENTS 0
18 
19 namespace artdaq {
20 
25 {
26 public:
31  art_config_file(fhicl::ParameterSet ps /*, uint32_t shm_key, uint32_t broadcast_key*/)
32  : dir_name_("/tmp/partition_" + std::to_string(GetPartitionNumber()))
33  , file_name_(dir_name_ + "/artConfig_" + std::to_string(my_rank) + "_" + std::to_string(artdaq::TimeUtils::gettimeofday_us()) + ".fcl")
34  {
35  mkdir(dir_name_.c_str(), 0777); // Allowed to fail if directory already exists
36 
37  std::ofstream of(file_name_, std::ofstream::trunc);
38  if (of.fail())
39  {
40  // Probably a permissions error...
41  dir_name_ = "/tmp/partition_" + std::to_string(GetPartitionNumber()) + "_" + std::to_string(getuid());
42  mkdir(dir_name_.c_str(), 0777); // Allowed to fail if directory already exists
43  file_name_ = dir_name_ + "/artConfig_" + std::to_string(my_rank) + "_" + std::to_string(artdaq::TimeUtils::gettimeofday_us()) + ".fcl";
44 
45  of.open(file_name_, std::ofstream::trunc);
46  if (of.fail())
47  {
48  TLOG(TLVL_ERROR) << "Failed to open configuration file after two attemps! ABORTING!";
49  exit(46);
50  }
51  }
52  of << ps.to_string();
53 
54  //if (ps.has_key("services.NetMonTransportServiceInterface"))
55  //{
56  // of << " services.NetMonTransportServiceInterface.shared_memory_key: 0x" << std::hex << shm_key;
57  // of << " services.NetMonTransportServiceInterface.broadcast_shared_memory_key: 0x" << std::hex << broadcast_key;
58  // of << " services.NetMonTransportServiceInterface.rank: " << std::dec << my_rank;
59  //}
60  if (!ps.has_key("services.message"))
61  {
62  of << " services.message: { " << generateMessageFacilityConfiguration("art") << "} ";
63  }
64  //of << " source.shared_memory_key: 0x" << std::hex << shm_key;
65  //of << " source.broadcast_shared_memory_key: 0x" << std::hex << broadcast_key;
66  //of << " source.rank: " << std::dec << my_rank;
67  of.close();
68  }
70  {
71  remove(file_name_.c_str());
72  rmdir(dir_name_.c_str()); // Will only delete directory if no config files are left over
73  }
78  std::string getFileName() const { return file_name_; }
79 
80 private:
81  std::string dir_name_;
82  std::string file_name_;
83 };
84 
88 class SharedMemoryEventManager : public SharedMemoryManager
89 {
90 public:
91  typedef RawEvent::run_id_t run_id_t;
92  typedef RawEvent::subrun_id_t subrun_id_t;
93  typedef Fragment::sequence_id_t sequence_id_t;
94  typedef std::map<sequence_id_t, RawEvent_ptr> EventMap;
95 
99  struct Config
100  {
103  fhicl::Atom<size_t> max_event_size_bytes{fhicl::Name{"max_event_size_bytes"}, fhicl::Comment{"Maximum event size (all Fragments), in bytes"}};
105  fhicl::Atom<size_t> stale_buffer_timeout_usec{fhicl::Name{"stale_buffer_timeout_usec"}, fhicl::Comment{"Maximum amount of time elapsed before a buffer is marked as abandoned. Time is reset each time an operation is performed on the buffer."}, 5000000};
107  fhicl::Atom<bool> overwrite_mode{fhicl::Name{"overwrite_mode"}, fhicl::Comment{"Whether buffers are allowed to be overwritten when safe (state == Full or Reading)"}, false};
109  fhicl::Atom<bool> restart_crashed_art_processes{fhicl::Name{"restart_crashed_art_processes"}, fhicl::Comment{"Whether to automatically restart art processes that fail for any reason"}, true};
111  fhicl::Atom<uint32_t> shared_memory_key{fhicl::Name{"shared_memory_key"}, fhicl::Comment{"Key to use for shared memory access"}, 0xBEE70000 + getpid()};
113  fhicl::Atom<size_t> buffer_count{fhicl::Name{"buffer_count"}, fhicl::Comment{"Number of events in the Shared Memory (incomplete + pending art)"}};
115  fhicl::Atom<size_t> max_subrun_lookup_table_size{fhicl::Name{"max_subrun_lookup_table_size"}, fhicl::Comment{"Maximum number of entries in the subrun rollover history"}, 100};
118  fhicl::Atom<size_t> max_fragment_size_bytes{fhicl::Name{"max_fragment_size_bytes"}, fhicl::Comment{" Maximum Fragment size, in bytes"}};
120  fhicl::Atom<size_t> event_queue_wait_time{fhicl::Name{"event_queue_wait_time"}, fhicl::Comment{"Amount of time (in seconds) an event can exist in shared memory before being released to art. Used as input to default parameter of \"stale_buffer_timeout_usec\"."}, 5};
122  fhicl::Atom<bool> broadcast_mode{fhicl::Name{"broadcast_mode"}, fhicl::Comment{"When true, buffers are not marked Empty when read, but return to Full state. Buffers are overwritten in order received."}, false};
124  fhicl::Atom<size_t> art_analyzer_count{fhicl::Name{"art_analyzer_count"}, fhicl::Comment{"Number of art procceses to start"}, 1};
126  fhicl::Atom<size_t> expected_fragments_per_event{fhicl::Name{"expected_fragments_per_event"}, fhicl::Comment{"Number of Fragments to expect per event"}};
128  fhicl::Atom<int> maximum_oversize_fragment_count{fhicl::Name{"maximum_oversize_fragment_count"}, fhicl::Comment{"Maximum number of over-size Fragments to drop before throwing an exception. Default is 1, which means to throw an exception if any over-size Fragments are dropped. Set to 0 to disable."}, 1};
130  fhicl::Atom<bool> update_run_ids_on_new_fragment{fhicl::Name{"update_run_ids_on_new_fragment"}, fhicl::Comment{"Whether the run and subrun ID of an event should be updated whenever a Fragment is added."}, true};
132  fhicl::Atom<bool> use_sequence_id_for_event_number{fhicl::Name{"use_sequence_id_for_event_number"}, fhicl::Comment{"Whether to use the artdaq Sequence ID (true) or the Timestamp (false) for art Event numbers"}, true};
134  fhicl::Atom<size_t> max_subrun_event_map_length{fhicl::Name{"max_subrun_event_map_length"}, fhicl::Comment{"The maximum number of entries to store in the sequence ID-SubRun ID lookup table"}, 100};
136  fhicl::Atom<bool> send_init_fragments{fhicl::Name{"send_init_fragments"}, fhicl::Comment{"Whether Init Fragments are expected to be sent to art. If true, a Warning message is printed when an Init Fragment is requested but none are available."}, true};
138  fhicl::Atom<int> incomplete_event_report_interval_ms{fhicl::Name{"incomplete_event_report_interval_ms"}, fhicl::Comment{"Interval at which an incomplete event report should be written"}, -1};
141  fhicl::Atom<int> fragment_broadcast_timeout_ms{fhicl::Name{"fragment_broadcast_timeout_ms"}, fhicl::Comment{"Amount of time broadcast fragments should live in the broadcast shared memory segment"}, 3000};
143  fhicl::Atom<double> minimum_art_lifetime_s{fhicl::Name{"minimum_art_lifetime_s"}, fhicl::Comment{"Amount of time that an art process should run to not be considered \"DOA\""}, 2.0};
146  fhicl::Atom<size_t> expected_art_event_processing_time_us{fhicl::Name{"expected_art_event_processing_time_us"}, fhicl::Comment{"During shutdown, SMEM will wait for this amount of time while it is checking that the art threads are done reading buffers."}, 100000};
148  fhicl::Atom<uint32_t> broadcast_shared_memory_key{fhicl::Name{"broadcast_shared_memory_key"}, fhicl::Comment{""}, 0xCEE70000 + getpid()};
150  fhicl::Atom<size_t> broadcast_buffer_count{fhicl::Name{"broadcast_buffer_count"}, fhicl::Comment{"Buffers in the broadcast shared memory segment"}, 10};
152  fhicl::Atom<size_t> broadcast_buffer_size{fhicl::Name{"broadcast_buffer_size"}, fhicl::Comment{"Size of the buffers in the broadcast shared memory segment"}, 0x100000};
154  fhicl::Atom<bool> use_art{fhicl::Name{"use_art"}, fhicl::Comment{"Whether to start and manage art threads (Sets art_analyzer count to 0 and overwrite_mode to true when false)"}, true};
156  fhicl::Atom<bool> manual_art{fhicl::Name{"manual_art"}, fhicl::Comment{"Prints the startup command line for the art process so that the user may (for example) run it in GDB or valgrind"}, false};
157 
158  fhicl::TableFragment<artdaq::RequestSender::Config> requestSenderConfig;
159  };
160  using Parameters = fhicl::WrappedTable<Config>;
161 
167  SharedMemoryEventManager(fhicl::ParameterSet pset, fhicl::ParameterSet art_pset);
171  virtual ~SharedMemoryEventManager();
172 
173 private:
180  bool AddFragment(detail::RawFragmentHeader frag, void* dataPtr);
181 
182 public:
190  bool AddFragment(FragmentPtr frag, size_t timeout_usec, FragmentPtr& outfrag);
191 
198  RawDataType* WriteFragmentHeader(detail::RawFragmentHeader frag, bool dropIfNoBuffersAvailable = false);
199 
204  void DoneWritingFragment(detail::RawFragmentHeader frag);
205 
210  size_t GetIncompleteEventCount() { return active_buffers_.size(); }
211 
216  size_t GetPendingEventCount() { return pending_buffers_.size(); }
217 
222  size_t GetLockedBufferCount() { return GetBuffersOwnedByManager().size(); }
223 
228  size_t GetArtEventCount() { return run_event_count_; }
229 
236  size_t GetFragmentCount(Fragment::sequence_id_t seqID, Fragment::type_t type = Fragment::InvalidFragmentType);
237 
244  size_t GetFragmentCountInBuffer(int buffer, Fragment::type_t type = Fragment::InvalidFragmentType);
245 
249  void RunArt(std::shared_ptr<art_config_file> config_file, std::shared_ptr<std::atomic<pid_t>> pid_out);
253  void StartArt();
254 
260  pid_t StartArtProcess(fhicl::ParameterSet pset);
261 
266  void ShutdownArtProcesses(std::set<pid_t>& pids);
267 
274  void ReconfigureArt(fhicl::ParameterSet art_pset, run_id_t newRun = 0, int n_art_processes = -1);
275 
285  bool endOfData();
286 
291  void startRun(run_id_t runID);
292 
297  run_id_t runID() const { return run_id_; }
298 
303  bool endRun();
304 
310  void rolloverSubrun(sequence_id_t boundary, subrun_id_t subrun);
311 
315  void rolloverSubrun();
316 
320  void sendMetrics();
321 
327  {
328  if (requests_) requests_->SetRequestMode(mode);
329  }
330 
335  void setOverwrite(bool overwrite) { overwrite_mode_ = overwrite; }
336 
340  void SetInitFragment(FragmentPtr frag);
341 
346  uint32_t GetBroadcastKey() { return broadcasts_.GetKey(); }
347 
352  RawDataType* GetDroppedDataAddress(Fragment::fragment_id_t frag) { return dropped_data_[frag]->dataBegin(); }
353 
363  void UpdateArtConfiguration(fhicl::ParameterSet art_pset);
364 
368  void CheckPendingBuffers();
369 
374  subrun_id_t GetSubrunForSequenceID(Fragment::sequence_id_t seqID);
375 
376  subrun_id_t GetCurrentSubrun() { return GetSubrunForSequenceID(Fragment::InvalidSequenceID); }
377 
378 private:
379  size_t get_art_process_count_()
380  {
381  std::unique_lock<std::mutex> lk(art_process_mutex_);
382  return art_processes_.size();
383  }
384 
385 private:
386  size_t num_art_processes_;
387  size_t const num_fragments_per_event_;
388  size_t const queue_size_;
389  run_id_t run_id_;
390 
391  std::map<sequence_id_t, subrun_id_t> subrun_event_map_;
392  size_t max_subrun_event_map_length_;
393  static std::mutex subrun_event_map_mutex_;
394 
395  std::set<int> active_buffers_;
396  std::set<int> pending_buffers_;
397  std::unordered_map<Fragment::sequence_id_t, size_t> released_incomplete_events_;
398 
399  bool update_run_ids_;
400  bool use_sequence_id_for_event_number_;
401  bool overwrite_mode_;
402  bool send_init_fragments_;
403  bool running_;
404 
405  std::unordered_map<int, std::atomic<int>> buffer_writes_pending_;
406  std::unordered_map<int, std::mutex> buffer_mutexes_;
407  static std::mutex sequence_id_mutex_;
408 
409  int incomplete_event_report_interval_ms_;
410  std::chrono::steady_clock::time_point last_incomplete_event_report_time_;
411  std::chrono::steady_clock::time_point last_shmem_buffer_metric_update_;
412  std::chrono::steady_clock::time_point last_backpressure_report_time_;
413  std::chrono::steady_clock::time_point last_fragment_header_write_time_;
414 
415 
416  struct MetricData
417  {
418  MetricData()
419  : event_count(0), event_size(0) {}
420  size_t event_count;
421  size_t event_size;
422  };
423  MetricData metric_data_;
424 
425  int broadcast_timeout_ms_;
426 
427  std::atomic<int> run_event_count_;
428  std::atomic<int> run_incomplete_event_count_;
429  std::atomic<int> subrun_event_count_;
430  std::atomic<int> subrun_incomplete_event_count_;
431  std::atomic<int> oversize_fragment_count_;
432  int maximum_oversize_fragment_count_;
433 
434  mutable std::mutex art_process_mutex_;
435  std::set<pid_t> art_processes_;
436  std::atomic<bool> restart_art_;
437  bool always_restart_art_;
438  std::atomic<bool> manual_art_;
439  fhicl::ParameterSet current_art_pset_;
440  std::shared_ptr<art_config_file> current_art_config_file_;
441  double minimum_art_lifetime_s_;
442  size_t art_event_processing_time_us_;
443 
444  std::unique_ptr<RequestSender> requests_;
445  fhicl::ParameterSet data_pset_;
446 
447  FragmentPtr init_fragment_;
448  std::unordered_map<Fragment::fragment_id_t, FragmentPtr> dropped_data_;
449 
450  bool broadcastFragment_(FragmentPtr frag, FragmentPtr& outFrag);
451 
452  detail::RawEventHeader* getEventHeader_(int buffer);
453 
454  int getBufferForSequenceID_(Fragment::sequence_id_t seqID, bool create_new, Fragment::timestamp_t timestamp = Fragment::InvalidTimestamp);
455  bool hasFragments_(int buffer);
456  void complete_buffer_(int buffer);
457  bool bufferComparator(int bufA, int bufB);
458  void check_pending_buffers_(std::unique_lock<std::mutex> const& lock);
459 
460  void send_init_frag_();
461  SharedMemoryManager broadcasts_;
462 };
463 } // namespace artdaq
464 
465 #endif //ARTDAQ_DAQRATE_SHAREDMEMORYEVENTMANAGER_HH
void RunArt(std::shared_ptr< art_config_file > config_file, std::shared_ptr< std::atomic< pid_t >> pid_out)
Run an art instance, recording the return codes and restarting it until the end flag is raised...
fhicl::Atom< size_t > event_queue_wait_time
&quot;event_queue_wait_time&quot; (Default: 5) : Amount of time(in seconds) an event can exist in shared memory...
fhicl::Atom< bool > overwrite_mode
&quot;overwite_mode&quot; (Default: false): Whether new data is allowed to overwrite buffers in the &quot;Full&quot; stat...
art_config_file wraps a temporary file used to configure art
void ShutdownArtProcesses(std::set< pid_t > &pids)
Shutdown a set of art processes.
fhicl::Atom< size_t > max_subrun_event_map_length
&quot;max_subrun_event_map_length&quot; (Default: 100): The maximum number of entries to store in the sequence ...
size_t GetLockedBufferCount()
Returns the number of buffers currently owned by this manager.
virtual ~SharedMemoryEventManager()
SharedMemoryEventManager Destructor.
The SharedMemoryEventManager is a SharedMemoryManger which tracks events as they are built...
Fragment::sequence_id_t sequence_id_t
Copy Fragment::sequence_id_t into local scope.
fhicl::Atom< size_t > broadcast_buffer_size
&quot;broadcast_buffer_size&quot; (Default: 0x100000): Size of the buffers in the broadcast shared memory segme...
fhicl::Atom< bool > restart_crashed_art_processes
&quot;restart_crashed_art_processes&quot; (Default: true) : Whether to automatically restart art processes that...
void ReconfigureArt(fhicl::ParameterSet art_pset, run_id_t newRun=0, int n_art_processes=-1)
Restart all art processes, using the given fhicl code to configure the new art processes.
pid_t StartArtProcess(fhicl::ParameterSet pset)
Start one art process.
fhicl::Atom< double > minimum_art_lifetime_s
&quot;minimum_art_lifetime_s&quot; (Default: 2 seconds): Amount of time that an art process should run to not b...
fhicl::Atom< int > maximum_oversize_fragment_count
&quot;maximum_oversize_fragment_count&quot; (Default: 1): Maximum number of over-size Fragments to drop before ...
fhicl::Atom< bool > update_run_ids_on_new_fragment
&quot;update_run_ids_on_new_fragment&quot; (Default: true) : Whether the run and subrun ID of an event should b...
RawDataType * WriteFragmentHeader(detail::RawFragmentHeader frag, bool dropIfNoBuffersAvailable=false)
Get a pointer to a reserved memory area for the given Fragment header.
void setRequestMode(detail::RequestMessageMode mode)
Set the RequestMessageMode for all outgoing data requests.
size_t GetArtEventCount()
Returns the number of events sent to art this run.
fhicl::Atom< size_t > art_analyzer_count
&quot;art_analyzer_count&quot; (Default: 1) : Number of art procceses to start
Configuration of the SharedMemoryEventManager. May be used for parameter validation ...
fhicl::Atom< bool > use_sequence_id_for_event_number
&quot;use_sequence_id_for_event_number&quot; (Default: true): Whether to use the artdaq Sequence ID (true) or t...
fhicl::Atom< uint32_t > broadcast_shared_memory_key
&quot;broadcast_shared_memory_key&quot; (Default: 0xCEE7000 + PID): Key to use for broadcast shared memory acce...
RawEvent::run_id_t run_id_t
Copy RawEvent::run_id_t into local scope.
size_t GetFragmentCount(Fragment::sequence_id_t seqID, Fragment::type_t type=Fragment::InvalidFragmentType)
Get the count of Fragments of a given type in an event.
void UpdateArtConfiguration(fhicl::ParameterSet art_pset)
Updates the internally-stored copy of the art configuration.
RawDataType * GetDroppedDataAddress(Fragment::fragment_id_t frag)
Gets the address of the &quot;dropped data&quot; fragment. Used for testing.
fhicl::Atom< size_t > max_subrun_lookup_table_size
&quot;max_subrun_lookup_table_size&quot; (Default: 100): Maximum number of entries in the subrun rollover histo...
size_t GetPendingEventCount()
Returns the number of events which are complete but waiting on lower sequenced events to finish...
void StartArt()
Start all the art processes.
run_id_t runID() const
Get the current Run number.
fhicl::Atom< size_t > buffer_count
&quot;buffer_count&quot; REQUIRED: Number of events in the Shared Memory(incomplete + pending art) ...
void SetInitFragment(FragmentPtr frag)
Set the stored Init fragment, if one has not yet been set already.
subrun_id_t GetSubrunForSequenceID(Fragment::sequence_id_t seqID)
Get the subrun number that the given Sequence ID would be assigned to.
fhicl::TableFragment< artdaq::RequestSender::Config > requestSenderConfig
Configuration of the RequestSender. See artdaq::RequestSender::Config.
fhicl::Atom< uint32_t > shared_memory_key
&quot;shared_memory_key&quot; (Default 0xBEE70000 + PID) : Key to use for shared memory access ...
std::map< sequence_id_t, RawEvent_ptr > EventMap
An EventMap is a map of RawEvent_ptr objects, keyed by sequence ID.
RequestMessageMode
Mode used to indicate current run conditions to the request receiver.
void rolloverSubrun()
Add a subrun transition immediately after the highest currently define sequence ID.
void sendMetrics()
Send metrics to the MetricManager, if one has been instantiated in the application.
fhicl::Atom< size_t > stale_buffer_timeout_usec
&quot;stale_buffer_timeout_usec&quot; (Default: event_queue_wait_time * 1, 000, 000) : Maximum amount of time e...
fhicl::Atom< size_t > expected_fragments_per_event
&quot;expected_fragments_per_event&quot; (REQUIRED) : Number of Fragments to expect per event ...
size_t GetIncompleteEventCount()
Returns the number of buffers which contain data but are not yet complete.
fhicl::Atom< bool > send_init_fragments
&quot;send_init_fragments&quot; (Default: true): Whether Init Fragments are expected to be sent to art...
SharedMemoryEventManager(fhicl::ParameterSet pset, fhicl::ParameterSet art_pset)
SharedMemoryEventManager Constructor.
bool endRun()
Send an EndOfRunFragment to the art thread.
void setOverwrite(bool overwrite)
Set the overwrite flag (non-reliable data transfer) for the Shared Memory.
fhicl::Atom< int > incomplete_event_report_interval_ms
&quot;incomplete_event_report_interval_ms&quot; (Default: -1): Interval at which an incomplete event report sho...
std::string getFileName() const
Get the path of the temporary file.
fhicl::Atom< bool > manual_art
&quot;manual_art&quot; (Default: false): Prints the startup command line for the art process so that the user m...
fhicl::Atom< bool > broadcast_mode
&quot;broadcast_mode&quot; (Default: false) : When true, buffers are not marked Empty when read, but return to Full state.Buffers are overwritten in order received.
fhicl::Atom< size_t > broadcast_buffer_count
&quot;broadcast_buffer_count&quot; (Default: 10): Buffers in the broadcast shared memory segment ...
void DoneWritingFragment(detail::RawFragmentHeader frag)
Used to indicate that the given Fragment is now completely in the buffer. Will check for buffer compl...
uint32_t GetBroadcastKey()
Gets the shared memory key of the broadcast SharedMemoryManager.
bool endOfData()
Indicate that the end of input has been reached to the art processes.
fhicl::Atom< bool > use_art
&quot;use_art&quot; (Default: true): Whether to start and manage art threads (Sets art_analyzer count to 0 and ...
RawEvent::subrun_id_t subrun_id_t
Copy RawEvent::subrun_id_t into local scope.
art_config_file(fhicl::ParameterSet ps)
art_config_file Constructor
void startRun(run_id_t runID)
Start a Run.
size_t GetFragmentCountInBuffer(int buffer, Fragment::type_t type=Fragment::InvalidFragmentType)
Get the count of Fragments of a given type in a buffer.
void CheckPendingBuffers()
Check for buffers which are ready to be marked incomplete and released to art and issue tokens for an...