from __future__ import print_function, division, absolute_import import io import os import sys import copy import logging import functools from multiprocessing import Process, Pipe, Queue import multiprocessing.connection import traceback from xopen import xopen from . import seqio from .modifiers import ZeroCapper from .report import Statistics from .filters import (Redirector, PairedRedirector, NoFilter, PairedNoFilter, InfoFileWriter, RestFileWriter, WildcardFileWriter, TooShortReadFilter, TooLongReadFilter, NContentFilter, CasavaFilter, DiscardTrimmedFilter, DiscardUntrimmedFilter, Demultiplexer, PairedEndDemultiplexer) from .seqio import read_chunks_from_file, read_paired_chunks logger = logging.getLogger() class OutputFiles(object): """ The attributes are open file-like objects except when demultiplex is True. In that case, untrimmed, untrimmed2 are file names, and out and out2 are file name templates containing '{name}'. If interleaved is True, then out is written interleaved. Files may also be None. """ # TODO interleaving for the other file pairs (too_short, too_long, untrimmed)? def __init__( self, out=None, out2=None, untrimmed=None, untrimmed2=None, too_short=None, too_short2=None, too_long=None, too_long2=None, info=None, rest=None, wildcard=None, demultiplex=False, interleaved=False, ): self.out = out self.out2 = out2 self.untrimmed = untrimmed self.untrimmed2 = untrimmed2 self.too_short = too_short self.too_short2 = too_short2 self.too_long = too_long self.too_long2 = too_long2 self.info = info self.rest = rest self.wildcard = wildcard self.demultiplex = demultiplex self.interleaved = interleaved def __iter__(self): yield self.out yield self.out2 yield self.untrimmed yield self.untrimmed2 yield self.too_short yield self.too_short2 yield self.too_long yield self.too_long2 yield self.info yield self.rest yield self.wildcard class Pipeline(object): """ Processing pipeline that loops over reads and applies modifiers and filters """ should_warn_legacy = False n_adapters = 0 def __init__(self, ): self._close_files = [] self._reader = None self._filters = [] self._modifiers = [] self._colorspace = None self._outfiles = None self._demultiplexer = None # Filter settings self._minimum_length = None self._maximum_length = None self.max_n = None self.discard_casava = False self.discard_trimmed = False self.discard_untrimmed = False def set_input(self, file1, file2=None, qualfile=None, colorspace=False, fileformat=None, interleaved=False): self._reader = seqio.open(file1, file2, qualfile, colorspace, fileformat, interleaved, mode='r') self._colorspace = colorspace # Special treatment: Disable zero-capping if no qualities are available if not self._reader.delivers_qualities: self._modifiers = [m for m in self._modifiers if not isinstance(m, ZeroCapper)] def _open_writer(self, file, file2, **kwargs): # TODO backwards-incompatible change (?) would be to use outfiles.interleaved # for all outputs return seqio.open(file, file2, mode='w', qualities=self.uses_qualities, colorspace=self._colorspace, **kwargs) def set_output(self, outfiles): self._filters = [] self._outfiles = outfiles filter_wrapper = self._filter_wrapper() if outfiles.rest: self._filters.append(RestFileWriter(outfiles.rest)) if outfiles.info: self._filters.append(InfoFileWriter(outfiles.info)) if outfiles.wildcard: self._filters.append(WildcardFileWriter(outfiles.wildcard)) # minimum length and maximum length for lengths, file1, file2, filter_class in ( (self._minimum_length, outfiles.too_short, outfiles.too_short2, TooShortReadFilter), (self._maximum_length, outfiles.too_long, outfiles.too_long2, TooLongReadFilter) ): writer = None if lengths is not None: if file1: writer = self._open_writer(file1, file2) f1 = filter_class(lengths[0]) if lengths[0] is not None else None if len(lengths) == 2 and lengths[1] is not None: f2 = filter_class(lengths[1]) else: f2 = None self._filters.append(filter_wrapper(writer, filter=f1, filter2=f2)) if self.max_n is not None: f1 = f2 = NContentFilter(self.max_n) self._filters.append(filter_wrapper(None, f1, f2)) if self.discard_casava: f1 = f2 = CasavaFilter() self._filters.append(filter_wrapper(None, f1, f2)) if int(self.discard_trimmed) + int(self.discard_untrimmed) + int(outfiles.untrimmed is not None) > 1: raise ValueError('discard_trimmed, discard_untrimmed and outfiles.untrimmed must not ' 'be set simultaneously') if outfiles.demultiplex: self._demultiplexer = self._create_demultiplexer(outfiles) self._filters.append(self._demultiplexer) else: # Set up the remaining filters to deal with --discard-trimmed, # --discard-untrimmed and --untrimmed-output. These options # are mutually exclusive in order to avoid brain damage. if self.discard_trimmed: self._filters.append(filter_wrapper(None, DiscardTrimmedFilter(), DiscardTrimmedFilter())) elif self.discard_untrimmed: self._filters.append(filter_wrapper(None, DiscardUntrimmedFilter(), DiscardUntrimmedFilter())) elif outfiles.untrimmed: untrimmed_writer = self._open_writer(outfiles.untrimmed, outfiles.untrimmed2) self._filters.append(filter_wrapper(untrimmed_writer, DiscardUntrimmedFilter(), DiscardUntrimmedFilter())) self._filters.append(self._final_filter(outfiles)) def close(self): for f in self._outfiles: # TODO do not use hasattr if f is not None and f is not sys.stdin and f is not sys.stdout and hasattr(f, 'close'): f.close() if self._demultiplexer is not None: self._demultiplexer.close() @property def uses_qualities(self): return self._reader.delivers_qualities def run(self): (n, total1_bp, total2_bp) = self.process_reads() # TODO m = self._modifiers m2 = getattr(self, '_modifiers2', []) stats = Statistics() stats.collect(n, total1_bp, total2_bp, m, m2, self._filters) return stats def process_reads(self): raise NotImplementedError() def _filter_wrapper(self): raise NotImplementedError() def _final_filter(self, outfiles): raise NotImplementedError() def _create_demultiplexer(self, outfiles): raise NotImplementedError() class SingleEndPipeline(Pipeline): """ Processing pipeline for single-end reads """ paired = False def __init__(self): super(SingleEndPipeline, self).__init__() self._modifiers = [] def add(self, modifier): self._modifiers.append(modifier) def add1(self, modifier): """An alias for the add() function. Makes the interface similar to PairedEndPipeline""" self.add(modifier) def process_reads(self): """Run the pipeline. Return statistics""" n = 0 # no. of processed reads # TODO turn into attribute total_bp = 0 for read in self._reader: n += 1 total_bp += len(read.sequence) matches = [] for modifier in self._modifiers: read = modifier(read, matches) for filter_ in self._filters: if filter_(read, matches): break return (n, total_bp, None) def _filter_wrapper(self): return Redirector def _final_filter(self, outfiles): writer = self._open_writer(outfiles.out, outfiles.out2) return NoFilter(writer) def _create_demultiplexer(self, outfiles): return Demultiplexer(outfiles.out, outfiles.untrimmed, qualities=self.uses_qualities, colorspace=self._colorspace) @property def minimum_length(self): return self._minimum_length @minimum_length.setter def minimum_length(self, value): assert value is None or len(value) == 1 self._minimum_length = value @property def maximum_length(self): return self._maximum_length @maximum_length.setter def maximum_length(self, value): assert value is None or len(value) == 1 self._maximum_length = value class PairedEndPipeline(Pipeline): """ Processing pipeline for paired-end reads. """ def __init__(self, pair_filter_mode, modify_first_read_only=False): """Setting modify_first_read_only to True enables "legacy mode" """ super(PairedEndPipeline, self).__init__() self._modifiers2 = [] self._pair_filter_mode = pair_filter_mode self._modify_first_read_only = modify_first_read_only self._add_both_called = False self._should_warn_legacy = False self._reader = None def set_input(self, *args, **kwargs): super(PairedEndPipeline, self).set_input(*args, **kwargs) if not self._reader.delivers_qualities: self._modifiers2 = [m for m in self._modifiers2 if not isinstance(m, ZeroCapper)] def add(self, modifier): """ Add a modifier for R1 and R2. If modify_first_read_only is True, the modifier is not added for R2. """ self._modifiers.append(modifier) if not self._modify_first_read_only: modifier2 = copy.copy(modifier) self._modifiers2.append(modifier2) else: self._should_warn_legacy = True def add1(self, modifier): """Add a modifier for R1 only""" self._modifiers.append(modifier) def add2(self, modifier): """Add a modifier for R2 only""" assert not self._modify_first_read_only self._modifiers2.append(modifier) def process_reads(self): n = 0 # no. of processed reads total1_bp = 0 total2_bp = 0 for read1, read2 in self._reader: n += 1 total1_bp += len(read1.sequence) total2_bp += len(read2.sequence) matches1 = [] matches2 = [] for modifier in self._modifiers: read1 = modifier(read1, matches1) for modifier in self._modifiers2: read2 = modifier(read2, matches2) for filter in self._filters: # Stop writing as soon as one of the filters was successful. if filter(read1, read2, matches1, matches2): break return (n, total1_bp, total2_bp) @property def should_warn_legacy(self): return self._should_warn_legacy @should_warn_legacy.setter def should_warn_legacy(self, value): self._should_warn_legacy = bool(value) @property def paired(self): return 'first' if self._modify_first_read_only else 'both' def _filter_wrapper(self): return functools.partial(PairedRedirector, pair_filter_mode=self._pair_filter_mode) def _final_filter(self, outfiles): writer = self._open_writer(outfiles.out, outfiles.out2, interleaved=outfiles.interleaved) return PairedNoFilter(writer) def _create_demultiplexer(self, outfiles): return PairedEndDemultiplexer(outfiles.out, outfiles.out2, outfiles.untrimmed, outfiles.untrimmed2, qualities=self.uses_qualities, colorspace=self._colorspace) @property def minimum_length(self): return self._minimum_length @minimum_length.setter def minimum_length(self, value): assert value is None or len(value) == 2 self._minimum_length = value @property def maximum_length(self): return self._maximum_length @maximum_length.setter def maximum_length(self, value): assert value is None or len(value) == 2 self._maximum_length = value def reader_process(file, file2, connections, queue, buffer_size, stdin_fd): """ Read chunks of FASTA or FASTQ data from *file* and send to a worker. queue -- a Queue of worker indices. A worker writes its own index into this queue to notify the reader that it is ready to receive more data. connections -- a list of Connection objects, one for each worker. The function repeatedly - reads a chunk from the file - reads a worker index from the Queue - sends the chunk to connections[index] and finally sends "poison pills" (the value -1) to all connections. """ if stdin_fd != -1: sys.stdin.close() sys.stdin = os.fdopen(stdin_fd) try: with xopen(file, 'rb') as f: if file2: with xopen(file2, 'rb') as f2: for chunk_index, (chunk1, chunk2) in enumerate(read_paired_chunks(f, f2, buffer_size)): # Determine the worker that should get this chunk worker_index = queue.get() pipe = connections[worker_index] pipe.send(chunk_index) pipe.send_bytes(chunk1) pipe.send_bytes(chunk2) else: for chunk_index, chunk in enumerate(read_chunks_from_file(f, buffer_size)): # Determine the worker that should get this chunk worker_index = queue.get() pipe = connections[worker_index] pipe.send(chunk_index) pipe.send_bytes(chunk) # Send poison pills to all workers for _ in range(len(connections)): worker_index = queue.get() connections[worker_index].send(-1) except Exception as e: # TODO better send this to a common "something went wrong" Queue for worker_index in range(len(connections)): connections[worker_index].send(-2) connections[worker_index].send((e, traceback.format_exc())) class WorkerProcess(Process): """ The worker repeatedly reads chunks of data from the read_pipe, runs the pipeline on it and sends the processed chunks to the write_pipe. To notify the reader process that it wants data, it puts its own identifier into the need_work_queue before attempting to read data from the read_pipe. """ def __init__(self, id_, pipeline, input_path1, input_path2, interleaved_input, orig_outfiles, read_pipe, write_pipe, need_work_queue): super(WorkerProcess, self).__init__() self._id = id_ self._pipeline = pipeline self._input_path1 = input_path1 self._input_path2 = input_path2 self._interleaved_input = interleaved_input self._orig_outfiles = orig_outfiles self._read_pipe = read_pipe self._write_pipe = write_pipe self._need_work_queue = need_work_queue def run(self): try: stats = Statistics() while True: # Notify reader that we need data self._need_work_queue.put(self._id) chunk_index = self._read_pipe.recv() if chunk_index == -1: # reader is done break elif chunk_index == -2: # An exception has occurred in the reader e, tb_str = self._read_pipe.recv() logger.error('%s', tb_str) raise e # Setting the .buffer.name attributess below is necessary because # file format detection uses the file name data = self._read_pipe.recv_bytes() input = io.TextIOWrapper(io.BytesIO(data), encoding='ascii') input.buffer.name = self._input_path1 if self._input_path2: data = self._read_pipe.recv_bytes() input2 = io.TextIOWrapper(io.BytesIO(data), encoding='ascii') input2.buffer.name = self._input_path2 else: input2 = None output = io.TextIOWrapper(io.BytesIO(), encoding='ascii') output.buffer.name = self._orig_outfiles.out.name if self._orig_outfiles.out2 is not None: output2 = io.TextIOWrapper(io.BytesIO(), encoding='ascii') output2.buffer.name = self._orig_outfiles.out2.name else: output2 = None outfiles = OutputFiles(out=output, out2=output2, interleaved=self._orig_outfiles.interleaved) self._pipeline.set_input(input, input2, interleaved=self._interleaved_input) self._pipeline.set_output(outfiles) (n, bp1, bp2) = self._pipeline.process_reads() cur_stats = Statistics() cur_stats.collect(n, bp1, bp2, [], [], self._pipeline._filters) stats += cur_stats output.flush() processed_chunk = output.buffer.getvalue() self._write_pipe.send(chunk_index) self._write_pipe.send_bytes(processed_chunk) if self._orig_outfiles.out2 is not None: output2.flush() processed_chunk2 = output2.buffer.getvalue() self._write_pipe.send_bytes(processed_chunk2) m = self._pipeline._modifiers m2 = getattr(self._pipeline, '_modifiers2', []) modifier_stats = Statistics() modifier_stats.collect(0, 0, 0 if self._pipeline.paired else None, m, m2, []) stats += modifier_stats self._write_pipe.send(-1) self._write_pipe.send(stats) except Exception as e: self._write_pipe.send(-2) self._write_pipe.send((e, traceback.format_exc())) class OrderedChunkWriter(object): """ We may receive chunks of processed data from worker processes in any order. This class writes them to an output file in the correct order. """ def __init__(self, outfile): self._chunks = dict() self._current_index = 0 self._outfile = outfile def write(self, data, chunk_index): """ """ self._chunks[chunk_index] = data while self._current_index in self._chunks: # TODO 1) do not decode 2) use .buffer.write self._outfile.write(self._chunks[self._current_index].decode('utf-8')) del self._chunks[self._current_index] self._current_index += 1 def wrote_everything(self): return not self._chunks class ParallelPipelineRunner(object): """ Run a Pipeline in parallel - When set_input() is called, a reader process is spawned. - When run() is called, as many worker processes as requested are spawned. - In the main process, results are written to the output files in the correct order, and statistics are aggregated. If a worker needs work, it puts its own index into a Queue() (_need_work_queue). The reader process listens on this queue and sends the raw data to the worker that has requested work. For sending the data from reader to worker, a Connection() is used. There is one such connection for each worker (self._pipes). For sending the processed data from the worker to the main process, there is a second set of connections, again one for each worker. When the reader is finished, it sends 'poison pills' to all workers. When a worker receives this, it sends a poison pill to the main process, followed by a Statistics object that contains statistics about all the reads processed by that worker. """ def __init__(self, pipeline, n_workers, buffer_size=4*1024**2): self._pipeline = pipeline self._pipes = [] # the workers read from these self._reader_process = None self._outfiles = None self._input_path1 = None self._input_path2 = None self._interleaved_input = None self._n_workers = n_workers self._need_work_queue = Queue() self._buffer_size = buffer_size def set_input(self, file1, file2=None, qualfile=None, colorspace=False, fileformat=None, interleaved=False): if self._reader_process is not None: raise RuntimeError('Do not call set_input more than once') assert qualfile is None and colorspace is False and fileformat is None self._input_path1 = file1 if type(file1) is str else file1.name self._input_path2 = file2 if type(file2) is str or file2 is None else file2.name self._interleaved_input = interleaved connections = [Pipe(duplex=False) for _ in range(self._n_workers)] self._pipes, connw = zip(*connections) try: fileno = sys.stdin.fileno() except io.UnsupportedOperation: # This happens during tests: pytest sets sys.stdin to an object # that does not have a file descriptor. fileno = -1 self._reader_process = Process(target=reader_process, args=(file1, file2, connw, self._need_work_queue, self._buffer_size, fileno)) self._reader_process.daemon = True self._reader_process.start() @staticmethod def can_output_to(outfiles): return ( outfiles.out is not None and outfiles.rest is None and outfiles.info is None and outfiles.wildcard is None and outfiles.too_short is None and outfiles.too_short2 is None and outfiles.too_long is None and outfiles.too_long2 is None and outfiles.untrimmed is None and outfiles.untrimmed2 is None and not outfiles.demultiplex ) def set_output(self, outfiles): if not self.can_output_to(outfiles): raise ValueError() self._outfiles = outfiles def _start_workers(self): workers = [] connections = [] for index in range(self._n_workers): conn_r, conn_w = Pipe(duplex=False) connections.append(conn_r) worker = WorkerProcess( index, self._pipeline, self._input_path1, self._input_path2, self._interleaved_input, self._outfiles, self._pipes[index], conn_w, self._need_work_queue) worker.daemon = True worker.start() workers.append(worker) return workers, connections def run(self): workers, connections = self._start_workers() writers = [] for outfile in [self._outfiles.out, self._outfiles.out2]: if outfile is None: continue writers.append(OrderedChunkWriter(outfile)) stats = None while connections: ready_connections = multiprocessing.connection.wait(connections) for connection in ready_connections: chunk_index = connection.recv() if chunk_index == -1: # the worker is done cur_stats = connection.recv() if stats == -2: # An exception has occurred in the worker (see below, # this happens only when there is an exception sending # the statistics) e, tb_str = connection.recv() # TODO traceback should only be printed in development logger.error('%s', tb_str) raise e if stats is None: stats = cur_stats else: stats += cur_stats connections.remove(connection) continue elif chunk_index == -2: # An exception has occurred in the worker e, tb_str = connection.recv() logger.error('%s', tb_str) # We should use the worker's actual traceback object # here, but traceback objects are not picklable. raise e for writer in writers: data = connection.recv_bytes() writer.write(data, chunk_index) for writer in writers: assert writer.wrote_everything() for w in workers: w.join() self._reader_process.join() return stats def close(self): for f in self._outfiles: # TODO do not use hasattr if f is not None and f is not sys.stdin and f is not sys.stdout and hasattr(f, 'close'): f.close()