""" Utilities for rechunking arrays through p2p shuffles ==================================================== Tensors (or n-D arrays) in dask are split up across the workers as regular n-D "chunks" or bricks. These bricks are stacked up to form the global array. A key algorithm for these tensors is to "rechunk" them. That is to reassemble the same global representation using differently shaped n-D bricks. For example, to take an FFT of an n-D array, one uses a sequence of 1D FFTs along each axis. The implementation in dask (and indeed almost all distributed array frameworks) requires that 1D axis along which the FFT is taken is local to a single brick. So to perform the global FFT we need to arrange that each axis in turn is local to bricks. This can be achieved through all-to-all communication between the workers to exchange sub-pieces of their individual bricks, given a "rechunking" scheme. To perform the redistribution, each input brick is cut up into some number of smaller pieces, each of which contributes to one of the output bricks. The mapping from input brick to output bricks decomposes into the Cartesian product of axis-by-axis mappings. To see this, consider first a 1D example. Suppose our array is split up into three equally sized bricks:: |----0----|----1----|----2----| And the requested output chunks are:: |--A--|--B--|----C----|---D---| So brick 0 contributes to output bricks A and B; brick 1 contributes to B and C; and brick 2 contributes to C and D. Now consider a 2D example of the same problem:: +----0----+----1----+----2----+ | | | | α | | | | | | | +---------+---------+---------+ | | | | β | | | | | | | +---------+---------+---------+ | | | | γ | | | | | | | +---------+---------+---------+ Each brick can be described as the ordered pair of row and column 1D bricks, (0, α), (0, β), ..., (2, γ). Since the rechunking does not also reshape the array, axes do not "interfere" with one another when determining output bricks:: +--A--+--B--+----C----+---D---+ | | | | | Σ | | | | | | | | | +-----+ ----+---------+-------+ | | | | | | | | | | | | | | | Π | | | | | | | | | | | | | | | | | | | +-----+-----+---------+-------+ Consider the output (B, Σ) brick. This is contributed to by the input (0, α) and (1, α) bricks. Determination of the subslices is just done by slicing the the axes separately and combining them. The key thing to note here is that we never need to create, and store, the dense 2D mapping, we can instead construct it on the fly for each output brick in turn as necessary. The implementation here uses :func:`split_axes` to construct these 1D rechunkings. The output partitioning in :meth:`~.ArrayRechunkRun.add_partition` then lazily constructs the subsection of the Cartesian product it needs to determine the slices of the current input brick. This approach relies on the generic p2p buffering machinery to ensure that there are not too many small messages exchanged, since no special effort is made to minimise messages between workers when a worker might have two adjacent input bricks that are sliced into the same output brick. """ from __future__ import annotations import math import mmap import os from collections import defaultdict from collections.abc import ( Callable, Collection, Generator, Hashable, Iterable, Iterator, Mapping, Sequence, ) from concurrent.futures import ThreadPoolExecutor from dataclasses import dataclass from itertools import chain, product from pathlib import Path from typing import TYPE_CHECKING, Any, NamedTuple, cast import toolz from tornado.ioloop import IOLoop import dask import dask.config from dask._task_spec import Task, TaskRef, parse_input from dask.highlevelgraph import HighLevelGraph from dask.layers import Layer from dask.tokenize import tokenize from dask.typing import Key from dask.utils import parse_bytes from distributed.core import PooledRPCCall from distributed.metrics import context_meter from distributed.shuffle._core import ( NDIndex, P2PBarrierTask, ShuffleId, ShuffleRun, ShuffleSpec, barrier_key, get_worker_plugin, handle_transfer_errors, handle_unpack_errors, p2p_barrier, ) from distributed.shuffle._limiter import ResourceLimiter from distributed.shuffle._pickle import unpickle_bytestream from distributed.shuffle._worker_plugin import ShuffleWorkerPlugin from distributed.sizeof import sizeof if TYPE_CHECKING: import numpy as np from typing_extensions import TypeAlias import dask.array as da _T_LowLevelGraph: TypeAlias = dict[Key, tuple] ChunkedAxis: TypeAlias = tuple[float, ...] # chunks must either be an int or NaN ChunkedAxes: TypeAlias = tuple[ChunkedAxis, ...] NDSlice: TypeAlias = tuple[slice, ...] SlicedAxis: TypeAlias = tuple[slice, ...] SlicedAxes: TypeAlias = tuple[SlicedAxis, ...] def rechunk_transfer( input: np.ndarray, id: ShuffleId, input_chunk: NDIndex, ) -> int: with handle_transfer_errors(id): return get_worker_plugin().add_partition( input, partition_id=input_chunk, id=id, ) def rechunk_unpack( id: ShuffleId, output_chunk: NDIndex, barrier_run_id: int ) -> np.ndarray: with handle_unpack_errors(id): return get_worker_plugin().get_output_partition( id, barrier_run_id, output_chunk ) class _Partial(NamedTuple): """Information used to perform a partial rechunk along a single axis""" #: Slice of the old chunks along this axis that belong to the partial old: slice #: Slice of the new chunks along this axis that belong to the partial new: slice class _NDPartial(NamedTuple): """Information used to perform a partial rechunk along all axes""" #: n-dimensional slice of the old chunks along each axis that belong to the partial old: NDSlice #: n-dimensional slice of the new chunks along each axis that belong to the partial new: NDSlice ix: NDIndex def rechunk_name(token: str) -> str: return f"rechunk-p2p-{token}" def rechunk_p2p( x: da.Array, chunks: ChunkedAxes, *, threshold: int | None = None, block_size_limit: int | None = None, balance: bool = False, ) -> da.Array: import dask.array as da if x.size == 0: # Special case for empty array, as the algorithm below does not behave correctly return da.empty(x.shape, chunks=chunks, dtype=x.dtype) from dask.array.core import new_da_object prechunked = _calculate_prechunking(x.chunks, chunks, x.dtype, block_size_limit) if prechunked != x.chunks: x = cast( "da.Array", x.rechunk( chunks=prechunked, threshold=threshold, block_size_limit=block_size_limit, balance=balance, method="tasks", ), ) token = tokenize(x, chunks) name = rechunk_name(token) disk: bool = dask.config.get("distributed.p2p.storage.disk") layer = P2PRechunkLayer( name=name, token=token, chunks=chunks, chunks_input=x.chunks, name_input=x.name, disk=disk, ) return new_da_object( HighLevelGraph.from_collections(name, layer, [x]), name, chunks, meta=x._meta, dtype=x.dtype, ) class P2PRechunkLayer(Layer): name: str token: str chunks: ChunkedAxes chunks_input: ChunkedAxes name_input: str disk: bool keepmap: np.ndarray _cached_dict: _T_LowLevelGraph | None def __init__( self, name: str, token: str, chunks: ChunkedAxes, chunks_input: ChunkedAxes, name_input: str, disk: bool, keepmap: np.ndarray | None = None, annotations: Mapping[str, Any] | None = None, ): import numpy as np self.name = name self.token = token self.chunks = chunks self.chunks_input = chunks_input self.name_input = name_input self.disk = disk if keepmap is not None: self.keepmap = keepmap else: shape = tuple(len(axis) for axis in chunks) self.keepmap = np.ones(shape, dtype=bool) self._cached_dict = None super().__init__(annotations=annotations) def __repr__(self) -> str: return f"{type(self).__name__}" def get_output_keys(self) -> set[Key]: import numpy as np return { (self.name,) + nindex for nindex in np.ndindex(tuple(len(axis) for axis in self.chunks)) if self.keepmap[nindex] } def is_materialized(self) -> bool: return self._cached_dict is not None @property def _dict(self) -> _T_LowLevelGraph: """Materialize full dict representation""" dsk: _T_LowLevelGraph if self._cached_dict is not None: return self._cached_dict else: dsk = self._construct_graph() self._cached_dict = dsk return self._cached_dict def __getitem__(self, key: Key) -> tuple: return self._dict[key] def __iter__(self) -> Iterator[Key]: return iter(self._dict) def __len__(self) -> int: return len(self._dict) def _cull(self, keepmap: np.ndarray) -> P2PRechunkLayer: return P2PRechunkLayer( name=self.name, token=self.token, chunks=self.chunks, chunks_input=self.chunks_input, name_input=self.name_input, disk=self.disk, keepmap=keepmap, annotations=self.annotations, ) def _keys_to_indices(self, keys: Iterable[Key]) -> set[tuple[int, ...]]: """Simple utility to convert keys to chunk indices.""" chunks = set() for key in keys: if not isinstance(key, tuple) or len(key) < 2 or key[0] != self.name: continue chunk = cast(tuple[int, ...], key[1:]) assert all(isinstance(index, int) for index in chunk) chunks.add(chunk) return chunks def cull( self, keys: set[Key], all_keys: Collection[Key] ) -> tuple[P2PRechunkLayer, dict]: """Cull a P2PRechunkLayer HighLevelGraph layer. The underlying graph will only include the necessary tasks to produce the keys (indices) included in `keepmap`. Therefore, "culling" the layer only requires us to reset this parameter. """ import numpy as np from dask.array.rechunk import old_to_new keepmap = np.zeros_like(self.keepmap, dtype=bool) indices_to_keep = self._keys_to_indices(keys) _old_to_new = old_to_new(self.chunks_input, self.chunks) for ndindex in indices_to_keep: keepmap[ndindex] = True culled_deps = {} # Identify the individual partial rechunks for ndpartial in _split_partials(_old_to_new): # Cull partials for which we do not keep any output tasks if not np.any(keepmap[ndpartial.new]): continue # Within partials, we have all-to-all communication. # Thus, all output tasks share the same input tasks. deps = frozenset( (self.name_input,) + ndindex for ndindex in _ndindices_of_slice(ndpartial.old) ) for ndindex in _ndindices_of_slice(ndpartial.new): culled_deps[(self.name,) + ndindex] = deps if np.array_equal(keepmap, self.keepmap): return self, culled_deps else: culled_layer = self._cull(keepmap) return culled_layer, culled_deps def _construct_graph(self) -> _T_LowLevelGraph: import numpy as np from dask.array.rechunk import old_to_new dsk: _T_LowLevelGraph = {} _old_to_new = old_to_new(self.chunks_input, self.chunks) for ndpartial in _split_partials(_old_to_new): partial_keepmap = self.keepmap[ndpartial.new] output_count = np.sum(partial_keepmap) if output_count == 0: continue elif output_count == 1: # Single output chunk dsk.update( partial_concatenate( input_name=self.name_input, input_chunks=self.chunks_input, ndpartial=ndpartial, token=self.token, keepmap=self.keepmap, old_to_new=_old_to_new, ) ) else: dsk.update( partial_rechunk( input_name=self.name_input, input_chunks=self.chunks_input, chunks=self.chunks, ndpartial=ndpartial, token=self.token, disk=self.disk, keepmap=self.keepmap, ) ) return dsk def _calculate_prechunking( old_chunks: ChunkedAxes, new_chunks: ChunkedAxes, dtype: np.dtype, block_size_limit: int | None, ) -> ChunkedAxes: """Calculate how to perform the pre-rechunking step During the pre-rechunking step, we 1. Split input chunks along partial boundaries to make partials completely independent of one another 2. Merge small chunks within partials to reduce the number of transfer tasks and corresponding overhead """ split_axes = _split_chunks_along_partial_boundaries(old_chunks, new_chunks) # We can only determine how to concatenate chunks if we can calculate block sizes. has_nans = (any(math.isnan(y) for y in x) for x in old_chunks) if len(new_chunks) <= 1 or not all(new_chunks) or any(has_nans): return tuple(tuple(chain(*axis)) for axis in split_axes) if dtype is None or dtype.hasobject or dtype.itemsize == 0: return tuple(tuple(chain(*axis)) for axis in split_axes) # We made sure that there are no NaNs in split_axes above return _concatenate_small_chunks( split_axes, old_chunks, new_chunks, dtype, block_size_limit # type: ignore[arg-type] ) def _concatenate_small_chunks( split_axes: list[list[list[int]]], old_chunks: ChunkedAxes, new_chunks: ChunkedAxes, dtype: np.dtype, block_size_limit: int | None, ) -> ChunkedAxes: """Concatenate small chunks within partials. By concatenating chunks within partials, we reduce the number of P2P transfer tasks and their corresponding overhead. The algorithm used in this function is very similar to :func:`dask.array.rechunk.find_merge_rechunk`, the main difference is that we have to make sure only to merge chunks within partials. """ import numpy as np block_size_limit = block_size_limit or dask.config.get("array.chunk-size") if isinstance(block_size_limit, str): block_size_limit = parse_bytes(block_size_limit) # Make it a number of elements block_size_limit //= dtype.itemsize # We verified earlier that we do not have any NaNs largest_old_block = _largest_block_size(old_chunks) # type: ignore[arg-type] largest_new_block = _largest_block_size(new_chunks) # type: ignore[arg-type] block_size_limit = max([block_size_limit, largest_old_block, largest_new_block]) old_largest_width = [max(chain(*axis)) for axis in split_axes] new_largest_width = [max(c) for c in new_chunks] # This represents how much each dimension increases (>1) or reduces (<1) # the graph size during rechunking graph_size_effect = { dim: len(new_axis) / sum(map(len, split_axis)) for dim, (split_axis, new_axis) in enumerate(zip(split_axes, new_chunks)) } ndim = len(old_chunks) # This represents how much each dimension increases (>1) or reduces (<1) the # largest block size during rechunking block_size_effect = { dim: new_largest_width[dim] / (old_largest_width[dim] or 1) for dim in range(ndim) } # Our goal is to reduce the number of nodes in the rechunk graph # by concatenating some adjacent chunks, so consider dimensions where we can # reduce the # of chunks candidates = [dim for dim in range(ndim) if graph_size_effect[dim] <= 1.0] # Concatenating along each dimension reduces the graph size by a certain factor # and increases memory largest block size by a certain factor. # We want to optimize the graph size while staying below the given # block_size_limit. This is in effect a knapsack problem, except with # multiplicative values and weights. Just use a greedy algorithm # by trying dimensions in decreasing value / weight order. def key(k: int) -> float: gse = graph_size_effect[k] bse = block_size_effect[k] if bse == 1: bse = 1 + 1e-9 return (np.log(gse) / np.log(bse)) if bse > 0 else 0 sorted_candidates = sorted(candidates, key=key) concatenated_axes: list[list[int]] = [[] for i in range(ndim)] # Sim all the axes that are no candidates for i in range(ndim): if i in candidates: continue concatenated_axes[i] = list(chain(*split_axes[i])) # We want to concatenate chunks for axis_index in sorted_candidates: concatenated_axis = concatenated_axes[axis_index] multiplier = math.prod( old_largest_width[:axis_index] + old_largest_width[axis_index + 1 :] ) axis_limit = block_size_limit // multiplier for partial in split_axes[axis_index]: current = partial[0] for chunk in partial[1:]: if (current + chunk) > axis_limit: concatenated_axis.append(current) current = chunk else: current += chunk concatenated_axis.append(current) old_largest_width[axis_index] = max(concatenated_axis) return tuple(tuple(axis) for axis in concatenated_axes) def _split_chunks_along_partial_boundaries( old_chunks: ChunkedAxes, new_chunks: ChunkedAxes ) -> list[list[list[float]]]: """Split the old chunks along the boundaries of partials, i.e., groups of new chunks that share the same inputs. By splitting along the boundaries before rechunkin their input tasks become disjunct and each partial conceptually operates on an independent sub-array. """ from dask.array.rechunk import old_to_new _old_to_new = old_to_new(old_chunks, new_chunks) partials = _slice_new_chunks_into_partials(_old_to_new) split_axes = [] # Along each axis, we want to figure out how we have to split input chunks in order to make # partials disjunct. We then group the resulting input chunks per partial before returning. for axis_index, slices in enumerate(partials): old_to_new_axis = _old_to_new[axis_index] old_axis = old_chunks[axis_index] split_axis = [] partial_chunks = [] for slice_ in slices: first_new_chunk = slice_.start first_old_chunk, first_old_slice = old_to_new_axis[first_new_chunk][0] last_new_chunk = slice_.stop - 1 last_old_chunk, last_old_slice = old_to_new_axis[last_new_chunk][-1] first_chunk_size = old_axis[first_old_chunk] last_chunk_size = old_axis[last_old_chunk] if first_old_chunk == last_old_chunk: chunk_size = first_chunk_size if ( last_old_slice.stop is not None and last_old_slice.stop != last_chunk_size ): chunk_size = last_old_slice.stop if first_old_slice.start != 0: chunk_size -= first_old_slice.start partial_chunks.append(chunk_size) else: partial_chunks.append(first_chunk_size - first_old_slice.start) partial_chunks.extend(old_axis[first_old_chunk + 1 : last_old_chunk]) if last_old_slice.stop is not None: chunk_size = last_old_slice.stop else: chunk_size = last_chunk_size partial_chunks.append(chunk_size) split_axis.append(partial_chunks) partial_chunks = [] if partial_chunks: split_axis.append(partial_chunks) split_axes.append(split_axis) return split_axes def _largest_block_size(chunks: tuple[tuple[int, ...], ...]) -> int: return math.prod(map(max, chunks)) def _split_partials( old_to_new: list[Any], ) -> Generator[_NDPartial]: """Split the rechunking into partials that can be performed separately""" partials_per_axis = _split_partials_per_axis(old_to_new) indices_per_axis = (range(len(partials)) for partials in partials_per_axis) for nindex, partial_per_axis in zip( product(*indices_per_axis), product(*partials_per_axis) ): old, new = zip(*partial_per_axis) yield _NDPartial(old, new, nindex) def _split_partials_per_axis(old_to_new: list[Any]) -> tuple[tuple[_Partial, ...], ...]: """Split the rechunking into partials that can be performed separately on each axis""" sliced_axes = _slice_new_chunks_into_partials(old_to_new) partial_axes = [] for axis_index, slices in enumerate(sliced_axes): partials = [] for slice_ in slices: last_old_chunk: int first_old_chunk, _ = old_to_new[axis_index][slice_.start][0] last_old_chunk, _ = old_to_new[axis_index][slice_.stop - 1][-1] partials.append( _Partial( old=slice(first_old_chunk, last_old_chunk + 1), new=slice_, ) ) partial_axes.append(tuple(partials)) return tuple(partial_axes) def _slice_new_chunks_into_partials( old_to_new: list[list[list[tuple[int, slice]]]] ) -> SlicedAxes: """Slice the new chunks into partials that can be computed separately""" sliced_axes = [] chunk_shape = tuple(len(axis) for axis in old_to_new) for axis_index, old_to_new_axis in enumerate(old_to_new): # Two consecutive output chunks A and B belong to the same partial rechunk # if A and B share the same input chunks, i.e., separating A and B would not # allow us to cull more input tasks. # Index of the last input chunk of this partial rechunk first_old_chunk: int | None = None partial_splits = [0] recipe: list[tuple[int, slice]] for new_chunk_index, recipe in enumerate(old_to_new_axis): if len(recipe) == 0: continue current_first_old_chunk, _ = recipe[0] current_last_old_chunk, _ = recipe[-1] if first_old_chunk is None: first_old_chunk = current_first_old_chunk elif first_old_chunk != current_last_old_chunk: partial_splits.append(new_chunk_index) first_old_chunk = current_first_old_chunk partial_splits.append(chunk_shape[axis_index]) sliced_axes.append( tuple(slice(a, b) for a, b in toolz.sliding_window(2, partial_splits)) ) return tuple(sliced_axes) def _ndindices_of_slice(ndslice: NDSlice) -> Iterator[NDIndex]: return product(*(range(slc.start, slc.stop) for slc in ndslice)) def _partial_index(global_index: NDIndex, partial_offset: NDIndex) -> NDIndex: return tuple(index - offset for index, offset in zip(global_index, partial_offset)) def partial_concatenate( input_name: str, input_chunks: ChunkedAxes, ndpartial: _NDPartial, token: str, keepmap: np.ndarray, old_to_new: list[Any], ) -> dict[Key, Any]: import numpy as np from dask.array.chunk import getitem from dask.array.core import concatenate3 dsk: dict[Key, Any] = {} slice_group = f"rechunk-slice-{token}" partial_keepmap = keepmap[ndpartial.new] assert np.sum(partial_keepmap) == 1 partial_new_index = np.argwhere(partial_keepmap)[0] global_new_index = tuple( int(ix) + slc.start for ix, slc in zip(partial_new_index, ndpartial.new) ) inputs = tuple( old_to_new_axis[ix] for ix, old_to_new_axis in zip(global_new_index, old_to_new) ) shape = tuple(len(axis) for axis in inputs) rec_cat_arg = np.empty(shape, dtype="O") for old_partial_index in np.ndindex(shape): old_global_index, old_slice = zip( *(input_axis[index] for index, input_axis in zip(old_partial_index, inputs)) ) original_shape = tuple( old_axis[index] for index, old_axis in zip(old_global_index, input_chunks) ) if _slicing_is_necessary(old_slice, original_shape): key = (slice_group,) + ndpartial.ix + old_global_index dsk[key] = t = Task( key, getitem, TaskRef((input_name,) + old_global_index), old_slice, ) rec_cat_arg[old_partial_index] = t.ref() else: rec_cat_arg[old_partial_index] = TaskRef((input_name,) + old_global_index) concat_task = Task( (rechunk_name(token),) + global_new_index, concatenate3, parse_input(rec_cat_arg.tolist()), ) dsk[concat_task.key] = concat_task return dsk def _slicing_is_necessary(slice: NDSlice, shape: tuple[int | None, ...]) -> bool: """Return True if applying the slice alters the shape, False otherwise.""" return not all( slc.start == 0 and (size is None and slc.stop is None or slc.stop == size) for slc, size in zip(slice, shape) ) def partial_rechunk( input_name: str, input_chunks: ChunkedAxes, chunks: ChunkedAxes, ndpartial: _NDPartial, token: str, disk: bool, keepmap: np.ndarray, ) -> dict[Key, Any]: dsk: dict[Key, Any] = {} old_partial_offset = tuple(slice_.start for slice_ in ndpartial.old) partial_token = tokenize(token, ndpartial.ix) # Use `token` to generate a canonical group for the entire rechunk transfer_group = f"rechunk-transfer-{token}" unpack_group = rechunk_name(token) # We can use `partial_token` here because the barrier task share their # group across all P2P shuffle-like operations # FIXME: Make this group unique per individual P2P shuffle-like operation _barrier_key = barrier_key(ShuffleId(partial_token)) ndim = len(input_chunks) partial_old = tuple( chunk_axis[partial_axis] for partial_axis, chunk_axis in zip(ndpartial.old, input_chunks) ) partial_new: ChunkedAxes = tuple( chunks[axis_index][ndpartial.new[axis_index]] for axis_index in range(ndim) ) transfer_keys = [] for global_index in _ndindices_of_slice(ndpartial.old): partial_index = _partial_index(global_index, old_partial_offset) input_key = TaskRef((input_name,) + global_index) key = (transfer_group,) + ndpartial.ix + global_index dsk[key] = t = Task( key, rechunk_transfer, input_key, ShuffleId(partial_token), partial_index, ) transfer_keys.append(t.ref()) dsk[_barrier_key] = barrier = P2PBarrierTask( _barrier_key, p2p_barrier, partial_token, *transfer_keys, spec=ArrayRechunkSpec( id=ShuffleId(partial_token), new=partial_new, old=partial_old, disk=disk ), ) new_partial_offset = tuple(axis.start for axis in ndpartial.new) for global_index in _ndindices_of_slice(ndpartial.new): partial_index = _partial_index(global_index, new_partial_offset) if keepmap[global_index]: k = (unpack_group,) + global_index dsk[k] = Task( k, rechunk_unpack, ShuffleId(partial_token), partial_index, barrier.ref(), ) return dsk class Split(NamedTuple): """Slice of a chunk that is concatenated with other splits to create a new chunk Splits define how to slice an input chunk on a single axis into small pieces that can be concatenated together with splits from other input chunks to create output chunks of a rechunk operation. """ #: Index of the new output chunk to which this split belongs. chunk_index: int #: Index of the split within the list of splits that are concatenated #: to create the new chunk. split_index: int #: Slice of the input chunk. slice: slice SplitChunk: TypeAlias = list[Split] SplitAxis: TypeAlias = list[SplitChunk] SplitAxes: TypeAlias = list[SplitAxis] def split_axes(old: ChunkedAxes, new: ChunkedAxes) -> SplitAxes: """Calculate how to split the old chunks on each axis to create the new chunks Parameters ---------- old : ChunkedAxes Chunks along each axis of the old array new : ChunkedAxes Chunks along each axis of the new array Returns ------- SplitAxes Splits along each axis that determine how to slice the input chunks to create the new chunks by concatenating the resulting shards. """ from dask.array.rechunk import old_to_new _old_to_new = old_to_new(old, new) axes = [] for axis_index, new_axis in enumerate(_old_to_new): old_axis: SplitAxis = [[] for _ in old[axis_index]] for new_chunk_index, new_chunk in enumerate(new_axis): for split_index, (old_chunk_index, slice) in enumerate(new_chunk): old_axis[old_chunk_index].append( Split(new_chunk_index, split_index, slice) ) for old_chunk in old_axis: old_chunk.sort(key=lambda split: split.slice.start) axes.append(old_axis) return axes def convert_chunk(shards: list[list[tuple[NDIndex, np.ndarray]]]) -> np.ndarray: import numpy as np from dask.array.core import concatenate3 indexed: dict[NDIndex, np.ndarray] = {} for sublist in shards: for index, shard in sublist: indexed[index] = shard subshape = [max(dim) + 1 for dim in zip(*indexed.keys())] assert len(indexed) == np.prod(subshape) rec_cat_arg = np.empty(subshape, dtype="O") for index, shard in indexed.items(): rec_cat_arg[tuple(index)] = shard arrs = rec_cat_arg.tolist() # This may block for several seconds, as it physically reads the memory-mapped # buffers from disk return concatenate3(arrs) class ArrayRechunkRun(ShuffleRun[NDIndex, "np.ndarray"]): """State for a single active rechunk execution This object is responsible for splitting, sending, receiving and combining data shards. It is entirely agnostic to the distributed system and can perform a rechunk with other run instances using `rpc``. The user of this needs to guarantee that only `ArrayRechunkRun`s of the same unique `ShuffleID` and `run_id` interact. Parameters ---------- worker_for: A mapping partition_id -> worker_address. old: Existing chunking of the array per dimension. new: Desired chunking of the array per dimension. id: A unique `ShuffleID` this belongs to. run_id: A unique identifier of the specific execution of the shuffle this belongs to. span_id: Span identifier; see :doc:`spans` local_address: The local address this Shuffle can be contacted by using `rpc`. directory: The scratch directory to buffer data in. executor: Thread pool to use for offloading compute. rpc: A callable returning a PooledRPCCall to contact other Shuffle instances. Typically a ConnectionPool. digest_metric: A callable to ingest a performance metric. Typically Server.digest_metric. scheduler: A PooledRPCCall to contact the scheduler. memory_limiter_disk: memory_limiter_comm: A ``ResourceLimiter`` limiting the total amount of memory used in either buffer. """ def __init__( self, worker_for: dict[NDIndex, str], old: ChunkedAxes, new: ChunkedAxes, id: ShuffleId, run_id: int, span_id: str | None, local_address: str, directory: str, executor: ThreadPoolExecutor, rpc: Callable[[str], PooledRPCCall], digest_metric: Callable[[Hashable, float], None], scheduler: PooledRPCCall, memory_limiter_disk: ResourceLimiter, memory_limiter_comms: ResourceLimiter, disk: bool, loop: IOLoop, ): super().__init__( id=id, run_id=run_id, span_id=span_id, local_address=local_address, directory=directory, executor=executor, rpc=rpc, digest_metric=digest_metric, scheduler=scheduler, memory_limiter_comms=memory_limiter_comms, memory_limiter_disk=memory_limiter_disk, disk=disk, loop=loop, ) self.old = old self.new = new partitions_of = defaultdict(list) for part, addr in worker_for.items(): partitions_of[addr].append(part) self.partitions_of = dict(partitions_of) self.worker_for = worker_for self.split_axes = split_axes(old, new) async def _receive( self, data: list[tuple[NDIndex, list[tuple[NDIndex, tuple[NDIndex, np.ndarray]]]]], ) -> None: self.raise_if_closed() # Repartition shards and filter out already received ones shards = defaultdict(list) for d in data: id1, payload = d if id1 in self.received: continue self.received.add(id1) for id2, shard in payload: shards[id2].append(shard) self.total_recvd += sizeof(d) del data if not shards: return try: await self._write_to_disk(shards) except Exception as e: self._exception = e raise def _shard_partition( self, data: np.ndarray, partition_id: NDIndex ) -> dict[str, tuple[NDIndex, list[tuple[NDIndex, tuple[NDIndex, np.ndarray]]]]]: out: dict[str, list[tuple[NDIndex, tuple[NDIndex, np.ndarray]]]] = defaultdict( list ) shards_size = 0 shards_count = 0 ndsplits = product(*(axis[i] for axis, i in zip(self.split_axes, partition_id))) for ndsplit in ndsplits: chunk_index, shard_index, ndslice = zip(*ndsplit) shard = data[ndslice] # Don't wait until all shards have been transferred over the network # before data can be released if shard.base is not None: shard = shard.copy() shards_size += shard.nbytes shards_count += 1 out[self.worker_for[chunk_index]].append( (chunk_index, (shard_index, shard)) ) context_meter.digest_metric("p2p-shards", shards_size, "bytes") context_meter.digest_metric("p2p-shards", shards_count, "count") return {k: (partition_id, v) for k, v in out.items()} def _get_output_partition( self, partition_id: NDIndex, key: Key, **kwargs: Any ) -> np.ndarray: # Quickly read metadata from disk. # This is a bunch of seek()'s interleaved with short reads. data = self._read_from_disk(partition_id) # Copy the memory-mapped buffers from disk into memory. # This is where we'll spend most time. return convert_chunk(data) def deserialize(self, buffer: Any) -> Any: return buffer def read(self, path: Path) -> tuple[list[list[tuple[NDIndex, np.ndarray]]], int]: """Open a memory-mapped file descriptor to disk, read all metadata, and unpickle all arrays. This is a fast sequence of short reads interleaved with seeks. Do not read in memory the actual data; the arrays' buffers will point to the memory-mapped area. The file descriptor will be automatically closed by the kernel when all the returned arrays are dereferenced, which will happen after the call to concatenate3. """ with path.open(mode="r+b") as fh: buffer = memoryview(mmap.mmap(fh.fileno(), 0)) # The file descriptor has *not* been closed! shards = list(unpickle_bytestream(buffer)) return shards, buffer.nbytes def _get_assigned_worker(self, id: NDIndex) -> str: return self.worker_for[id] @dataclass(frozen=True) class ArrayRechunkSpec(ShuffleSpec[NDIndex]): new: ChunkedAxes old: ChunkedAxes @property def output_partitions(self) -> Generator[NDIndex]: yield from product(*(range(len(c)) for c in self.new)) def pick_worker(self, partition: NDIndex, workers: Sequence[str]) -> str: npartitions = 1 for c in self.new: npartitions *= len(c) ix = 0 for dim, pos in enumerate(partition): if dim > 0: ix += len(self.new[dim - 1]) * pos else: ix += pos i = len(workers) * ix // npartitions return workers[i] def create_run_on_worker( self, run_id: int, span_id: str | None, worker_for: dict[NDIndex, str], plugin: ShuffleWorkerPlugin, ) -> ShuffleRun: return ArrayRechunkRun( worker_for=worker_for, old=self.old, new=self.new, id=self.id, run_id=run_id, span_id=span_id, directory=os.path.join( plugin.worker.local_directory, f"shuffle-{self.id}-{run_id}", ), executor=plugin._executor, local_address=plugin.worker.address, rpc=plugin.worker.rpc, digest_metric=plugin.worker.digest_metric, scheduler=plugin.worker.scheduler, memory_limiter_disk=plugin.memory_limiter_disk, memory_limiter_comms=plugin.memory_limiter_comms, disk=self.disk, loop=plugin.worker.loop, )