##// END OF EJS Templates
exchange: improve computation of relevant markers for large repos...
exchange: improve computation of relevant markers for large repos Compute the candidate nodes with relevant markers directly from keys of the predecessors/successors/children dictionaries of obsstore. This is faster than iterating over all nodes directly. This test could be further improved for repositories with relative few markers compared to the repository size, but this is no longer hot already. With the current loop structure, the obshashrange use works as well as before as it passes lists with a single node. Adjust the interface by allowing revision lists as well as node lists. This helps cases that computes ancestors as it reduces the materialisation cost. Use this in _pushdiscoveryobsmarker and _getbundleobsmarkerpart. Improve the latter further by directly using ancestors(). Performance benchmarks show notable and welcome improvement to no-op push and pull (that would also apply to other push/pull). This apply to push and pull done without evolve. ### push/pull Benchmark parameter # bin-env-vars.hg.flavor = default # benchmark.variants.explicit-rev = none # benchmark.variants.protocol = ssh # benchmark.variants.revs = none ## benchmark.name = hg.command.pull # data-env-vars.name = mercurial-devel-2024-03-22-zstd-sparse-revlog before: 5.968537 seconds after: 5.668507 seconds (-5.03%, -0.30) # data-env-vars.name = tryton-devel-2024-03-22-zstd-sparse-revlog before: 1.446232 seconds after: 0.835553 seconds (-42.23%, -0.61) # data-env-vars.name = netbsd-src-draft-2024-09-19-zstd-sparse-revlog before: 5.777412 seconds after: 2.523454 seconds (-56.32%, -3.25) ## benchmark.name = hg.command.push # data-env-vars.name = mercurial-devel-2024-03-22-zstd-sparse-revlog before: 6.155501 seconds after: 5.885072 seconds (-4.39%, -0.27) # data-env-vars.name = tryton-devel-2024-03-22-zstd-sparse-revlog before: 1.491054 seconds after: 0.934882 seconds (-37.30%, -0.56) # data-env-vars.name = netbsd-src-draft-2024-09-19-zstd-sparse-revlog before: 5.902494 seconds after: 2.957644 seconds (-49.89%, -2.94) There is not notable different in these result using the "rust" flavor instead of the "default". The performance impact on the same operation when using evolve were also tested and no impact was noted.

File last commit:

r52756:f4733654 default
r52789:8583d138 default
Show More
v2.py
429 lines | 15.0 KiB | text/x-python | PythonLexer
# v2.py - Pure-Python implementation of the dirstate-v2 file format
#
# Copyright Mercurial Contributors
#
# This software may be used and distributed according to the terms of the
# GNU General Public License version 2 or any later version.
from __future__ import annotations
import struct
import typing
from ..thirdparty import attr
# Force pytype to use the non-vendored package
if typing.TYPE_CHECKING:
# noinspection PyPackageRequirements
import attr
from .. import error, policy
parsers = policy.importmod('parsers')
# Must match the constant of the same name in
# `rust/hg-core/src/dirstate_tree/on_disk.rs`
TREE_METADATA_SIZE = 44
NODE_SIZE = 44
# Must match the `TreeMetadata` Rust struct in
# `rust/hg-core/src/dirstate_tree/on_disk.rs`. See doc-comments there.
#
# * 4 bytes: start offset of root nodes
# * 4 bytes: number of root nodes
# * 4 bytes: total number of nodes in the tree that have an entry
# * 4 bytes: total number of nodes in the tree that have a copy source
# * 4 bytes: number of bytes in the data file that are not used anymore
# * 4 bytes: unused
# * 20 bytes: SHA-1 hash of ignore patterns
TREE_METADATA = struct.Struct('>LLLLL4s20s')
# Must match the `Node` Rust struct in
# `rust/hg-core/src/dirstate_tree/on_disk.rs`. See doc-comments there.
#
# * 4 bytes: start offset of full path
# * 2 bytes: length of the full path
# * 2 bytes: length within the full path before its "base name"
# * 4 bytes: start offset of the copy source if any, or zero for no copy source
# * 2 bytes: length of the copy source if any, or unused
# * 4 bytes: start offset of child nodes
# * 4 bytes: number of child nodes
# * 4 bytes: number of descendant nodes that have an entry
# * 4 bytes: number of descendant nodes that have a "tracked" state
# * 1 byte: flags
# * 4 bytes: expected size
# * 4 bytes: mtime seconds
# * 4 bytes: mtime nanoseconds
NODE = struct.Struct('>LHHLHLLLLHlll')
assert TREE_METADATA_SIZE == TREE_METADATA.size
assert NODE_SIZE == NODE.size
# match constant in mercurial/pure/parsers.py
DIRSTATE_V2_DIRECTORY = 1 << 13
def parse_dirstate(map, copy_map, data, tree_metadata):
"""parse a full v2-dirstate from a binary data into dictionaries:
- map: a {path: entry} mapping that will be filled
- copy_map: a {path: copy-source} mapping that will be filled
- data: a binary blob contains v2 nodes data
- tree_metadata:: a binary blob of the top level node (from the docket)
"""
(
root_nodes_start,
root_nodes_len,
_nodes_with_entry_count,
_nodes_with_copy_source_count,
_unreachable_bytes,
_unused,
_ignore_patterns_hash,
) = TREE_METADATA.unpack(tree_metadata)
parse_nodes(map, copy_map, data, root_nodes_start, root_nodes_len)
def parse_nodes(map, copy_map, data, start, len):
"""parse <len> nodes from <data> starting at offset <start>
This is used by parse_dirstate to recursively fill `map` and `copy_map`.
All directory specific information is ignored and do not need any
processing (DIRECTORY, ALL_UNKNOWN_RECORDED, ALL_IGNORED_RECORDED)
"""
for i in range(len):
node_start = start + NODE_SIZE * i
node_bytes = slice_with_len(data, node_start, NODE_SIZE)
(
path_start,
path_len,
_basename_start,
copy_source_start,
copy_source_len,
children_start,
children_count,
_descendants_with_entry_count,
_tracked_descendants_count,
flags,
size,
mtime_s,
mtime_ns,
) = NODE.unpack(node_bytes)
# Parse child nodes of this node recursively
parse_nodes(map, copy_map, data, children_start, children_count)
item = parsers.DirstateItem.from_v2_data(flags, size, mtime_s, mtime_ns)
if not item.any_tracked:
continue
path = slice_with_len(data, path_start, path_len)
map[path] = item
if copy_source_start:
copy_map[path] = slice_with_len(
data, copy_source_start, copy_source_len
)
def slice_with_len(data, start, len):
return data[start : start + len]
@attr.s
class Node:
path = attr.ib()
entry = attr.ib()
parent = attr.ib(default=None)
children_count = attr.ib(default=0)
children_offset = attr.ib(default=0)
descendants_with_entry = attr.ib(default=0)
tracked_descendants = attr.ib(default=0)
def pack(self, copy_map, paths_offset):
path = self.path
copy = copy_map.get(path)
entry = self.entry
path_start = paths_offset
path_len = len(path)
basename_start = path.rfind(b'/') + 1 # 0 if rfind returns -1
if copy is not None:
copy_source_start = paths_offset + len(path)
copy_source_len = len(copy)
else:
copy_source_start = 0
copy_source_len = 0
if entry is not None:
flags, size, mtime_s, mtime_ns = entry.v2_data()
else:
# There are no mtime-cached directories in the Python implementation
flags = DIRSTATE_V2_DIRECTORY
size = 0
mtime_s = 0
mtime_ns = 0
return NODE.pack(
path_start,
path_len,
basename_start,
copy_source_start,
copy_source_len,
self.children_offset,
self.children_count,
self.descendants_with_entry,
self.tracked_descendants,
flags,
size,
mtime_s,
mtime_ns,
)
def pack_dirstate(map, copy_map):
"""
Pack `map` and `copy_map` into the dirstate v2 binary format and return
the tuple of (data, metadata) bytearrays.
The on-disk format expects a tree-like structure where the leaves are
written first (and sorted per-directory), going up levels until the root
node and writing that one to the docket. See more details on the on-disk
format in `mercurial/helptext/internals/dirstate-v2`.
Since both `map` and `copy_map` are flat dicts we need to figure out the
hierarchy. This algorithm does so without having to build the entire tree
in-memory: it only keeps the minimum number of nodes around to satisfy the
format.
# Algorithm explanation
This explanation does not talk about the different counters for tracked
descendants and storing the copies, but that work is pretty simple once this
algorithm is in place.
## Building a subtree
First, sort `map`: this makes it so the leaves of the tree are contiguous
per directory (i.e. a/b/c and a/b/d will be next to each other in the list),
and enables us to use the ordering of folders to have a "cursor" of the
current folder we're in without ever going twice in the same branch of the
tree. The cursor is a node that remembers its parent and any information
relevant to the format (see the `Node` class), building the relevant part
of the tree lazily.
Then, for each file in `map`, move the cursor into the tree to the
corresponding folder of the file: for example, if the very first file
is "a/b/c", we start from `Node[""]`, create `Node["a"]` which points to
its parent `Node[""]`, then create `Node["a/b"]`, which points to its parent
`Node["a"]`. These nodes are kept around in a stack.
If the next file in `map` is in the same subtree ("a/b/d" or "a/b/e/f"), we
add it to the stack and keep looping with the same logic of creating the
tree nodes as needed. If however the next file in `map` is *not* in the same
subtree ("a/other", if we're still in the "a/b" folder), then we know that
the subtree we're in is complete.
## Writing the subtree
We have the entire subtree in the stack, so we start writing it to disk
folder by folder. The way we write a folder is to pop the stack into a list
until the folder changes, revert this list of direct children (to satisfy
the format requirement that children be sorted). This process repeats until
we hit the "other" subtree.
An example:
a
dir1/b
dir1/c
dir2/dir3/d
dir2/dir3/e
dir2/f
Would have us:
- add to the stack until "dir2/dir3/e"
- realize that "dir2/f" is in a different subtree
- pop "dir2/dir3/e", "dir2/dir3/d", reverse them so they're sorted and
pack them since the next entry is "dir2/dir3"
- go back up to "dir2"
- add "dir2/f" to the stack
- realize we're done with the map
- pop "dir2/f", "dir2/dir3" from the stack, reverse and pack them
- go up to the root node, do the same to write "a", "dir1" and "dir2" in
that order
## Special case for the root node
The root node is not serialized in the format, but its information is
written to the docket. Again, see more details on the on-disk format in
`mercurial/helptext/internals/dirstate-v2`.
"""
data = bytearray()
root_nodes_start = 0
root_nodes_len = 0
nodes_with_entry_count = 0
nodes_with_copy_source_count = 0
# Will always be 0 since this implementation always re-writes everything
# to disk
unreachable_bytes = 0
unused = b'\x00' * 4
# This is an optimization that's only useful for the Rust implementation
ignore_patterns_hash = b'\x00' * 20
if len(map) == 0:
tree_metadata = TREE_METADATA.pack(
root_nodes_start,
root_nodes_len,
nodes_with_entry_count,
nodes_with_copy_source_count,
unreachable_bytes,
unused,
ignore_patterns_hash,
)
return data, tree_metadata
sorted_map = sorted(map.items(), key=lambda x: x[0].split(b"/"))
# Use a stack to have to only remember the nodes we currently need
# instead of building the entire tree in memory
stack = []
current_node = Node(b"", None)
stack.append(current_node)
for index, (path, entry) in enumerate(sorted_map, 1):
nodes_with_entry_count += 1
if path in copy_map:
nodes_with_copy_source_count += 1
current_folder = get_folder(path)
current_node = move_to_correct_node_in_tree(
current_folder, current_node, stack
)
current_node.children_count += 1
# Entries from `map` are never `None`
if entry.tracked:
current_node.tracked_descendants += 1
current_node.descendants_with_entry += 1
stack.append(Node(path, entry, current_node))
should_pack = True
next_path = None
if index < len(sorted_map):
# Determine if the next entry is in the same sub-tree, if so don't
# pack yet
next_path = sorted_map[index][0]
should_pack = not is_ancestor(next_path, current_folder)
if should_pack:
pack_directory_children(current_node, copy_map, data, stack)
while stack and current_node.path != b"":
# Go up the tree and write until we reach the folder of the next
# entry (if any, otherwise the root)
parent = current_node.parent
in_ancestor_of_next_path = next_path is not None and (
is_ancestor(next_path, get_folder(stack[-1].path))
)
if parent is None or in_ancestor_of_next_path:
break
pack_directory_children(parent, copy_map, data, stack)
current_node = parent
# Special case for the root node since we don't write it to disk, only its
# children to the docket
current_node = stack.pop()
assert current_node.path == b"", current_node.path
assert len(stack) == 0, len(stack)
tree_metadata = TREE_METADATA.pack(
current_node.children_offset,
current_node.children_count,
nodes_with_entry_count,
nodes_with_copy_source_count,
unreachable_bytes,
unused,
ignore_patterns_hash,
)
return data, tree_metadata
def get_folder(path):
"""
Return the folder of the path that's given, an empty string for root paths.
"""
return path.rsplit(b'/', 1)[0] if b'/' in path else b''
def is_ancestor(path, maybe_ancestor):
"""Returns whether `maybe_ancestor` is an ancestor of `path`.
>>> is_ancestor(b"a", b"")
True
>>> is_ancestor(b"a/b/c", b"a/b/c")
False
>>> is_ancestor(b"hgext3rd/__init__.py", b"hgext")
False
>>> is_ancestor(b"hgext3rd/__init__.py", b"hgext3rd")
True
"""
if maybe_ancestor == b"":
return True
if path <= maybe_ancestor:
return False
path_components = path.split(b"/")
ancestor_components = maybe_ancestor.split(b"/")
return all(c == o for c, o in zip(path_components, ancestor_components))
def move_to_correct_node_in_tree(target_folder, current_node, stack):
"""
Move inside the dirstate node tree to the node corresponding to
`target_folder`, creating the missing nodes along the way if needed.
"""
while target_folder != current_node.path:
if is_ancestor(target_folder, current_node.path):
# We need to go down a folder
prefix = target_folder[len(current_node.path) :].lstrip(b'/')
subfolder_name = prefix.split(b'/', 1)[0]
if current_node.path:
subfolder_path = current_node.path + b'/' + subfolder_name
else:
subfolder_path = subfolder_name
next_node = stack[-1]
if next_node.path == target_folder:
# This folder is now a file and only contains removed entries
# merge with the last node
current_node = next_node
else:
current_node.children_count += 1
current_node = Node(subfolder_path, None, current_node)
stack.append(current_node)
else:
# We need to go up a folder
current_node = current_node.parent
return current_node
def pack_directory_children(node, copy_map, data, stack):
"""
Write the binary representation of the direct sorted children of `node` to
`data`
"""
direct_children = []
while stack[-1].path != b"" and get_folder(stack[-1].path) == node.path:
direct_children.append(stack.pop())
if not direct_children:
raise error.ProgrammingError(b"no direct children for %r" % node.path)
# Reverse the stack to get the correct sorted order
direct_children.reverse()
packed_children = bytearray()
# Write the paths to `data`. Pack child nodes but don't write them yet
for child in direct_children:
packed = child.pack(copy_map=copy_map, paths_offset=len(data))
packed_children.extend(packed)
data.extend(child.path)
data.extend(copy_map.get(child.path, b""))
node.tracked_descendants += child.tracked_descendants
node.descendants_with_entry += child.descendants_with_entry
# Write the fixed-size child nodes all together
node.children_offset = len(data)
data.extend(packed_children)