diff --git a/docs/design.md b/docs/design.md index e5e6ff4..8e2189e 100644 --- a/docs/design.md +++ b/docs/design.md @@ -1,250 +1,244 @@ # Software Heritage Filesystem (SwhFS) --- Design notes The [Software Heritage](https://www.softwareheritage.org/) {ref}`data model ` is a [Direct Acyclic Graph](https://en.wikipedia.org/wiki/Directed_acyclic_graph) (DAG) with nodes of different types that correspond to source code artifacts such as directories, commits, etc. Using this [FUSE](https://en.wikipedia.org/wiki/Filesystem_in_Userspace) module (*SwhFS* for short) you can locally mount, and then navigate as a filesystem, parts of the archive identified by {ref}`Software Heritage identifiers ` (SWHIDs). To retrieve information about the source code artifacts, SwhFS interacts over the network with the Software Heritage archive via its {ref}`Web API `. ## Architecture SwhFS in context ([C4](https://en.wikipedia.org/wiki/C4_model) context diagram): ```{image} images/arch-context.svg :align: center ``` Main components of SwhFS (C4 container diagram): ```{image} images/arch-container.svg :align: center ``` ## Command-line interface $ swh fs mount [SWHID]... will mount the Software Heritage archive at the local ``, the *SwhFS mount point*. From there, the user will be able to lazily load and navigate the archive using SWHID at entry points. If one or more SWHIDs are also specified, the corresponding objects will be pre- fetched from the archive at mount-time and available at `/archive/`. For more details see the {ref}`CLI documentation `. ## Mount point The SwhFS mount point contain: -- `archive/`: initially empty, this directory is lazily populated with one -entry per accessed SWHID, having actual SWHIDs as names (possibly sharded into -`xy/../SWHID` paths to avoid overcrowding `archive/`). +- `archive/`: initially empty, this directory is lazily populated with one entry + per accessed SWHID, having actual SWHIDs as names (possibly sharded into + `xy/../SWHID` paths to avoid overcrowding `archive/`). - `meta/`: initially empty, this directory contains one `.json` file for -each `` entry under `archive/`. The JSON file contain all available meta -information about the given SWHID, as returned by the Software Heritage Web API -for that object. Note that, in case of pagination (e.g., snapshot objects with -many branches) the JSON file will contain a complete version with all pages -merged together. + each `` entry under `archive/`. The JSON file contain all available + meta information about the given SWHID, as returned by the Software Heritage + Web API for that object. Note that, in case of pagination (e.g., snapshot + objects with many branches) the JSON file will contain a complete version with + all pages merged together. -- `origin/`: initially empty, this directory is lazily populated with one -entry per accessed origin URL, having encoded URL as names. The URL encoding is -done using the percent-encoding mechanism described in RFC 3986. +- `origin/`: initially empty, this directory is lazily populated with one entry + per accessed origin URL, having encoded URL as names. The URL encoding is done + using the percent-encoding mechanism described in + [RFC 3986](https://tools.ietf.org/html/rfc3986.html). ## File system representation SWHID are represented differently on the file-system depending on the associated node types in the Software Heritage graph. Details are given below, for each node type. ### `cnt` nodes (blobs) Content leaves (AKA blobs) are represented on disks as regular files, containing the corresponding bytes, as archived. Note that permissions are associated to blobs only in the context of directories. Hence, when accessing blobs from the top-level `archive/` directory, the permissions of the `archive/SWHID` file will be arbitrary and not meaningful (e.g., `0x644`). ### `dir` nodes (directories) Directory nodes are represented as directories on the file-system, containing one entry for each entry of the archived directory. Entry names and other metadata, including permissions, will correspond to the archived entry metadata. Note that SwhFS is mounted read-only, no matter what the permissions say. So it is possible that, in the context of a directory, a file is presented as writable, whereas actually writing to it will fail with `EPERM`. ### `rev` nodes (commits) Revision (AKA commit) nodes are represented on the file-system as directories with the following entries: - `root`: source tree at the time of the commit, as a symlink pointing into -`archive/`, to a SWHID of type `dir` + `archive/`, to a SWHID of type `dir` - `parents/` (note the plural): a virtual directory containing entries named -`1`, `2`, `3`, etc., one for each parent commit. Each of these entry is a -symlink pointing into `archive/`, to the SWHID file for the given parent commit -- `parent` (note the singular): present if and only if the current commit has a -single parent commit (which is the most common case). When present it is a -symlink pointing into `archive/` to the SWHID for the sole parent commit -- `history`: a virtual directory containing all the parents commit until the -root commit. Entries are listed as symlinks with the SWHID as directory name, -pointing into `archive/SWHID`, and are returned in a topological ordering -similar to `git log` ordering. + `1`, `2`, `3`, etc., one for each parent commit. Each of these entry is a + symlink pointing into `archive/`, to the SWHID file for the given parent + commit +- `parent` (note the singular): present if and only if the current commit has at + least one parent commit (which is the most common case). When present it is a + symlink pointing into `parents/1/` +- `history`: a virtual directory listing all its revision ancestors, sorted in + reverse topological order. The history can be listed through `by-date/`, + `by-hash/` or `by-page/` with each its own sharding policy. - `meta.json`: metadata for the current node, as a symlink pointing to the -relevant `meta/.json` file + relevant `meta/.json` file ### `rel` nodes (releases) Release nodes are represented on the file-system as directories with the following entries: - `target`: target node, as a symlink to `archive/` -- `target_type`: type of the target SWHID, as a 3-letter code +- `target_type`: regular file containing the type of the target SWHID - `root`: present if and only if the release points to something that -(transitively) resolves to a directory. When present it is a symlink pointing -into `archive/` to the SWHID of the given directory + (transitively) resolves to a directory. When present it is a symlink pointing + into `archive/` to the SWHID of the given directory - `meta.json`: metadata for the current node, as a symlink pointing to the -relevant `meta/.json` file + relevant `meta/.json` file ### `snp` nodes (snapshots) Snapshot nodes are represented on the file-system as directories with one entry for each branch in the snapshot. Branch names are subject to URL encoding, in order to avoid problematic characters (e.g., `/` are replaced by `%2F`). Each entry is a symlink named as the branch name, URL encoded (to avoid problematic characters such as `/`, which becomes `%2F`). The symlink target points into `archive/` to the SWHID corresponding to the branch target. ### `ori` nodes (origins) Origin nodes are represented on the file-system as directories with one entry for each origin visit. The visits directories are named after the visit date (`YYYY-MM-DD`, if multiple visits occur the same day only the first one is kept). Each visit directory contains a `meta.json` with associated metadata for the origin node, and potentially a `snapshot` symlink pointing to the visit's snapshot node. ## Caching SwhFS retrieves both metadata and file contents from the Software Heritage archive via the network. In order to obtain reasonable performances several caches are used to minimize network transfer. Caches are stored on disk in SQLite DB(s) located under `$XDG_CACHE_HOME/swh/fuse/`. ```{todo} - potential improvement: store blobs larger than a threshold on disk as files rather than in SQLite, e.g., under `$XDG_CACHE_HOME/swh/fuse/objects/` ``` All caches are persistent (i.e., they survive the restart of the SwhFS process) and global (i.e., they are shared by concurrent SwhFS processes). We assume that no cache *invalidation* is necessary, due to intrinsic properties of the Software Heritage archive, such as integrity verification and append-only archive changes. To clean the caches one can just remove the corresponding files from disk. ### Metadata cache - SWHID → JSON metadata + Artifact id → JSON metadata -The metadata cache map each SWHID to the complete metadata of the referenced +The metadata cache map each artifact to the complete metadata of the referenced object. This is analogous to what is available in `meta/.json` file (and generally used as data source for returning the content of those files). +Artifacts are identified using their SWHIDs, or in the case of origin visits, +using their URLs. Cache location on-disk: `$XDG_CACHE_HOME/swh/fuse/metadata.sqlite` ### Blob cache cnt SWHID → bytes The blob cache map SWHIDs of type `cnt` to the bytes of their archived content. In general, each SWHID that has an entry in the blob cache also has a matching entry in the metadata cache for other blob attributes (e.g., checksums, size, etc.). The blob cache entry for a given content object is populated, at the latest, the first time the object is `open()`-d. It might be populated earlier on due to prefetching, e.g., when a directory pointing to the given content is listed for the first time. Cache location on-disk: `$XDG_CACHE_HOME/swh/fuse/blob.sqlite` -### Dentry cache - - dir SWHID → directory entries - -The dentry (directory entry) cache map SWHIDs of type `dir` to the directory -entries they contain. Each entry comes with its name as well as file attributes -(i.e., all its needed to perform a detailed directory listing). - -Additional attributes of each directory entry should be looked up on a entry by -entry basis, possibly hitting the metadata cache. - -The dentry cache for a given dir is populated, at the latest, when the content -of the directory is listed. More aggressive prefetching might happen. For -instance, when first opening a dir a recursive listing of it can be retrieved -from the remote backend and used to recursively populate the dentry cache for -all (transitive) sub-directories. - - -### Parents cache - - rev SWHID → parent SWHIDs - -The parents cache map SWHIDs of type `rev` to the list of their parent commits. - -The parents cache for a given rev is populated, at the latest, when the content -of the revision virtual directory is listed. More aggressive prefetching might -happen. For instance, when first opening a rev virtual directory a recursive -listing of all its ancestor can be retrieved from the remote backend and used to -recursively populate the parents cache for all ancestors. - - ### History cache rev SWHID → ancestor SWHIDs The history cache map SWHIDs of type `rev` to a list of `rev` SWHIDs corresponding to all its revision ancestors, sorted in reverse topological order. As the parents cache, the history cache is lazily populated and can be prefetched. To efficiently store the ancestor lists, the history cache represents ancestors as graph edges (a pair of two SWHID nodes), meaning the history cache is shared amongst all revisions parents. + +Cache location on-disk: `$XDG_CACHE_HOME/swh/fuse/metadata.sqlite` + + +### Direntry cache + + dir inode → directory entries + +The direntry cache map inode representing directories to the entries they +contain. Each entry comes with its name as well as file attributes (i.e., all +its needed to perform a detailed directory listing). + +Additional attributes of each directory entry should be looked up on a entry by +entry basis, possibly hitting the metadata cache. + +The direntry cache for a given dir is populated, at the latest, when the content +of the directory is listed. More aggressive prefetching might happen. For +instance, when first opening a dir a recursive listing of it can be retrieved +from the remote backend and used to recursively populate the direntry cache for +all (transitive) sub-directories. + +Cache location: in-memory. diff --git a/swh/fuse/cache.py b/swh/fuse/cache.py index f96db19..c4e0296 100644 --- a/swh/fuse/cache.py +++ b/swh/fuse/cache.py @@ -1,380 +1,380 @@ # Copyright (C) 2020 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information from abc import ABC from collections import OrderedDict from dataclasses import dataclass, field import json import logging from pathlib import Path import re import sys from typing import Any, AsyncGenerator, Dict, List, Optional, Tuple import aiosqlite import dateutil.parser from psutil import virtual_memory from swh.fuse.fs.artifact import RevisionHistoryShardByDate from swh.fuse.fs.entry import FuseDirEntry, FuseEntry from swh.fuse.fs.mountpoint import ArchiveDir, MetaDir, OriginDir from swh.model.exceptions import ValidationError from swh.model.identifiers import REVISION, SWHID, parse_swhid from swh.web.client.client import ORIGIN_VISIT, typify_json class FuseCache: """SwhFS retrieves both metadata and file contents from the Software Heritage archive via the network. In order to obtain reasonable performances several caches are used to minimize network transfer. Caches are stored on disk in SQLite databases located at `$XDG_CACHE_HOME/swh/fuse/`. All caches are persistent (i.e., they survive the restart of the SwhFS process) and global (i.e., they are shared by concurrent SwhFS processes). We assume that no cache *invalidation* is necessary, due to intrinsic properties of the Software Heritage archive, such as integrity verification and append-only archive changes. To clean the caches one can just remove the corresponding files from disk. """ def __init__(self, cache_conf: Dict[str, Any]): self.cache_conf = cache_conf async def __aenter__(self): # History and raw metadata share the same SQLite db self.metadata = MetadataCache(self.cache_conf["metadata"]) self.history = HistoryCache(self.cache_conf["metadata"]) self.blob = BlobCache(self.cache_conf["blob"]) self.direntry = DirEntryCache(self.cache_conf["direntry"]) await self.metadata.__aenter__() await self.blob.__aenter__() await self.history.__aenter__() return self async def __aexit__(self, type=None, val=None, tb=None) -> None: await self.metadata.__aexit__() await self.blob.__aexit__() await self.history.__aexit__() async def get_cached_swhids(self) -> AsyncGenerator[SWHID, None]: """ Return a list of all previously cached SWHID """ # Use the metadata db since it should always contain all accessed SWHIDs metadata_cursor = await self.metadata.conn.execute( "select swhid from metadata_cache" ) swhids = await metadata_cursor.fetchall() for raw_swhid in swhids: yield parse_swhid(raw_swhid[0]) async def get_cached_visits(self) -> AsyncGenerator[str, None]: """ Return a list of all previously cached visit URL """ cursor = await self.metadata.conn.execute("select url from visits_cache") urls = await cursor.fetchall() for raw_url in urls: yield raw_url[0] class AbstractCache(ABC): """ Abstract cache implementation to share common behavior between cache types (such as: YAML config parsing, SQLite context manager) """ def __init__(self, conf: Dict[str, Any]): self.conf = conf async def __aenter__(self): # In-memory (thus temporary) caching is useful for testing purposes if self.conf.get("in-memory", False): path = "file::memory:?cache=shared" uri = True else: path = Path(self.conf["path"]) path.parent.mkdir(parents=True, exist_ok=True) uri = False self.conn = await aiosqlite.connect(path, uri=uri) return self async def __aexit__(self, type=None, val=None, tb=None) -> None: await self.conn.close() class MetadataCache(AbstractCache): """ The metadata cache map each artifact to the complete metadata of the referenced object. This is analogous to what is available in `meta/.json` file (and generally used as data source for returning the content of those files). Artifacts are identified using their SWHIDs, or - in the case of origins visits using their URLs. """ + in the case of origin visits, using their URLs. """ DB_SCHEMA = """ create table if not exists metadata_cache ( swhid text, metadata blob, date text ); create index if not exists idx_metadata on metadata_cache(swhid); create table if not exists visits_cache ( url text, metadata blob ); create index if not exists idx_visits on visits_cache(url); """ async def __aenter__(self): await super().__aenter__() await self.conn.executescript(self.DB_SCHEMA) await self.conn.commit() return self async def get(self, swhid: SWHID, typify: bool = True) -> Any: cursor = await self.conn.execute( "select metadata from metadata_cache where swhid=?", (str(swhid),) ) cache = await cursor.fetchone() if cache: metadata = json.loads(cache[0]) return typify_json(metadata, swhid.object_type) if typify else metadata else: return None async def get_visits(self, url_encoded: str) -> Optional[List[Dict[str, Any]]]: cursor = await self.conn.execute( "select metadata from visits_cache where url=?", (url_encoded,) ) cache = await cursor.fetchone() if cache: visits = json.loads(cache[0]) visits_typed = [typify_json(v, ORIGIN_VISIT) for v in visits] return visits_typed else: return None async def set(self, swhid: SWHID, metadata: Any) -> None: # Fill in the date column for revisions (used as cache for history/by-date/) swhid_date = "" if swhid.object_type == REVISION: date = dateutil.parser.parse(metadata["date"]) swhid_date = RevisionHistoryShardByDate.DATE_FMT.format( year=date.year, month=date.month, day=date.day ) await self.conn.execute( "insert into metadata_cache values (?, ?, ?)", (str(swhid), json.dumps(metadata), swhid_date), ) await self.conn.commit() async def set_visits(self, url_encoded: str, visits: List[Dict[str, Any]]) -> None: await self.conn.execute( "insert into visits_cache values (?, ?)", (url_encoded, json.dumps(visits)), ) await self.conn.commit() class BlobCache(AbstractCache): """ The blob cache map SWHIDs of type `cnt` to the bytes of their archived content. The blob cache entry for a given content object is populated, at the latest, the first time the object is `read()`-d. It might be populated earlier on due to prefetching, e.g., when a directory pointing to the given content is listed for the first time. """ DB_SCHEMA = """ create table if not exists blob_cache ( swhid text, blob blob ); create index if not exists idx_blob on blob_cache(swhid); """ async def __aenter__(self): await super().__aenter__() await self.conn.executescript(self.DB_SCHEMA) await self.conn.commit() return self async def get(self, swhid: SWHID) -> Optional[bytes]: cursor = await self.conn.execute( "select blob from blob_cache where swhid=?", (str(swhid),) ) cache = await cursor.fetchone() if cache: blob = cache[0] return blob else: return None async def set(self, swhid: SWHID, blob: bytes) -> None: await self.conn.execute( "insert into blob_cache values (?, ?)", (str(swhid), blob) ) await self.conn.commit() class HistoryCache(AbstractCache): """ The history cache map SWHIDs of type `rev` to a list of `rev` SWHIDs corresponding to all its revision ancestors, sorted in reverse topological order. As the parents cache, the history cache is lazily populated and can be prefetched. To efficiently store the ancestor lists, the history cache represents ancestors as graph edges (a pair of two SWHID nodes), meaning the history cache is shared amongst all revisions parents. """ DB_SCHEMA = """ create table if not exists history_graph ( src text not null, dst text not null, unique(src, dst) ); create index if not exists idx_history on history_graph(src); """ async def __aenter__(self): await super().__aenter__() await self.conn.executescript(self.DB_SCHEMA) await self.conn.commit() return self HISTORY_REC_QUERY = """ with recursive dfs(node) AS ( values(?) union select history_graph.dst from history_graph join dfs on history_graph.src = dfs.node ) -- Do not keep the root node since it is not an ancestor select * from dfs limit -1 offset 1 """ async def get(self, swhid: SWHID) -> Optional[List[SWHID]]: cursor = await self.conn.execute(self.HISTORY_REC_QUERY, (str(swhid),),) cache = await cursor.fetchall() if not cache: return None history = [] for row in cache: parent = row[0] try: history.append(parse_swhid(parent)) except ValidationError: logging.warning("Cannot parse object from history cache: %s", parent) return history async def get_with_date_prefix( self, swhid: SWHID, date_prefix: str ) -> List[Tuple[SWHID, str]]: cursor = await self.conn.execute( f""" select swhid, date from ( {self.HISTORY_REC_QUERY} ) as history join metadata_cache on history.node = metadata_cache.swhid where metadata_cache.date like '{date_prefix}%' """, (str(swhid),), ) cache = await cursor.fetchall() if not cache: return [] history = [] for row in cache: parent, date = row[0], row[1] try: history.append((parse_swhid(parent), date)) except ValidationError: logging.warning("Cannot parse object from history cache: %s", parent) return history async def set(self, history: str) -> None: history = history.strip() if history: edges = [edge.split(" ") for edge in history.split("\n")] await self.conn.executemany( "insert or ignore into history_graph values (?, ?)", edges ) await self.conn.commit() class DirEntryCache: """ The direntry cache map inode representing directories to the entries they contain. Each entry comes with its name as well as file attributes (i.e., all its needed to perform a detailed directory listing). Additional attributes of each directory entry should be looked up on a entry by entry basis, possibly hitting other caches. The direntry cache for a given dir is populated, at the latest, when the content of the directory is listed. More aggressive prefetching might happen. For instance, when first opening a dir a recursive listing of it can be retrieved from the remote backend and used to recursively populate the direntry cache for all (transitive) sub-directories. """ @dataclass class LRU(OrderedDict): max_ram: int used_ram: int = field(init=False, default=0) def sizeof(self, value: Any) -> int: # Rough size estimate in bytes for a list of entries return len(value) * 1000 def __getitem__(self, key: Any) -> Any: value = super().__getitem__(key) self.move_to_end(key) return value def __setitem__(self, key: Any, value: Any) -> None: if key in self: self.move_to_end(key) else: self.used_ram += self.sizeof(value) super().__setitem__(key, value) while self.used_ram > self.max_ram and self: oldest = next(iter(self)) self.used_ram -= self.sizeof(oldest) del self[oldest] def __init__(self, conf: Dict[str, Any]): m = re.match(r"(\d+)\s*(.+)\s*", conf["maxram"]) if not m: logging.error("Cannot parse direntry maxram config: %s", conf["maxram"]) sys.exit(1) num = float(m.group(1)) unit = m.group(2).upper() if unit == "%": max_ram = int(num * virtual_memory().available / 100) else: units = {"B": 1, "KB": 10 ** 3, "MB": 10 ** 6, "GB": 10 ** 9} max_ram = int(float(num) * units[unit]) self.lru_cache = self.LRU(max_ram) def get(self, direntry: FuseDirEntry) -> Optional[List[FuseEntry]]: return self.lru_cache.get(direntry.inode, None) def set(self, direntry: FuseDirEntry, entries: List[FuseEntry]) -> None: if isinstance(direntry, (ArchiveDir, MetaDir, OriginDir)): # The `archive/`, `meta/`, and `origin/` are populated on the fly so # we should never cache them pass elif ( isinstance(direntry, RevisionHistoryShardByDate) and not direntry.is_status_done ): # The `by-date/' directory is populated in parallel so only cache it # once it has finished fetching all data from the API pass else: self.lru_cache[direntry.inode] = entries diff --git a/swh/fuse/fs/artifact.py b/swh/fuse/fs/artifact.py index c40fe12..54f75ac 100644 --- a/swh/fuse/fs/artifact.py +++ b/swh/fuse/fs/artifact.py @@ -1,561 +1,561 @@ # Copyright (C) 2020 The Software Heritage developers # See the AUTHORS file at the top-level directory of this distribution # License: GNU General Public License version 3, or any later version # See top-level LICENSE file for more information import asyncio from dataclasses import dataclass, field import json import logging from pathlib import Path from typing import Any, AsyncIterator, Dict, List import urllib.parse from swh.fuse.fs.entry import ( EntryMode, FuseDirEntry, FuseEntry, FuseFileEntry, FuseSymlinkEntry, ) from swh.model.from_disk import DentryPerms from swh.model.identifiers import CONTENT, DIRECTORY, RELEASE, REVISION, SNAPSHOT, SWHID @dataclass class Content(FuseFileEntry): """ Software Heritage content artifact. Attributes: swhid: Software Heritage persistent identifier prefetch: optional prefetched metadata used to set entry attributes Content leaves (AKA blobs) are represented on disks as regular files, containing the corresponding bytes, as archived. Note that permissions are associated to blobs only in the context of directories. Hence, when accessing blobs from the top-level `archive/` directory, the permissions of the `archive/SWHID` file will be arbitrary and not meaningful (e.g., `0x644`). """ swhid: SWHID prefetch: Any = None async def get_content(self) -> bytes: data = await self.fuse.get_blob(self.swhid) if not self.prefetch: self.prefetch = {"length": len(data)} return data async def size(self) -> int: if self.prefetch: return self.prefetch["length"] else: return await super().size() @dataclass class Directory(FuseDirEntry): """ Software Heritage directory artifact. Attributes: swhid: Software Heritage persistent identifier Directory nodes are represented as directories on the file-system, containing one entry for each entry of the archived directory. Entry names and other metadata, including permissions, will correspond to the archived entry metadata. Note that the FUSE mount is read-only, no matter what the permissions say. So it is possible that, in the context of a directory, a file is presented as writable, whereas actually writing to it will fail with `EPERM`. """ swhid: SWHID async def compute_entries(self) -> AsyncIterator[FuseEntry]: metadata = await self.fuse.get_metadata(self.swhid) for entry in metadata: name = entry["name"] swhid = entry["target"] mode = ( # Archived permissions for directories are always set to # 0o040000 so use a read-only permission instead int(EntryMode.RDONLY_DIR) if swhid.object_type == DIRECTORY else entry["perms"] ) # 1. Regular file if swhid.object_type == CONTENT: yield self.create_child( Content, name=name, mode=mode, swhid=swhid, # The directory API has extra info we can use to set # attributes without additional Software Heritage API call prefetch=entry, ) # 2. Regular directory elif swhid.object_type == DIRECTORY: yield self.create_child( Directory, name=name, mode=mode, swhid=swhid, ) # 3. Symlink elif mode == DentryPerms.symlink: yield self.create_child( FuseSymlinkEntry, name=name, # Symlink target is stored in the blob content target=await self.fuse.get_blob(swhid), ) # 4. Submodule elif swhid.object_type == REVISION: # Make sure the revision metadata is fetched and create a # symlink to distinguish it with regular directories await self.fuse.get_metadata(swhid) yield self.create_child( FuseSymlinkEntry, name=name, target=Path(self.get_relative_root_path(), f"archive/{swhid}"), ) else: raise ValueError("Unknown directory entry type: {swhid.object_type}") @dataclass class Revision(FuseDirEntry): """ Software Heritage revision artifact. Attributes: swhid: Software Heritage persistent identifier Revision (AKA commit) nodes are represented on the file-system as directories with the following entries: - `root`: source tree at the time of the commit, as a symlink pointing into `archive/`, to a SWHID of type `dir` - `parents/` (note the plural): a virtual directory containing entries named `1`, `2`, `3`, etc., one for each parent commit. Each of these entry is a symlink pointing into `archive/`, to the SWHID file for the given parent commit - `parent` (note the singular): present if and only if the current commit has at least one parent commit (which is the most common case). When present it is a symlink pointing into `parents/1/` - `history`: a virtual directory listing all its revision ancestors, sorted - in reverse topological order. Each entry is a symlink pointing into - `archive/SWHID`. + in reverse topological order. The history can be listed through + `by-date/`, `by-hash/` or `by-page/` with each its own sharding policy. - `meta.json`: metadata for the current node, as a symlink pointing to the relevant `meta/.json` file """ swhid: SWHID async def compute_entries(self) -> AsyncIterator[FuseEntry]: metadata = await self.fuse.get_metadata(self.swhid) directory = metadata["directory"] parents = metadata["parents"] root_path = self.get_relative_root_path() yield self.create_child( FuseSymlinkEntry, name="root", target=Path(root_path, f"archive/{directory}"), ) yield self.create_child( FuseSymlinkEntry, name="meta.json", target=Path(root_path, f"meta/{self.swhid}.json"), ) yield self.create_child( RevisionParents, name="parents", mode=int(EntryMode.RDONLY_DIR), parents=[x["id"] for x in parents], ) if len(parents) >= 1: yield self.create_child( FuseSymlinkEntry, name="parent", target="parents/1/", ) yield self.create_child( RevisionHistory, name="history", mode=int(EntryMode.RDONLY_DIR), swhid=self.swhid, ) @dataclass class RevisionParents(FuseDirEntry): """ Revision virtual `parents/` directory """ parents: List[SWHID] async def compute_entries(self) -> AsyncIterator[FuseEntry]: root_path = self.get_relative_root_path() for i, parent in enumerate(self.parents): yield self.create_child( FuseSymlinkEntry, name=str(i + 1), target=Path(root_path, f"archive/{parent}"), ) @dataclass class RevisionHistory(FuseDirEntry): """ Revision virtual `history/` directory """ swhid: SWHID async def prefill_caches(self) -> None: history = await self.fuse.get_history(self.swhid) for swhid in history: await self.fuse.get_metadata(swhid) async def compute_entries(self) -> AsyncIterator[FuseEntry]: # Run it concurrently because of the many API calls necessary asyncio.create_task(self.prefill_caches()) yield self.create_child( RevisionHistoryShardByDate, name="by-date", mode=int(EntryMode.RDONLY_DIR), history_swhid=self.swhid, ) yield self.create_child( RevisionHistoryShardByHash, name="by-hash", mode=int(EntryMode.RDONLY_DIR), history_swhid=self.swhid, ) yield self.create_child( RevisionHistoryShardByPage, name="by-page", mode=int(EntryMode.RDONLY_DIR), history_swhid=self.swhid, ) @dataclass class RevisionHistoryShardByDate(FuseDirEntry): """ Revision virtual `history/by-date` sharded directory """ history_swhid: SWHID prefix: str = field(default="") is_status_done: bool = field(default=False) DATE_FMT = "{year:04d}/{month:02d}/{day:02d}/" @dataclass class StatusFile(FuseFileEntry): """ Temporary file used to indicate loading progress in by-date/ """ name: str = field(init=False, default=".status") mode: int = field(init=False, default=int(EntryMode.RDONLY_FILE)) done: int todo: int async def get_content(self) -> bytes: fmt = f"Done: {self.done}/{self.todo}\n" return fmt.encode() async def compute_entries(self) -> AsyncIterator[FuseEntry]: history = await self.fuse.get_history(self.history_swhid) # Only check for cached revisions with the appropriate prefix, since # fetching all of them with the Web API would take too long swhids = await self.fuse.cache.history.get_with_date_prefix( self.history_swhid, date_prefix=self.prefix ) depth = self.prefix.count("/") root_path = self.get_relative_root_path() sharded_dirs = set() for (swhid, sharded_name) in swhids: if not sharded_name.startswith(self.prefix): continue if depth == 3: yield self.create_child( FuseSymlinkEntry, name=str(swhid), target=Path(root_path, f"archive/{swhid}"), ) # Create sharded directories else: next_prefix = sharded_name.split("/")[depth] if next_prefix not in sharded_dirs: sharded_dirs.add(next_prefix) yield self.create_child( RevisionHistoryShardByDate, name=next_prefix, mode=int(EntryMode.RDONLY_DIR), prefix=f"{self.prefix}{next_prefix}/", history_swhid=self.history_swhid, ) # TODO: store len(history) somewhere to avoid recompute? self.is_status_done = len(swhids) == len(history) if not self.is_status_done and depth == 0: yield self.create_child( RevisionHistoryShardByDate.StatusFile, done=len(swhids), todo=len(history), ) @dataclass class RevisionHistoryShardByHash(FuseDirEntry): """ Revision virtual `history/by-hash` sharded directory """ history_swhid: SWHID prefix: str = field(default="") SHARDING_LENGTH = 2 async def compute_entries(self) -> AsyncIterator[FuseEntry]: history = await self.fuse.get_history(self.history_swhid) if self.prefix: root_path = self.get_relative_root_path() for swhid in history: if swhid.object_id.startswith(self.prefix): yield self.create_child( FuseSymlinkEntry, name=str(swhid), target=Path(root_path, f"archive/{swhid}"), ) # Create sharded directories else: sharded_dirs = set() for swhid in history: next_prefix = swhid.object_id[: self.SHARDING_LENGTH] if next_prefix not in sharded_dirs: sharded_dirs.add(next_prefix) yield self.create_child( RevisionHistoryShardByHash, name=next_prefix, mode=int(EntryMode.RDONLY_DIR), prefix=next_prefix, history_swhid=self.history_swhid, ) @dataclass class RevisionHistoryShardByPage(FuseDirEntry): """ Revision virtual `history/by-page` sharded directory """ history_swhid: SWHID prefix: int = field(default=None) PAGE_SIZE = 10_000 PAGE_FMT = "{page_number:03d}" async def compute_entries(self) -> AsyncIterator[FuseEntry]: history = await self.fuse.get_history(self.history_swhid) if self.prefix is not None: current_page = self.prefix root_path = self.get_relative_root_path() max_idx = min(len(history), (current_page + 1) * self.PAGE_SIZE) for i in range(current_page * self.PAGE_SIZE, max_idx): swhid = history[i] yield self.create_child( FuseSymlinkEntry, name=str(swhid), target=Path(root_path, f"archive/{swhid}"), ) # Create sharded directories else: for i in range(0, len(history), self.PAGE_SIZE): page_number = i // self.PAGE_SIZE yield self.create_child( RevisionHistoryShardByPage, name=self.PAGE_FMT.format(page_number=page_number), mode=int(EntryMode.RDONLY_DIR), history_swhid=self.history_swhid, prefix=page_number, ) @dataclass class Release(FuseDirEntry): """ Software Heritage release artifact. Attributes: swhid: Software Heritage persistent identifier Release nodes are represented on the file-system as directories with the following entries: - `target`: target node, as a symlink to `archive/` - `target_type`: regular file containing the type of the target SWHID - `root`: present if and only if the release points to something that (transitively) resolves to a directory. When present it is a symlink pointing into `archive/` to the SWHID of the given directory - `meta.json`: metadata for the current node, as a symlink pointing to the relevant `meta/.json` file """ swhid: SWHID async def find_root_directory(self, swhid: SWHID) -> SWHID: if swhid.object_type == RELEASE: metadata = await self.fuse.get_metadata(swhid) return await self.find_root_directory(metadata["target"]) elif swhid.object_type == REVISION: metadata = await self.fuse.get_metadata(swhid) return metadata["directory"] elif swhid.object_type == DIRECTORY: return swhid else: return None async def compute_entries(self) -> AsyncIterator[FuseEntry]: metadata = await self.fuse.get_metadata(self.swhid) root_path = self.get_relative_root_path() yield self.create_child( FuseSymlinkEntry, name="meta.json", target=Path(root_path, f"meta/{self.swhid}.json"), ) target = metadata["target"] yield self.create_child( FuseSymlinkEntry, name="target", target=Path(root_path, f"archive/{target}") ) yield self.create_child( ReleaseType, name="target_type", mode=int(EntryMode.RDONLY_FILE), target_type=target.object_type, ) target_dir = await self.find_root_directory(target) if target_dir is not None: yield self.create_child( FuseSymlinkEntry, name="root", target=Path(root_path, f"archive/{target_dir}"), ) @dataclass class ReleaseType(FuseFileEntry): """ Release type virtual file """ target_type: str async def get_content(self) -> bytes: return str.encode(self.target_type + "\n") @dataclass class Snapshot(FuseDirEntry): """ Software Heritage snapshot artifact. Attributes: swhid: Software Heritage persistent identifier Snapshot nodes are represented on the file-system as directories with one entry for each branch in the snapshot. Each entry is a symlink pointing into `archive/` to the branch target SWHID. Branch names are URL encoded (hence '/' are replaced with '%2F'). """ swhid: SWHID async def compute_entries(self) -> AsyncIterator[FuseEntry]: metadata = await self.fuse.get_metadata(self.swhid) root_path = self.get_relative_root_path() for branch_name, branch_meta in metadata.items(): # Mangle branch name to create a valid UNIX filename name = urllib.parse.quote_plus(branch_name) yield self.create_child( FuseSymlinkEntry, name=name, target=Path(root_path, f"archive/{branch_meta['target']}"), ) @dataclass class Origin(FuseDirEntry): """ Software Heritage origin artifact. Origin nodes are represented on the file-system as directories with one entry for each origin visit. The visits directories are named after the visit date (`YYYY-MM-DD`, if multiple visits occur the same day only the first one is kept). Each visit directory contains a `meta.json` with associated metadata for the origin node, and potentially a `snapshot` symlink pointing to the visit's snapshot node. """ DATE_FMT = "{year:04d}-{month:02d}-{day:02d}" async def compute_entries(self) -> AsyncIterator[FuseEntry]: # The origin's name is always its URL (encoded to create a valid UNIX filename) visits = await self.fuse.get_visits(self.name) seen_date = set() for visit in visits: date = visit["date"] name = self.DATE_FMT.format(year=date.year, month=date.month, day=date.day) if name in seen_date: logging.debug( "Conflict date on origin: %s, %s", visit["origin"], str(name) ) else: seen_date.add(name) yield self.create_child( OriginVisit, name=name, mode=int(EntryMode.RDONLY_DIR), meta=visit, ) @dataclass class OriginVisit(FuseDirEntry): """ Origin visit virtual directory """ meta: Dict[str, Any] @dataclass class MetaFile(FuseFileEntry): content: str async def get_content(self) -> bytes: return str.encode(self.content + "\n") async def compute_entries(self) -> AsyncIterator[FuseEntry]: snapshot_swhid = self.meta["snapshot"] if snapshot_swhid: root_path = self.get_relative_root_path() yield self.create_child( FuseSymlinkEntry, name="snapshot", target=Path(root_path, f"archive/{snapshot_swhid}"), ) yield self.create_child( OriginVisit.MetaFile, name="meta.json", mode=int(EntryMode.RDONLY_FILE), content=json.dumps( self.meta, indent=self.fuse.conf["json-indent"], default=lambda x: str(x), ), ) OBJTYPE_GETTERS = { CONTENT: Content, DIRECTORY: Directory, REVISION: Revision, RELEASE: Release, SNAPSHOT: Snapshot, }