metaflow 2.12.39__py2.py3-none-any.whl → 2.13.1__py2.py3-none-any.whl

metaflow/plugins/pypi/conda_environment.py

@@ -5,21 +5,17 @@ import functools
 import io
 import json
 import os
-import sys
 import tarfile
-import time
-from concurrent.futures import ThreadPoolExecutor
+import threading
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from functools import wraps
 from hashlib import sha256
 from io import BufferedIOBase, BytesIO
-from itertools import chain
 from urllib.parse import unquote, urlparse
 
-import requests
-
 from metaflow.exception import MetaflowException
 from metaflow.metaflow_config import get_pinned_conda_libs
 from metaflow.metaflow_environment import MetaflowEnvironment
-from metaflow.metaflow_profile import profile
 
 from . import MAGIC_FILE, _datastore_packageroot
 from .utils import conda_platform
@@ -50,7 +46,6 @@ class CondaEnvironment(MetaflowEnvironment):
 
     def validate_environment(self, logger, datastore_type):
         self.datastore_type = datastore_type
-        self.logger = logger
 
         # Avoiding circular imports.
         from metaflow.plugins import DATASTORES
@@ -62,8 +57,21 @@ class CondaEnvironment(MetaflowEnvironment):
         from .micromamba import Micromamba
         from .pip import Pip
 
-        micromamba = Micromamba()
-        self.solvers = {"conda": micromamba, "pypi": Pip(micromamba)}
+        print_lock = threading.Lock()
+
+        def make_thread_safe(func):
+            @wraps(func)
+            def wrapper(*args, **kwargs):
+                with print_lock:
+                    return func(*args, **kwargs)
+
+            return wrapper
+
+        self.logger = make_thread_safe(logger)
+
+        # TODO: Wire up logging
+        micromamba = Micromamba(self.logger)
+        self.solvers = {"conda": micromamba, "pypi": Pip(micromamba, self.logger)}
 
     def init_environment(self, echo, only_steps=None):
         # The implementation optimizes for latency to ensure as many operations can
@@ -150,6 +158,9 @@ class CondaEnvironment(MetaflowEnvironment):
                 (
                     package["path"],
                     # Lazily fetch package from the interweb if needed.
+                    # TODO: Depending on the len_hint, the package might be downloaded from
+                    #       the interweb prematurely. save_bytes needs to be adjusted to handle
+                    #       this scenario.
                     LazyOpen(
                         package["local_path"],
                         "rb",
@@ -166,22 +177,60 @@ class CondaEnvironment(MetaflowEnvironment):
                 if id_ in dirty:
                     self.write_to_environment_manifest([id_, platform, type_], packages)
 
-        # First resolve environments through Conda, before PyPI.
+        storage = None
+        if self.datastore_type not in ["local"]:
+            # Initialize storage for caching if using a remote datastore
+            storage = self.datastore(_datastore_packageroot(self.datastore, echo))
+
         self.logger("Bootstrapping virtual environment(s) ...")
-        for solver in ["conda", "pypi"]:
-            with ThreadPoolExecutor() as executor:
-                results = list(
-                    executor.map(lambda x: solve(*x, solver), environments(solver))
-                )
-            _ = list(map(lambda x: self.solvers[solver].download(*x), results))
-            with ThreadPoolExecutor() as executor:
-                _ = list(
-                    executor.map(lambda x: self.solvers[solver].create(*x), results)
-                )
-            if self.datastore_type not in ["local"]:
-                # Cache packages only when a remote datastore is in play.
-                storage = self.datastore(_datastore_packageroot(self.datastore, echo))
-                cache(storage, results, solver)
+        # Sequence of operations:
+        #  1. Start all conda solves in parallel
+        #  2. Download conda packages sequentially
+        #  3. Create and cache conda environments in parallel
+        #  4. Start PyPI solves in parallel after each conda environment is created
+        #  5. Download PyPI packages sequentially
+        #  6. Create and cache PyPI environments in parallel
+
+        with ThreadPoolExecutor() as executor:
+            # Start all conda solves in parallel
+            conda_futures = [
+                executor.submit(lambda x: solve(*x, "conda"), env)
+                for env in environments("conda")
+            ]
+
+            pypi_envs = {env[0]: env for env in environments("pypi")}
+            pypi_futures = []
+
+            # Process conda results sequentially for downloads
+            for future in as_completed(conda_futures):
+                result = future.result()
+                # Sequential conda download
+                self.solvers["conda"].download(*result)
+                # Parallel conda create and cache
+                create_future = executor.submit(self.solvers["conda"].create, *result)
+                if storage:
+                    executor.submit(cache, storage, [result], "conda")
+
+                # Queue PyPI solve to start after conda create
+                if result[0] in pypi_envs:
+
+                    def pypi_solve(env):
+                        create_future.result()  # Wait for conda create
+                        return solve(*env, "pypi")
+
+                    pypi_futures.append(
+                        executor.submit(pypi_solve, pypi_envs[result[0]])
+                    )
+
+            # Process PyPI results sequentially for downloads
+            for solve_future in pypi_futures:
+                result = solve_future.result()
+                # Sequential PyPI download
+                self.solvers["pypi"].download(*result)
+                # Parallel PyPI create and cache
+                executor.submit(self.solvers["pypi"].create, *result)
+                if storage:
+                    executor.submit(cache, storage, [result], "pypi")
         self.logger("Virtual environment(s) bootstrapped!")
 
     def executable(self, step_name, default=None):
@@ -382,7 +431,8 @@ class CondaEnvironment(MetaflowEnvironment):
                 'DISABLE_TRACING=True python -m metaflow.plugins.pypi.bootstrap "%s" %s "%s" linux-64'
                 % (self.flow.name, id_, self.datastore_type),
                 "echo 'Environment bootstrapped.'",
-                "export PATH=$PATH:$(pwd)/micromamba",
+                # To avoid having to install micromamba in the PATH in micromamba.py, we add it to the PATH here.
+                "export PATH=$PATH:$(pwd)/micromamba/bin",
             ]
         else:
             # for @conda/@pypi(disabled=True).
@@ -443,6 +493,7 @@ class LazyOpen(BufferedIOBase):
         self._file = None
         self._buffer = None
         self._position = 0
+        self.requests = None
 
     def _ensure_file(self):
         if not self._file:
@@ -459,8 +510,13 @@ class LazyOpen(BufferedIOBase):
                 raise ValueError("Both filename and url are missing")
 
     def _download_to_buffer(self):
+        if self.requests is None:
+            # TODO: Remove dependency on requests
+            import requests
+
+            self.requests = requests
         # TODO: Stream it in chunks?
-        response = requests.get(self.url, stream=True)
+        response = self.requests.get(self.url, stream=True)
         response.raise_for_status()
         return response.content
 

metaflow/plugins/pypi/micromamba.py

@@ -1,12 +1,14 @@
+import functools
 import json
 import os
 import subprocess
 import tempfile
+import time
 
 from metaflow.exception import MetaflowException
 from metaflow.util import which
 
-from .utils import conda_platform
+from .utils import MICROMAMBA_MIRROR_URL, MICROMAMBA_URL, conda_platform
 
 
 class MicromambaException(MetaflowException):
@@ -19,8 +21,11 @@ class MicromambaException(MetaflowException):
         super(MicromambaException, self).__init__(msg)
 
 
+GLIBC_VERSION = os.environ.get("CONDA_OVERRIDE_GLIBC", "2.38")
+
+
 class Micromamba(object):
-    def __init__(self):
+    def __init__(self, logger=None):
         # micromamba is a tiny version of the mamba package manager and comes with
         # metaflow specific performance enhancements.
 
@@ -33,6 +38,12 @@ class Micromamba(object):
             os.path.expanduser(_home),
             "micromamba",
         )
+
+        if logger:
+            self.logger = logger
+        else:
+            self.logger = lambda *args, **kwargs: None  # No-op logger if not provided
+
         self.bin = (
             which(os.environ.get("METAFLOW_PATH_TO_MICROMAMBA") or "micromamba")
             or which("./micromamba")  # to support remote execution
@@ -70,6 +81,9 @@ class Micromamba(object):
                 "MAMBA_ADD_PIP_AS_PYTHON_DEPENDENCY": "true",
                 "CONDA_SUBDIR": platform,
                 # "CONDA_UNSATISFIABLE_HINTS_CHECK_DEPTH": "0" # https://github.com/conda/conda/issues/9862
+                # Add a default glibc version for linux-64 environments (ignored for other platforms)
+                # TODO: Make the version configurable
+                "CONDA_OVERRIDE_GLIBC": GLIBC_VERSION,
             }
             cmd = [
                 "create",
@@ -78,6 +92,7 @@ class Micromamba(object):
                 "--dry-run",
                 "--no-extra-safety-checks",
                 "--repodata-ttl=86400",
+                "--safety-checks=disabled",
                 "--retry-clean-cache",
                 "--prefix=%s/prefix" % tmp_dir,
             ]
@@ -91,10 +106,11 @@ class Micromamba(object):
                 cmd.append("python==%s" % python)
             # TODO: Ensure a human readable message is returned when the environment
             #       can't be resolved for any and all reasons.
-            return [
+            solved_packages = [
                 {k: v for k, v in item.items() if k in ["url"]}
                 for item in self._call(cmd, env)["actions"]["LINK"]
             ]
+            return solved_packages
 
     def download(self, id_, packages, python, platform):
         # Unfortunately all the packages need to be catalogued in package cache
@@ -103,8 +119,6 @@ class Micromamba(object):
         # Micromamba is painfully slow in determining if many packages are infact
         # already cached. As a perf heuristic, we check if the environment already
         # exists to short circuit package downloads.
-        if self.path_to_environment(id_, platform):
-            return
 
         prefix = "{env_dirs}/{keyword}/{platform}/{id}".format(
             env_dirs=self.info()["envs_dirs"][0],
@@ -113,13 +127,18 @@ class Micromamba(object):
             id=id_,
         )
 
-        # Another forced perf heuristic to skip cross-platform downloads.
+        # cheap check
         if os.path.exists(f"{prefix}/fake.done"):
             return
 
+        # somewhat expensive check
+        if self.path_to_environment(id_, platform):
+            return
+
         with tempfile.TemporaryDirectory() as tmp_dir:
             env = {
                 "CONDA_SUBDIR": platform,
+                "CONDA_OVERRIDE_GLIBC": GLIBC_VERSION,
             }
             cmd = [
                 "create",
@@ -159,6 +178,7 @@ class Micromamba(object):
             # use hardlinks when possible, otherwise copy files
             # disabled for now since it adds to environment creation latencies
             "CONDA_ALLOW_SOFTLINKS": "0",
+            "CONDA_OVERRIDE_GLIBC": GLIBC_VERSION,
         }
         cmd = [
             "create",
@@ -174,6 +194,7 @@ class Micromamba(object):
             cmd.append("{url}".format(**package))
         self._call(cmd, env)
 
+    @functools.lru_cache(maxsize=None)
     def info(self):
         return self._call(["config", "list", "-a"])
 
@@ -198,18 +219,24 @@ class Micromamba(object):
         }
         directories = self.info()["pkgs_dirs"]
         # search all package caches for packages
-        metadata = {
-            url: os.path.join(d, file)
+
+        file_to_path = {}
+        for d in directories:
+            if os.path.isdir(d):
+                try:
+                    with os.scandir(d) as entries:
+                        for entry in entries:
+                            if entry.is_file():
+                                # Prefer the first occurrence if the file exists in multiple directories
+                                file_to_path.setdefault(entry.name, entry.path)
+                except OSError:
+                    continue
+        ret = {
+            # set package tarball local paths to None if package tarballs are missing
+            url: file_to_path.get(file)
             for url, file in packages_to_filenames.items()
-            for d in directories
-            if os.path.isdir(d)
-            and file in os.listdir(d)
-            and os.path.isfile(os.path.join(d, file))
         }
-        # set package tarball local paths to None if package tarballs are missing
-        for url in packages_to_filenames:
-            metadata.setdefault(url, None)
-        return metadata
+        return ret
 
     def interpreter(self, id_):
         return os.path.join(self.path_to_environment(id_), "bin/python")
@@ -296,7 +323,7 @@ class Micromamba(object):
                         stderr="\n".join(err),
                     )
                 )
-            except (TypeError, ValueError) as ve:
+            except (TypeError, ValueError):
                 pass
             raise MicromambaException(
                 msg.format(
@@ -312,23 +339,37 @@ def _install_micromamba(installation_location):
     # Unfortunately no 32bit binaries are available for micromamba, which ideally
     # shouldn't be much of a problem in today's world.
     platform = conda_platform()
-    try:
-        subprocess.Popen(f"mkdir -p {installation_location}", shell=True).wait()
-        # https://mamba.readthedocs.io/en/latest/micromamba-installation.html#manual-installation
-        # requires bzip2
-        result = subprocess.Popen(
-            f"curl -Ls https://micro.mamba.pm/api/micromamba/{platform}/1.5.7 | tar -xvj -C {installation_location} bin/micromamba",
-            shell=True,
-            stderr=subprocess.PIPE,
-            stdout=subprocess.PIPE,
-        )
-        _, err = result.communicate()
-        if result.returncode != 0:
-            raise MicromambaException(
-                f"Micromamba installation '{result.args}' failed:\n{err.decode()}"
-            )
+    url = MICROMAMBA_URL.format(platform=platform, version="1.5.7")
+    mirror_url = MICROMAMBA_MIRROR_URL.format(platform=platform, version="1.5.7")
+    os.makedirs(installation_location, exist_ok=True)
 
-    except subprocess.CalledProcessError as e:
-        raise MicromambaException(
-            "Micromamba installation failed:\n{}".format(e.stderr.decode())
-        )
+    def _download_and_extract(url):
+        max_retries = 3
+        for attempt in range(max_retries):
+            try:
+                # https://mamba.readthedocs.io/en/latest/micromamba-installation.html#manual-installation
+                # requires bzip2
+                result = subprocess.Popen(
+                    f"curl -Ls {url} | tar -xvj -C {installation_location} bin/micromamba",
+                    shell=True,
+                    stderr=subprocess.PIPE,
+                    stdout=subprocess.PIPE,
+                )
+                _, err = result.communicate()
+                if result.returncode != 0:
+                    raise MicromambaException(
+                        f"Micromamba installation '{result.args}' failed:\n{err.decode()}"
+                    )
+            except subprocess.CalledProcessError as e:
+                if attempt == max_retries - 1:
+                    raise MicromambaException(
+                        "Micromamba installation failed:\n{}".format(e.stderr.decode())
+                    )
+                time.sleep(2**attempt)
+
+    try:
+        # prioritize downloading from mirror
+        _download_and_extract(mirror_url)
+    except Exception:
+        # download from official source as a fallback
+        _download_and_extract(url)

metaflow/plugins/pypi/pip.py

@@ -50,10 +50,14 @@ INSTALLATION_MARKER = "{prefix}/.pip/id"
 
 
 class Pip(object):
-    def __init__(self, micromamba=None):
+    def __init__(self, micromamba=None, logger=None):
         # pip is assumed to be installed inside a conda environment managed by
         # micromamba. pip commands are executed using `micromamba run --prefix`
-        self.micromamba = micromamba or Micromamba()
+        self.micromamba = micromamba or Micromamba(logger)
+        if logger:
+            self.logger = logger
+        else:
+            self.logger = lambda *args, **kwargs: None  # No-op logger if not provided
 
     def solve(self, id_, packages, python, platform):
         prefix = self.micromamba.path_to_environment(id_)
@@ -102,9 +106,8 @@ class Pip(object):
             except PipPackageNotFound as ex:
                 # pretty print package errors
                 raise PipException(
-                    "Could not find a binary distribution for %s \n"
-                    "for the platform %s\n\n"
-                    "Note that ***@pypi*** does not currently support source distributions"
+                    "Unable to find a binary distribution compatible with %s for %s.\n\n"
+                    "Note: ***@pypi*** does not currently support source distributions"
                     % (ex.package_spec, platform)
                 )
 
@@ -123,7 +126,7 @@ class Pip(object):
                         **res,
                         subdir_str=(
                             "#subdirectory=%s" % subdirectory if subdirectory else ""
-                        )
+                        ),
                     )
                     # used to deduplicate the storage location in case wheel does not
                     # build with enough unique identifiers.

metaflow/plugins/pypi/utils.py

@@ -1,4 +1,3 @@
-import os
 import platform
 import sys
 
@@ -17,10 +16,13 @@ else:
     from metaflow._vendor.packaging import tags
     from metaflow._vendor.packaging.utils import parse_wheel_filename
 
-from urllib.parse import unquote, urlparse
+from urllib.parse import unquote
 
 from metaflow.exception import MetaflowException
 
+MICROMAMBA_URL = "https://micro.mamba.pm/api/micromamba/{platform}/{version}"
+MICROMAMBA_MIRROR_URL = "https://micromamba.outerbounds.sh/{platform}/{version}.tar.bz2"
+
 
 def conda_platform():
     # Returns the conda platform for the Python interpreter