Azure import try/except over doubles pylint memory usage

[Jamie-]

Bug description

We run pylint on a reasonably large codebase, ~900 .py files, ~350k lines.
Recently we've needed to support two similar versions of the Azure SDK and as such made the change below in a single file.

from

from azure.mgmt.network.v2022_01_01 import NetworkManagementClient
from azure.mgmt.network.v2022_01_01.models import Resource as NetworkResource

to

try:
    from azure.mgmt.network.v2022_01_01 import NetworkManagementClient
    from azure.mgmt.network.v2022_01_01.models import Resource as NetworkResource
except ImportError:
    from azure.mgmt.network import NetworkManagementClient
    from azure.mgmt.network.v2022_07_01.models import Resource as NetworkResource

This change has increased pylint's run time but more importantly, over doubled resident memory usage! I've tried various combinations of disables and module ignores but cannot get anywhere close to previous figures without reverting the code change. Including attempting to ignore the azure module altogether with --ignored-modules=azure

Numbers below are taken from /usr/bin/time -v python3 -m pylint package1 package2 -f colorized -r n -j 1

Old code:

Elapsed (wall clock) time (h:mm:ss or m:ss): 5:13.65
Maximum resident set size (kbytes): 1589112

New code:

Elapsed (wall clock) time (h:mm:ss or m:ss): 6:30.88
Maximum resident set size (kbytes): 3993300

New code with --ignored-modules=azure:

Elapsed (wall clock) time (h:mm:ss or m:ss): 6:42.64
Maximum resident set size (kbytes): 3994000

To make this more generic, I'm running without a pylintrc file using the latest version from PyPI. This has the side effect of producing masses of warning/error output as usually we have a fair number of disables, however the issue described is still present in this state.

Configuration

No response

Command used

/usr/bin/time -v python3 -m pylint package1 package2 -f colorized -r n -j 1
and
/usr/bin/time -v python3 -m pylint --ignored-modules=azure package1 package2 -f colorized -r n -j 1

Pylint output

n/a

Expected behavior

Original run times and memory usage

Pylint version

pylint 3.0.3
astroid 3.0.3
Python 3.9.18 (main, Aug 25 2023, 13:20:14) 
[GCC 11.4.0]

OS / Environment

Ubuntu 22.04

Additional dependencies

# non-Azure modules removed
azure-common==1.1.28
azure-core==1.26.4
azure-identity==1.12.0
azure-keyvault-certificates==4.7.0
azure-keyvault-secrets==4.7.0
azure-mgmt-automation==1.0.0
azure-mgmt-compute==29.1.0
azure-mgmt-core==1.4.0
azure-mgmt-keyvault==10.2.1
azure-mgmt-monitor==6.0.0
azure-mgmt-network==21.0.1
azure-mgmt-resource==23.0.0
azure-mgmt-storage==21.0.0
azure-storage-blob==12.16.0
msrestazure==0.6.4

[jacobtylerwalls]

Thanks for the report! Sounds reasonable to expect ignored-modules to be more useful than it is.

I'm showing this is solved with two tiny patches to pylint and astroid. If you're up for it, would you be willing to confirm and open the PRs? We'd need just a little test and some documentation spruce-ups for the ignored-modules option.

diff --git a/pylint/lint/pylinter.py b/pylint/lint/pylinter.py
index 30250154e..473cd0900 100644
--- a/pylint/lint/pylinter.py
+++ b/pylint/lint/pylinter.py
@@ -1073,6 +1073,7 @@ class PyLinter(
         MANAGER.always_load_extensions = self.config.unsafe_load_any_extension
         MANAGER.max_inferable_values = self.config.limit_inference_results
         MANAGER.extension_package_whitelist.update(self.config.extension_pkg_allow_list)
+        MANAGER.module_denylist.extend(self.config.ignored_modules)
         if self.config.extension_pkg_whitelist:
             MANAGER.extension_package_whitelist.update(
                 self.config.extension_pkg_whitelist

diff --git a/astroid/manager.py b/astroid/manager.py
index c499fe55..386a3838 100644
--- a/astroid/manager.py
+++ b/astroid/manager.py
@@ -59,6 +59,7 @@ class AstroidManager:
         "optimize_ast": False,
         "max_inferable_values": 100,
         "extension_package_whitelist": set(),
+        "module_denylist": [],
         "_transform": TransformVisitor(),
     }
 
@@ -70,6 +71,7 @@ class AstroidManager:
         self.extension_package_whitelist = AstroidManager.brain[
             "extension_package_whitelist"
         ]
+        self.module_denylist = AstroidManager.brain["module_denylist"]
         self._transform = AstroidManager.brain["_transform"]
 
     @property
@@ -200,6 +203,8 @@ class AstroidManager:
         # importing a module with the same name as the file that is importing
         # we want to fallback on the import system to make sure we get the correct
         # module.
+        if modname in self.module_denylist:
+            raise AstroidImportError("Skipping")
         if modname in self.astroid_cache and use_cache:
             return self.astroid_cache[modname]
         if modname == "__main__":

[Jamie-]

Aha ok. I incorrectly assumed ignored-modules did that already.
I checked out the pylint and astroid repos, made those changes locally and ran some benchmarks on my codebase with (this time using all cores to get answers faster).

From what I'm seeing, those changes make no effect to the speed or memory usage when adding --ignored-modules=azure. I'm not quite sure why as adding some debug around the modified code, we are hitting the raise AstroidImportError("Skipping") line with the azure module.

Any pointers?

[jacobtylerwalls]

Make sure you're using the full module name, e.g. azure.mgmt.network.v2022_07_01.models

[Jamie-]

Yep, that helps! Seeing a considerable improvement now. Thanks for that!
Will look to get the changes PRed up.

[jacobtylerwalls]

Wonderful. Feel free to make improvements (e.g. for consistency, that collection should probably be a set...)