GH-38364: [Python] Initialize S3 on first use

python/pyarrow/_s3fs.pyx

@@ -70,6 +70,13 @@ def finalize_s3():
     check_status(CFinalizeS3())
 
 
+def ensure_s3_finalized():
+    """
+    Finalize S3 if already initialized
+    """
+    check_status(CEnsureS3Finalized())
+
+
 def resolve_s3_region(bucket):
     """
     Resolve the S3 region of a bucket.
@@ -93,6 +100,8 @@ def resolve_s3_region(bucket):
         c_string c_bucket
         c_string c_region
 
+    ensure_s3_initialized()
+
     c_bucket = tobytes(bucket)
     with nogil:
         c_region = GetResultValue(ResolveS3BucketRegion(c_bucket))
@@ -260,6 +269,8 @@ cdef class S3FileSystem(FileSystem):
                  load_frequency=900, proxy_options=None,
                  allow_bucket_creation=False, allow_bucket_deletion=False,
                  retry_strategy: S3RetryStrategy = AwsStandardS3RetryStrategy(max_attempts=3)):
+        ensure_s3_initialized()
+
         cdef:
             optional[CS3Options] options
             shared_ptr[CS3FileSystem] wrapped

python/pyarrow/fs.py

@@ -54,13 +54,16 @@
     from pyarrow._s3fs import (  # noqa
         AwsDefaultS3RetryStrategy, AwsStandardS3RetryStrategy,
         S3FileSystem, S3LogLevel, S3RetryStrategy, ensure_s3_initialized,
-        finalize_s3, initialize_s3, resolve_s3_region)
+        finalize_s3, ensure_s3_finalized, initialize_s3, resolve_s3_region)
 except ImportError:
     _not_imported.append("S3FileSystem")
 else:
-    ensure_s3_initialized()
+    # GH-38364: we don't initialize S3 eagerly as that could lead
+    # to crashes at shutdown even when S3 isn't used.
+    # Instead, S3 is initialized lazily using `ensure_s3_initialized`
+    # in assorted places.
     import atexit
-    atexit.register(finalize_s3)
+    atexit.register(ensure_s3_finalized)
 
 
 def __getattr__(name):

python/pyarrow/includes/libarrow_fs.pxd

@@ -211,6 +211,7 @@ cdef extern from "arrow/filesystem/api.h" namespace "arrow::fs" nogil:
         const CS3GlobalOptions& options)
     cdef CStatus CEnsureS3Initialized "arrow::fs::EnsureS3Initialized"()
     cdef CStatus CFinalizeS3 "arrow::fs::FinalizeS3"()
+    cdef CStatus CEnsureS3Finalized "arrow::fs::EnsureS3Finalized"()
 
     cdef CResult[c_string] ResolveS3BucketRegion(const c_string& bucket)