Handle BucketAlreadyOwnedByYou race condition in S3 bucket initialization

The default S3 bucket name is deterministic per account+region, so
concurrent cluster creation calls (e.g. parallel integration tests)
can race: multiple callers get a 404 from head_bucket, then all
attempt create_bucket. The first succeeds; the rest fail with
BucketAlreadyOwnedByYou, which was previously unhandled and surfaced
as "Unable to initialize s3 bucket."

Treat BucketAlreadyOwnedByYou as a success condition since it confirms
the bucket exists and is owned by the caller, which is the desired
end state. Log at info level and proceed normally.

Author: hanwen-cluster

CHANGELOG.md

@@ -1,6 +1,12 @@
 CHANGELOG
 =========
 
+3.16.0
+------
+
+**BUG FIXES**
+- Fix sporadic S3 bucket creation failure when multiple create-cluster commands are running simultaneously in the same region.
+
 3.15.0
 ------
 

cli/src/pcluster/models/s3_bucket.py

@@ -478,8 +478,11 @@ def _create_bucket(cls, bucket: S3Bucket, e: AWSClientError):
             try:
                 bucket.create_bucket()
             except AWSClientError as error:
-                LOGGER.error("Unable to create S3 bucket %s.", bucket.name)
-                raise error
+                if error.error_code == "BucketAlreadyOwnedByYou":
+                    LOGGER.info("S3 bucket %s was created shortly before by another process. Proceeding.", bucket.name)
+                else:
+                    LOGGER.error("Unable to create S3 bucket %s.", bucket.name)
+                    raise error
         else:
             LOGGER.error("Unable to check S3 bucket %s existence.", bucket.name)
             raise e