From: Oguzhan Ozmen <oozmen@bloomberg.net>
Date: Mon, 24 Nov 2025 03:03:18 +0000 (+0000)
Subject: RGW/test_multi: testcase for rgw instance hang/crash during realm reload
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=5e6a9c8f9de457406c3ef5f44bf2069c60bc6cd5;p=ceph.git

RGW/test_multi: testcase for rgw instance hang/crash during realm reload

Tests: https://tracker.ceph.com/issues/66100
Signed-off-by: Oguzhan Ozmen <oozmen@bloomberg.net>
---

diff --git a/src/test/rgw/rgw_multi/tests.py b/src/test/rgw/rgw_multi/tests.py
index 466a509f978..c732dbb92fe 100644
--- a/src/test/rgw/rgw_multi/tests.py
+++ b/src/test/rgw/rgw_multi/tests.py
@@ -6173,4 +6173,90 @@ def test_object_lock_sync():
     assert(response['ObjectLockConfiguration'] == lock_config)
 
 
-    
+def test_period_update_commit():
+    wkld_concurrency = 25
+    num_objects_to_upload = 2500
+    number_of_period_updates = 5
+    test_passed = False
+
+    zonegroup = realm.master_zonegroup()
+    zonegroup_conns = ZonegroupConns(zonegroup)
+    primary_zone_client_conn = zonegroup_conns.rw_zones[0]
+    secondary_zone_cluster_conn = zonegroup.zones[1]
+
+    bucket = primary_zone_client_conn.create_bucket(gen_bucket_name())
+    log.info(f"created bucket={bucket.name}")
+
+    def run_client_wkld(stop_event: threading.Event, key_range):
+        log.info(f"upload objects within range {key_range} to bucket={bucket.name}")
+        num_uploads = 0
+        while not stop_event.is_set():
+            start, end = key_range
+            for i in range(start, end + 1):
+                try:
+                    primary_zone_client_conn.s3_client.put_object(
+                        Bucket=bucket.name, Key=f"obj-{i:04d}", Body="..."
+                    )
+                    num_uploads += 1
+                except Exception as e:
+                    log.debug(f"failed to upload object to bucket={bucket.name}: {e}")
+        log.info(
+            f"uploaded {num_uploads} times for the range {key_range} to bucket={bucket.name}"
+        )
+
+    try:
+        log.info("verify cluster is healthy before moving on")
+        try:
+            zonegroup_data_checkpoint(zonegroup_conns)
+        except:
+            log.info("restart gateways for a clean start")
+            for z in zonegroup_conns.rw_zones:
+                z.zone.start()
+
+        log.info("start client write-only workload to generate replication traffic")
+        client_write_only_wkld_thread_stop = threading.Event()
+        step = num_objects_to_upload // wkld_concurrency
+        key_ranges = [(i * step, (i + 1) * step - 1) for i in range(wkld_concurrency)]
+        client_write_only_wkld_threads = []
+        for key_range in key_ranges:
+            thread = threading.Thread(
+                target=run_client_wkld,
+                args=(client_write_only_wkld_thread_stop, key_range),
+                daemon=False,
+            )
+            thread.start()
+            client_write_only_wkld_threads.append(thread)
+
+        log.info("run period-update-commits and verify rgw instances reload properly")
+        for _ in range(number_of_period_updates):
+            log.info("issue period update commit")
+            zonegroup.period.update(secondary_zone_cluster_conn, commit=True)
+            log.info("verify data sync is making progress")
+            if not data_sync_making_progress(secondary_zone_cluster_conn):
+                break  # the issue of realm reload freezing reproduced
+        client_write_only_wkld_thread_stop.set()  # stop client wkld
+        zonegroup_data_checkpoint(zonegroup_conns)
+        test_passed = True
+    except Exception as e:
+        log.error(f"test_period_update_commit failed: {e}")
+        raise
+    finally:
+        client_write_only_wkld_thread_stop.set()
+        for t in client_write_only_wkld_threads:
+            t.join()
+
+        # without a fix, an rgw instance may crash or freeze
+        # so restart all instances before further cleaning up
+        if not test_passed:
+            for z in zonegroup_conns.rw_zones:
+                z.zone.start()
+
+        log.info(f"delete {num_objects_to_upload} objects from bucket={bucket.name}")
+        for i in range(num_objects_to_upload):
+            primary_zone_client_conn.s3_client.delete_object(
+                Bucket=bucket.name,
+                Key=f"obj-{i:04d}",
+            )
+        log.info(f"delete bucket={bucket.name}")
+        primary_zone_client_conn.s3_client.delete_bucket(Bucket=bucket.name)
+