Merge pull request ceph#989 from leonidc/fix_rebalance_exeption_handling

fix exception handling in rebalane thread
gbregman · Dec 19, 2024 · 719a6ee · 719a6ee
2 parents 2539ac4 + e8d349c
commit 719a6ee
Show file tree

Hide file tree

Showing 2 changed files with 9 additions and 2 deletions.
diff --git a/control/rebalance.py b/control/rebalance.py
@@ -27,10 +27,11 @@ def __init__(self, gateway_service):
         self.ceph_utils = gateway_service.ceph_utils
         self.rebalance_period_sec              = gateway_service.config.getint_with_default("gateway",  "rebalance_period_sec", 7)
         self.rebalance_max_ns_to_change_lb_grp = gateway_service.config.getint_with_default("gateway", "max_ns_to_change_lb_grp", 8)
-        self.auto_rebalance = threading.Thread(target=self.auto_rebalance_task, daemon=True)
+        self.rebalance_event = threading.Event()
+        self.auto_rebalance = threading.Thread(target=self.auto_rebalance_task, daemon=True, args=(self.rebalance_event,))
         self.auto_rebalance.start() #start the thread
 
-    def auto_rebalance_task(self ):
+    def auto_rebalance_task(self, death_event):
         """Periodically calls for auto rebalance."""
         while (self.rebalance_period_sec > 0):
             for i in range(self.rebalance_max_ns_to_change_lb_grp):
@@ -41,6 +42,8 @@ def auto_rebalance_task(self ):
                         break
                 except Exception:
                         self.logger.exception(f"Exception in auto rebalance")
+                        if death_event:
+                            death_event.set()
                         raise
                 time.sleep(0.01) #release lock for 10ms after rebalancing each 1 NS
             time.sleep(self.rebalance_period_sec)

diff --git a/control/server.py b/control/server.py
@@ -692,6 +692,10 @@ def keep_alive(self):
             spdk_ping_interval_in_seconds = 0.0
 
         while True:
+            if self.gateway_rpc:
+                if self.gateway_rpc.rebalance.rebalance_event.is_set():
+                    self.logger.critical(f"Failure in rebalance, aborting")
+                    raise SystemExit(f"Failure in rebalance, quitting gateway")
             timedout = self.server.wait_for_termination(timeout=1)
             if not timedout:
                 break