[neighsyncd] increase neighbor syncd restore timeout to 110 seconds (#745)

yxieca · web-flow · commit d680ce2bc1c8 · 2019-02-07T14:41:21.000-08:00
* [neighsyncd] increase neighbor syncd restore timeout to 120 seconds

Neighbor syncd is restoring important information for teamd and BGP.
our timeout should not be shorter than the down stream service.

Signed-off-by: Ying Xie &lt;ying.xie@microsoft.com&gt;

* [restore_neighbor] improve restore neighbor timeouts

Try to get the bgp timeout and use it for restoring neighbor timeout.
When unavailable, use default 110 seconds.

Signed-off-by: Ying Xie &lt;ying.xie@microsoft.com&gt;

* Set default values according group discussion result

- restore_neighbors.py timeout at 110 seconds due to observed requirement
  of greater than 70 seconds.
- neighbor syncd timeout at 120 seconds (longer than 110 seconds).

Signed-off-by: Ying Xie &lt;ying.xie@microsoft.com&gt;
diff --git a/neighsyncd/neighsync.h b/neighsyncd/neighsync.h
@@ -11,10 +11,10 @@
 
 /*
  * This is the timer value (in seconds) that the neighsyncd waits for restore_neighbors
- * service to finish, should be longer than the restore_neighbors timeout value (60)
+ * service to finish, should be longer than the restore_neighbors timeout value (110)
  * This should not happen, if happens, system is in a unknown state, we should exit.
  */
-#define RESTORE_NEIGH_WAIT_TIME_OUT 70
+#define RESTORE_NEIGH_WAIT_TIME_OUT 120
 
 namespace swss {
 
diff --git a/neighsyncd/restore_neighbors.py b/neighsyncd/restore_neighbors.py
@@ -30,11 +30,12 @@
 logger.setLevel(logging.WARNING)
 logger.addHandler(logging.NullHandler())
 
-# timeout the restore process in 1 min if not finished
+# timeout the restore process in 110 seconds if not finished
 # This is mostly to wait for interfaces to be created and up after system warm-reboot
 # and this process is started by supervisord in swss docker.
-# It would be good to keep that time below routing reconciliation time-out.
-TIME_OUT = 60
+# There had been devices taking close to 70 seconds to complete restoration, setting
+# default timeout to 110 seconds.
+DEF_TIME_OUT = 110
 
 # every 5 seconds to check interfaces states
 CHECK_INTERVAL = 5
@@ -189,13 +190,13 @@ def set_statedb_neigh_restore_done():
 # Once all the entries are restored, this function is returned.
 # The interfaces' states were checked in a loop with an interval (CHECK_INTERVAL)
 # The function will timeout in case interfaces' states never meet the condition
-# after some time (TIME_OUT).
-def restore_update_kernel_neighbors(intf_neigh_map):
+# after some time (DEF_TIME_OUT).
+def restore_update_kernel_neighbors(intf_neigh_map, timeout=DEF_TIME_OUT):
     # create object for netlink calls to kernel
     ipclass = IPRoute()
     mtime = monotonic.time.time
     start_time = mtime()
-    while (mtime() - start_time) < TIME_OUT:
+    while (mtime() - start_time) < timeout:
         for intf, family_neigh_map in intf_neigh_map.items():
             # only try to restore to kernel when link is up
             if is_intf_oper_state_up(intf):
