This Article explains the reason behind high memory utilization on the backup routing engine: user@device> show chassis routing-engine | match "slot|elec|state|time|Memory " Slot 0: Current state Master Election priority Master (default) Memory utilization 29 percent ... Slot 1: Current state Backup Election priority Backup (default) Memory utilization 89 percent
The following log may help identify the issue: user@device> show system processes extensive | match % | except " 0.00%" | except "Inuse|idle|nhinfo" 67129 root 36 0 26768M 25700M CPU5 5 55:05 23.00% rpd{rpd} user@device> show system processes extensive | match % | except " 0.00%" | except "Inuse|idle|nhinfo" 67129 root 23 0 27488M 26273M kqread 5 67:36 6.40% rpd{rpd}
Memory leak may occur on Backup RE, when the contribute nh flap , such as member interface of AE is brought down or brought up. The memory leak may occur when building unilist nexthop in backup rpd. On all Junos platforms with dual REs, rpd memory leak may be seen when an AE member interface flaps ,or contribute nh flap, or immediate restart of master RE. The memory leak was observed be around 32 bytes per session, the leak is only seen when an unilist nexthop have more than 8 legs. This issue might be seen if the following conditions are met: * On all Junos platforms with dual REs * GRES (Graceful Routing Engine Switchover) and NSR (Nonstop Active Routing) are enabled * On AE member interface flap or contribute nh flap triggered unilist nh udpate * On immediate restart of master RE * On unilist nh with more than 8 legs
In order to resolve the issue, upgrade to a fixed release. Refer to PR1539601 .We can temperately address the problem by deactivating/activating GRES and NSR before Junos upgrade. The work plan is: