Description

The customer was upgrading Junos. After the upgrade was completed and they were preparing to restart the two REs, the equipment did not respond.

The system was restored by reinserting and removing the two RE devices. Please analyze.

Symptoms

Oct 10 01:01:06 2025 JTAC mgd[58688]: %DAEMON-5: /usr/libexec/ui/package -X update /var/tmp/junos-vmhost-install-ptx-x86-64-22.4R3-S6.5.tgz -no-validate

Oct 10 01:02:20 2025 JTAC root[59647]: %USER-5: Starting unbundled pkg cfg ptx10016

Oct 10 01:03:31 2025 JTAC chassisd[10555]: %DAEMON-4-CHASSISD_IPC_CONNECTION_DROPPED: Dropped IPC connection for SPMB 0

Oct 10 01:03:31 2025 JTAC alarmd[11033]: %DAEMON-4: Alarm set: SPMB id=33554547, color=RED, class=CHASSIS, reason=SPMB 0 not online

 

Solution

In platforms like Ptx10008,16 that have two management interfaces, using a single variable to track the saved link status (saved_linkstat) was causing frequent and unnecessary updates to /var/run/mgmt_inf_status.

This led to status flooding, as changes in either interface triggered updates regardless of which one actually changed.

This was causing memory over consumption and when the vmhost upgrade command was triggered leading to higher memory need and due to less availability leading to RE hang.

 

labroot@jtac-ptx10008-r2601-re0> start shell user root

Password:

root@jtac-ptx10008-r2601-re0:/var/home/labroot # vhclient -s

Last login: Wed Oct 29 17:42:13 2025 from 192.168.1.2

root@jtac-ptx10008-r2601-re0-node:~# cd /run

root@jtac-ptx10008-r2601-re0-node:/run# du -ah | sort -hr | head -n 10

3.7M .

1.8M ./udev/data

1.8M ./udev

1.5M ./mgmt_intf_status

236K ./systemd

72K ./systemd/journal

68K ./systemd/journal/streams

60K ./systemd/netif

56K ./systemd/netif/links

52K ./systemd/generator.late

root@jtac-ptx10008-r2601-re0-node:/run# more mgmt_intf_status

 

Wed Oct 29 14:17:51 IST 2025 eth0 PCS Links Status

Wed Oct 29 14:17:52 IST 2025 eth1 PCS Links Status

Wed Oct 29 14:17:53 IST 2025 eth0 PCS Links Status

Wed Oct 29 14:17:53 IST 2025 eth1 PCS Links Status

 

This is getting fixed via PR1915464, by creating extra variable to track both port individually.

Modification History

2025-11-27 : Article Created