In a JFE topology, RAM utilization increases continuously on satellites over a longer period of time (approx. 3 months) to over approx. 95%. When this value (approx. 95%) is exceeded, the satellite is no longer accessible and visible on the aggregation device and can only be recovered by switching it off and on again.

Check SNMP OID JUNIPER-MIB::jnxOperatingBuffer over the course of multiple days/weeks - utilization will continue to increase.

This issue can also be identified by gathering "top" command from each satellite and checking for "cpd" process memory utilization piling up on the SD node, as it can be seen in this example:

request chassis satellite login fpc-slot 93

# Login to SD PFE -

vty fpc0

top -b -n 3 | grep cpd

Timestamp: April 5th

2869 root   20  0 33032 29m 1188 S  0 1.5 11:39.04 cpd

Timestamp:

 2869 root   20  0 49352 45m 1188 S  0 2.3 18:12.46 cpd

Timestamp:

2869 root   20  0 70472 66m 1188 S  0 3.3 26:45.41 cpd

This issue is solved in PR 1577977 - [EX9200/EX3400] Junos Fusion Satellite Devices keep dropping out of cluster randomly with spfe and ppman core dumps

2023-06-14: initial publication