On MX10003 platforms, 100G interfaces may fail to come up after a connected Cisco UCS server is rebooted. This behavior has been observed after upgrading to Junos version 23.4R2.13, where interfaces remain down/flaps despite replacing transceivers, cables, or even server hardware components.

This issue is linked to delays in signal detection and DFE tuning when bringing the link up, particularly when using third-party servers (e.g., Cisco UCS) over 100G links.

DFE tuning failure logs observed

[Jan 28 15:46:15.032 LOG: Notice] qsfp-0/1/7 Chan# 2: Rx loss cleared

[Jan 28 15:46:15.032 LOG: Notice] qsfp-0/1/7 Chan# 3: Rx loss cleared

[Jan 28 15:46:16.770 LOG: Info] SMIC_PHY_DFE_TUNING: smic_phy_dfe_tuning_state: et-0/1/7 - DFE coarse/fine tuning completed (took 1000 ms); enabling DFE adaptive tuning.

[Jan 28 15:46:16.773 LOG: Info] MQSS(1): CHMAC1: Cleared Ethernet MAC Local Fault Delta Event for Port 0 (et-0/1/7)

[Jan 28 15:46:16.773 LOG: Info] MQSS(1): CHMAC1: Ethernet PCS Multilane Alignment Done Delta Event for Port 0 (et-0/1/7)

[Jan 28 15:46:17.523 LOG: Info] MQSS(1): CHMAC1: Detected Ethernet MAC Local Fault Delta Event for Port 0 (et-0/1/7)

[Jan 28 15:46:17.523 LOG: Info] MQSS(1): CHMAC1: Ethernet PCS Multilane Alignment Not Done Delta Event for Port 0 (et-0/1/7)

Temporarily configure soft loopback on the interface, then removing loopback after couple of mins, often restores the issue and the interface comes up.

Upgrade to the fix version of the PR 1876314

https://prsearch.juniper.net/problemreport/PR1876314