This article outlines an issue observed on QFX5200 platforms running Junos, where exhaustion of ECMP (Equal-Cost Multi-Path) resources can leave the Packet Forwarding Engine (PFE) in an inconsistent state. As a result, ECMP next-hops (NHs) may fail to install even after ECMP resource utilization returns to normal. This condition may cause traffic impact due to incomplete or missing next-hop installations, despite the apparent recovery of ECMP resources. The issue persists until the PFE is manually reset or the device is rebooted.

Root Cause:

When the ECMP table reaches full capacity, failed next-hop (NH) entries are temporarily assigned a hold next-hop index. Once ECMP resources become available, logic introduced by PR 1854995 attempts to reinstall these held NH entries. However, due to a flaw in this optimization, the reinstallation process fails, resulting in incomplete ECMP group programming in hardware.

Affected Releases:
This issue is a regression introduced by PR 1854995. Any Junos release that includes the fix for PR 1854995 but does not include the fix for PR 1870380 is affected.

Prevention:
To avoid triggering this issue, ensure ECMP usage remains within the supported scale limits to prevent resource exhaustion.

Workaround:
If the system enters this inconsistent state, the issue can be temporarily resolved by either:
    •    Restarting the Packet Forwarding Engine (PFE), or
    •    Rebooting the switch.
 

Fix:
This issue is resolved in PR 1870380. For information on which Junos releases include this fix, refer to the PR details or contact JTAC.