When this issue happens, the dcpfe crash might be seen due to bad optics or speed mismatch scenarios. Unstable connections with auto-channelization enabled may cause the memory to get exhausted and the system may crash. Continuous or unexplained Interface Device (IFD) creation and deletion can be monitored to check the symptom.
On QFX Series and EX Series switches with auto-channelization support, an optic speed mismatch connection might cause the auto-channelization to get into an infinite loop trying to match a proper speed. In this case, due to some memory leaks, the resources get exhausted, resulting in system crash. The traffic gets disrupted when the system dcpfe restarts. The logs might also show a significant number of errors related to the qfsp-0/0/20, 21, 22 and 23. For example, the configuration might have those ports set for 25g, but they have 40g optics inserted.
https://prsearch.juniper.net/problemreport/PR1484336
As a workaround we may try the following steps:
1) Manually configure with speed to match the inserted optics or remove the optics to avoid the mismatch.
2) Manually configure to disable-auto-speed-detection (even if no optics are inserted).
For a permanent fix the following releases are available per the PR previously shared:
18.4R2-S9 >>> EOS 06/22/2022
19.4R3-S8 >>> EOS 06/26/2023 (For customers in FIPS mode, Junos OS 19.4R3 EOE and EOS are extended to 30 Sept 2024.)
20.1R3 >>> EOS 09/27/2022
20.2R3-S4
20.3R2 >>> EOS 03/29/2023
20.3R3 >>> EOS 03/29/2023
20.3X75-D30 >>> EOS 03/29/2023
20.3X75-D40 >>> EOS 03/29/2023
20.4R2
21.1R1
21.2R1