Help us improve your experience.

Let us know what you think.

Do you have time for a two-minute survey?


Open Issues

Learn about open issues in Junos OS Release 21.4R3 for QFX Series switches.

For the most complete and latest information about known Junos OS defects, use the Juniper Networks online Junos Problem Report Search application.

General Routing

  • When VLAN is added as an action for changing the VLAN in both ingress and egress filters, the filter is not installed. PR1362609

  • VXLAN VNI (multicast learning) scaling on QFX5110 traffic issue is seen from VXLAN tunnel to Layer 2 interface. PR1462548

  • When running the command: show pfe filter hw filter-name filter name, the command fails to retrieve the PFE programming details of the filter. PR1495712

  • On PTX Series routers and QFX Series switches running Junos OS, the JFlow service might not report the accurate throughput rate. This issue is seen when there is high sampled traffic rate with low flow cache hit ratio.PR1502645

  • On QFX5100 devices not running the qfx-5e codes (non-TVP architecture), when an image with the Broadcom SDK upgrade (6.5.x) is installed, the CPU utilization may go up by around 5%. PR1534234

  • On the QFX5000 line of switches, route leaking does not work for IPv4 routes if the mask is less than /16 and for IPV6 routes if the mask is less than /64. PR1538853

  • The FPC process may not get spawned after hard reboot in a rare case, which causes the FPC to not come online successfully. PR1540107

  • 5M DAC connected between QFX10002-60C and MX2010 doesn't link up. But with 1M and 3M DAC this interop works as expected. Also it is to be noted QFX10002-60C and ACX or Traffic generator the same 5M DAC works seamlessly. There seems to be certain SI or link level configuration on both QFX10002-60C and MX2010 which needs to be debugged with the help from HW and SI teams and resolved. PR1555955

  • To avoid the additional interface flap , interface hold time needs to be configured . PR1562857

  • In EVPN/VXLAN scenario with OSPF configured over the IRB, OSPF sessions might not get established due to connectivity issues. PR1577183

  • On QFX5100, while checking DHCP Smart Relay over IRB interfaces, the renew-ack's may not be seen in the dhcp client. PR1581025

  • In a fully loaded devices, at times, firewall programming was failing due to scaled prefix configuration with more than 64800 entries. However, this issue is not observed in development setup. PR1581767

  • On PTX10K and QFX10K platforms running Junos, file permissions might be changed for /var/db/scripts files after rebooting the device. This issue might have an impact on the scripts running on the box. PR1583839

  • On QFX5000 line of switches, the Flexible PIC Concentrator (FPC) or dcpfe process might go into a very uncommon state when multiple Broadcom Counter (bcmCNTR) threads are running or spawned in FPC. This state causes the dcpfe process to crash or the FPC to reboot. The purpose of bcmCNTR is to poll statistics from hardware. PR1588704

  • On QFX/EX series switches with Broadcom chip as Packet Forwarding Engine (PFE), if IS-IS is enabled on an integrated routing and bridging (IRB) interface and the maximum transmission unit (MTU) size of the IRB interface is configured with a value great than 1496 bytes, the IS-IS hello (IIH) PDUs with jumbo frame size (i.e., great than 1496 bytes) might be dropped and not sent to the IS-IS neighbors. The following is the product list of QFX/EX series switches with Broadcom chip as PFE. QFX5100/QFX5110/QFX5120/QFX5130/QFX5200/QFX5210/QFX5220 EX2300/EX3400/EX4300/EX4600/EX4650 PR1595823

  • Pim Vxlan not working on TD3 chipsets enabling VxLAN flexflow after Junos OS Release 21.3R1. Customers Pim Vxlan or data plane VxLAN can use the Junos OS Release 21.3R1. PR1597276

  • Comparing convergence time with 21.1R1.5 , seen degradation in isisv6 , ospfv2 and ospfv3 convergence time. As it is a convergence time issue, many components will be involved and hence need investigation from multiple teams (RPD, Kernel, PFE). These kind of issue will require multiple iteration of experiments to narrow it down, a bit difficult to do RCA and it takes some good amount of time. Considering these points, marking this as known issue. PR1602334

  • On the PTX1000 and PTX10000 lines of routers and the QFX10000 line of switches with sFlow enabled, the sFlow samples might not be generated for transit MPLS traffic carrying IPv6 packets.PR1607497

  • In QFX10002-60C under mac statistics "output-mac-control-frames" and "output-mac-pause-frames" does not increment. PR1610745

  • On QFX5100 VC, when 118 (Max) lag groups are configured then there may be traffic loss of few packets intermittently. PR1611162

  • When IFA2.0 Init feature enabled on switch and flows are sampled, we will see incorrect pps and bps stats at IFL level on ingress and egress ports. PR1620139

  • The led port init was done for SXE port PR1621630

  • On Junos QFX10k platforms with scaled number of BFD (Bidirectional Forwarding Detection) sessions configured, addition of a new BFD session might cause flapping in newly added session and other existing BFD sessions.PR1621976

  • On QFX platforms, ipv6 traffic output byte(ipv6-transit-statistics) would not be in expected range as per traffic generator stats.PR1653671

  • S/W to add contentPR1663804

  • When the remote end server/system reboots, QFX5100 platform ports with SFP-T 1G inserted may go into a hung state and remain in that state even after the reboot is complete. This may affect traffic after the remote end system comes online and resumes traffic transmission.PR1665800

  • On QFX5100, for copper media, the media type may be shown as Fiber instead of copperPR1672293

  • Each locally learned ARP/ND Nexthop requires unique fabric token from Kernel. This unique token maps to physical address in HW and this address points to EDF memory for the ARP/ND nexthops. Token pool in kernel is also used by different features like flood nexthops, arp/ndp nexthops. Token usage has increased as tokens are now used by IRB interfaces and default mesh groups also. 96k ARP/ND scale might not be achievable always. It is recommended to scale upto 95000 ARP/ND.PR1673626

  • On QFX5110, dc-pfe core might be seen with PTP IRB configuration. This is because of a timing issue between the periodic poll and stream addition. PR1683308


  • Multiple remote DCI MAC (IRB and Host) entries go missing after the I-ESI modification in the local DC GW nodes.PR1600600

  • EVPN Local ESI Mac limit config mayn't not get effective immediately when it has already learned remote MH Macs. Clear the Mac table from all MH PEs and configure the Mac limit over local ESI interfaces. PR1619299

Interfaces and Chassis

  • If mc-lag is enabled on interfaces within a routing-instance (with consistency-check config enabled), then mclag-cfgchkd core will happen. consistency-check config: set multi-chassis mc-lag consistency-check .. Impact: mclag-cfgchkd daemon will core, but that will not impact any functionality. This is a day-1 issue, and is seen on all the devices which support mclag consistency-check. PR1599025

Layer 2 Features

  • In case of the access-side interfaces used as SP-style interfaces, when a new logical interface is added and if there is already a logical interface on the physical interface, there is 20--50 ms traffic drop on the existing logical interface. PR1367488

Layer 2 Ethernet Services

  • The DHCP client config is coming from two places, i.e AIU script and vsdk sandbox. The DHCP client config coming from AIU script has the serial Id in vendor id where as the default config from sandbox doesn't have. There is no impact on functionality or service. PR1601504

Platform and Infrastructure

  • When the DHCP relay mode is configured as no-snoop, we are observing the offer gets dropped due to incorrect asic programming. This issue only affects while running DHCP relay on EVPN/VXLAN environment. PR1530160

  • On all Junos and Junos OS Evolved platforms, while using source-address NTP configuration parameter and issue the command "set ntp date" from the CLI, packets will be sent with the source address of the outgoing interface rather than the manually configured IP address. Typically, the manually configured IP address would be a loopback address. The problem does not apply to automatically generated NTP poll packets.PR1545022

Routing Protocols

  • Multicast traffic is hogging the switch core when igmp-snooping is removed. The mcsnoopd might crash due to the changes in mrouter interfaces and routes.PR1569436

  • When the knob accept-remote-source under PIM is removed, the PIM SG entries may not be updated with the correct RPF. Clearing of the states would take care of the issue. This is day-1 behavior. PR1593283

  • MCSNOOPD core is seen sometimes due to Nexthop index being quickly reused by the Kernel. As a result when application is still holding old Nexthop reference which is waiting for deletion response from Kernel, the same Nexthop Index can be hence be received from other applications like RPD for EVPN core-NH updates as in current case. This will lead to MCSNOOPD wrongly manipulating the Nexthop ref counting, leading to using a freed Nexthop memory when this Nexthop-index is finally being freed. This will be fixed via a feature enhancement underway where the Kernel will maintain a timer to ensure a Nexthop-index is not put to free pool immediately for reuse and hence can be reused post the new timer expiry. PR1605393

  • In EVPN-OISM (Optimized inter-subnet multicast), with listener behind Server Leaf (SL) and source outside fabric, if the Server leaf does not have a valid supported unicast route (EVPN Type-5 unicast route is not supported with EVPN-OISM) to the source, might drop the flow.PR1665791

  • All Junos (other than MX) configured with sBFD responder with the following command: set protocols bfd sbfd local-discriminator which generates FPC core file and leads to traffic drop. PR1678016