Workaround for Antrea-NSX Manager Registration Issues in Tanzu with vSphere NAT Deployment
Article ID: 317179
Updated On:
Products
Issue/Introduction
This KB article provides a comprehensive guide to diagnose and temporarily resolve the issue related to NSX manager registration in a Tanzu environment with vSphere in NAT mode.
Symptoms:
In a VMware Tanzu environment deployed with vSphere in NAT mode, when deploying more than 20 workload clusters, and with Antrea-NSX integration enabled, some clusters fail to register with the NSX manager. The primary symptoms include:
- Controller service on Manager node to Transport node communication is down for at least three minutes.
Controller service on Manager node 192.168.x.y (uuid1) to Transport node (uuid2) down for at least three minutes from Controller service's point of view.
- Errors in the
dmesglog indicating dropped connections due to rate limits.
dmesg: [355518.907296] Dropped per conn limit: IN=eth0 OUT= MAC=mac1 SRC=192.168.0.1 DST=192.168.0.2 LEN=60 TOS=0x00 PREC=0x00 TTL=61 ID=49906 DF PROTO=TCP SPT=49611 DPT=1234 WINDOW=64240 RES=0x00 SYN URGP=0 [355519.938940] Dropped per conn limit: IN=eth0 OUT= MAC=mac2 SRC=192.168.0.1 DST=192.168.0.2 LEN=60 TOS=0x00 PREC=0x00 TTL=61 ID=49907 DF PROTO=TCP SPT=49611 DPT=1234 WINDOW=64240 RES=0x00 SYN URGP=0
- Repeated failure messages in the interworking pod logs, particularly for initializing version handshake.
- It is a known issue since NSX-T 3.2.0 Version
Environment
VMware NSX-T Data Center 3.x
Cause
Resolution
Workaround:
Diagnose the issue
1. check the interworking pod and register pod, if there is connection issue, it is maybe the rate limit issue, for example:
# kubectl --kubeconfig ${workload_cluster_config} logs -nvmware-system-antrea interworking-####-####
E0530 10:18:25.559290 13 controller.go:368] Failed to initialize versionhandshake for antrea_monitoring: rpc error: code = INTERNAL desc = , keep trying E0530 10:18:25.559359 13 controller.go:368] Failed to initialize versionhandshake for antrea_traceflow: rpc error: code = INTERNAL desc = , keep trying E0530 10:18:25.559469 13 controller.go:368] Failed to initialize versionhandshake for antrea_traceflow: rpc error: code = INTERNAL desc = , keep trying E0530 10:18:25.559847 13 controller.go:368] Failed to initialize versionhandshake for antrea_traceflow: rpc error: code = INTERNAL desc = , keep trying
2. Check NSX manager- Login to NSX manager, and run dmesg -T, if there is following errors/warnings.
IN=eth0 OUT= MAC=00:50:56:##:##:##:00:50:56:##:##:##:##:## SRC=10.0.0.1 DST=10.0.0.2 LEN=60 TOS=0x00 PREC=0x00 TTL=61 ID=41118 DF PROTO=TCP SPT=30439 DPT=1235 WINDOW=64240 RES=0x00 SYN URGP=0 [Tue May dd 10:09:57 20yy] IPTables-Dropped: IN=eth0 OUT= MAC=00:50:56:##:##:##:00:50:56:##:##:##:##:## SRC=10.0.0.1 DST=10.0.0.2 LEN=60 TOS=0x00 PREC=0x00 TTL=61 ID=41119 DF PROTO=TCP SPT=30439 DPT=1235 WINDOW=64240 RES=0x00 SYN URGP=0 [Tue May dd 10:09:59 20yy] IPTables-Dropped: IN=eth0 OUT= MAC=00:50:56:##:##:##:00:50:56:##:##:##:##:## SRC=10.0.0.1 DST=10.0.0.2 LEN=60 TOS=0x00 PREC=0x00 TTL=61 ID=41120 DF PROTO=TCP SPT=30439 DPT=1235 WINDOW=64240 RES=0x00 SYN URGP=0 [Tue May dd 10:10:04 20yy] IPTables-Dropped: IN=eth0 OUT= MAC=00:50:56:##:##:##:00:50:56:##:##:##:##:## SRC=10.0.0.1 DST=10.0.0.2 LEN=60 TOS=0x00 PREC=0x00 TTL=61 ID=41121 DF PROTO=TCP SPT=30439 DPT=1235 WINDOW=64240 RES=0x00 SYN URGP=0
If SRC is gateway IP address or the interworking pod IP address, then we hit the rate limit issue, this is because on NSX manager, there is following iptables rules.
# iptables -L INPUTLOG_DROP2 tcp -- anywhere anywhere multiport dports 1234,rmtcfg state NEW,ESTABLISHED #conn src/32 > 10LOG_DROP2 tcp -- anywhere anywhere multiport dports 1234,rmtcfg state NEW,ESTABLISHED limit: above 10000/sec burst 20 mode srcip LOG_DROP tcp -- anywhere anywhere tcp dpt:1235 state NEW,ESTABLISHED limit: above 10000/sec burst 20 mode srcip # iptables -L LOG_DROP LOG all -- anywhere anywhere limit: avg 10/sec burst 5 LOG level warning prefix "IPTables-Dropped: " DROP all -- anywhere anywhere # iptables -L LOG_DROP2 LOG all -- anywhere anywhere limit: avg 10/sec burst 5 LOG level warning prefix "Dropped per conn limit: " DROP all -- anywhere anywhere
In this scenario, it requests from one IP address exceeds 10/s, iptables will drop the request, it is very easy to exceeds 10/s with a gateway and multiple workload clusters.
Workaround
To address the issue, we should login to each of the 3 NSX managers and add a new rule in INPUT:
iptables -A INPUT -i eth0 -p tcp -m multiport --dports 1234,1236 -m state --state NEW,ESTABLISHED -m connlimit --connlimit-above 1000 --connlimit-mask 32 --connlimit-saddr -j LOG_DROP2
iptables -L INPUT --line-numbers
... 13 LOG_DROP2 tcp -- anywhere anywhere tcp dpt:1235 state NEW,ESTABLISHED #conn src/32 > 10 14 LOG_DROP tcp -- anywhere anywhere tcp dpt:1235 state NEW,ESTABLISHED #conn src/32 > 10 15 LOG_DROP tcp -- anywhere anywhere tcp dpt:1235 state NEW,ESTABLISHED limit: above 10000/sec burst 20 mode srcip ... 27 LOG_DROP2 tcp -- anywhere anywhere multiport dports 1234,rmtcfg state NEW,ESTABLISHED #conn src/32 > 1000
Delete the rule that defines rate-limit 10, here is 13th and 14th rule, rule 27 will take effect then.
iptables -D INPUT 14
iptables -D INPUT 13
iptables -L INPUT --line-numbers Chain INPUT (policy DROP) num target prot opt source destination 1 ACCEPT all -- anywhere anywhere state RELATED,ESTABLISHED 2 ACCEPT all -- anywhere anywhere 3 ACCEPT icmp -- anywhere anywhere icmp echo-request 4 ACCEPT tcp -- anywhere anywhere multiport dports ssh,http,https,9000,9040,7070,7071,9090,65000:65002,65010:65012 tcp flags:FIN,SYN,RST,ACK/SYN 5 ACCEPT udp -- anywhere anywhere multiport dports ntp,snmp,65000:65002,65010:65012 6 ACCEPT tcp -- anywhere anywhere tcp spt:domain state ESTABLISHED 7 ACCEPT udp -- anywhere anywhere udp spt:domain 8 ACCEPT tcp -- anywhere anywhere tcp spt:9092 state ESTABLISHED 9 LOG_DROP2 tcp -- anywhere anywhere multiport dports 1234,rmtcfg state NEW,ESTABLISHED limit: above 10000/sec burst 20 mode srcip 10 ACCEPT tcp -- anywhere anywhere multiport dports 1234,rmtcfg state NEW,ESTABLISHED 11 ACCEPT tcp -- anywhere anywhere tcp spt:https state ESTABLISHED 12 ACCEPT tcp -- anywhere anywhere tcp spt:https state ESTABLISHED 13 ACCEPT all -- anywhere anywhere 14 LOG_DROP tcp -- anywhere anywhere tcp dpt:1235 state NEW,ESTABLISHED limit: above 10000/sec burst 20 mode srcip 15 ACCEPT tcp -- anywhere anywhere multiport dports 1235,7777,ssh state NEW,ESTABLISHED 16 ACCEPT tcp -- anywhere anywhere multiport sports 7777,ssh,https,syslog-tls,shell state ESTABLISHED 17 ACCEPT tcp -- anywhere anywhere tcp spt:9000 state ESTABLISHED 18 ACCEPT udp -- anywhere anywhere udp dpts:11000:11004 19 ACCEPT udp -- anywhere anywhere udp spts:11000:11004 20 ACCEPT udp -- anywhere anywhere udp spt:bootps dpt:bootpc 21 ACCEPT icmp -- anywhere anywhere icmp echo-request 22 ACCEPT icmp -- anywhere anywhere icmp echo-reply 23 REJECT udp -- anywhere anywhere udp dpts:33434:33523 reject-with icmp-port-unreachable 24 ACCEPT icmp -- anywhere anywhere icmp destination-unreachable 25 ACCEPT icmp -- anywhere anywhere icmp time-exceeded 26 LOG_DROP2 tcp -- anywhere anywhere multiport dports 1234,rmtcfg state NEW,ESTABLISHED #conn src/32 > 1000
and change the files that defines the rule in /etc/iptables, you can use `grep LOG_DROP2` to search the files, and change rate limit from 10 to 1000, note the files are:
/etc/iptables/nsx-common.v6rules /etc/iptables/nsx-saved-iptables.v4rules /etc/iptables/nsx-common.v4rules
Additional Information
STEPS TO REPRODUCE THE ISSUE
Deploy Tanzu with vSphere in NAT mode and deploy multiple workload clusters (more than 20) and with Antrea-NSX integration enabled, at the same time, some clusters cannot register to NSX manager.
- Currently setup has 12 Antrea work clusters registered with NSX.
- Registration of New Antrea work Cluster is failing as Control Channel to Transport Node Down communication is down.
Impact/Risks:
The primary impact of this issue includes:
- NSX-T adapters cannot register to the NSX manager.
- Antrea-NSX integration fails to function as expected, hindering network functionality in the Tanzu environment.
Feedback
