F5 Distributed Cloud (XC) Custom Routes: Capabilities, Limitations, and Key Design Considerations
This article explores how Custom Routes work in F5 Distributed Cloud (XC), why they differ architecturally from standard Load Balancer routes, and what to watch out for in real-world deployments, covering backend abstraction, Endpoint/Cluster dependencies, and critical TLS trust and Root CA requirements.XC Distributed Cloud and how to keep the Source IP from changing with customer edges(CE)!
Code is community submitted, community supported, and recognized as ‘Use At Your Own Risk’. Old applications sometimes do not accept a different IP address to be used by the clients during the session/connection. How can make certain the IP stays the same for a client? The best will always be the application to stop tracking users based on something primitive as an ip address and sometimes the issue is in the Load Balancer or ADC after the XC RE and then if the persistence is based on source IP address on the ADC to be changed in case it is BIG-IP to Cookie or Universal or SSL session based if the Load Balancer is doing no decryption and it is just TCP/UDP layer . As an XC Regional Edge (RE) has many ip addresses it can connect to the origin servers adding a CE for the legacy apps is a good option to keep the source IP from changing for the same client HTTP requests during the session/transaction. Before going through this article I recommend reading the links below: F5 Distributed Cloud – CE High Availability Options: A Comparative Exploration | DevCentral F5 Distributed Cloud - Customer Edge | F5 Distributed Cloud Technical Knowledge Create Two Node HA Infrastructure for Load Balancing Using Virtual Sites with Customer Edges | F5 Distributed Cloud Technical Knowledge RE to CE cluster of 3 nodes The new SNAT prefix option under the origin pool allows no mater what CE connects to the origin pool the same IP address to be seen by the origin. Be careful as if you have more than a single IP with /32 then again the client may get each time different IP address. This may cause "inet port exhaustion " ( that is what it is called in F5BIG-IP) if there are too many connections to the origin server, so be careful as the SNAT option was added primary for that use case. There was an older option called "LB source IP persistence" but better not use it as it was not so optimized and clean as this one. RE to 2 CE nodes in a virtual site The same option with SNAT pool is not allowed for a virtual site made of 2 standalone CE. For this we can use the ring hash algorithm. Why this works? Well as Kayvan explained to me the hashing of the origin is taking into account the CE name, so the same origin under 2 different CE will get the same ring hash and the same source IP address will be send to the same CE to access the Origin Server. This will not work for a single 3-node CE cluster as it all 3 nodes have the same name. I have seen 503 errors when ring hash is enabled under the HTTP LB so enable it only under the XC route object and the attached origin pool to it! CE hosted HTTP LB with Advertise policy In XC with CE you can do do HA with 3-cluster CE that can be layer2 HA based on VRRP and arp or Layer 3 persistence based bgp that can work 3 node CE cluster or 2 CE in a virtual site and it's control options like weight, as prepend or local preference options at the router level. For the Layer 2 I will just mention that you need to allow 224.0.0.8 for the VRRP if you are migrating from BIG-IP HA and that XC selects 1 CE to hold active IP that is seen in the XC logs and at the moment the selection for some reason can't be controlled. if a CE can't reach the origin servers in the origin pool it should stop advertising the HTTP LB IP address through BGP. For those options Deploying F5 Distributed Cloud (XC) Services in Cisco ACI - Layer Three Attached Deployment is a great example as it shows ECMP BGP but with the BGP attributes you can easily select one CE to be active and processing connections, so that just one ip address is seen by the origin server. When a CE gets traffic by default it does prefer to send it to the origin as by default "Local Preferred" is enabled under the origin pool. In the clouds like AWS/Azure just a cloud native LB is added In front of the 3 CE cluster and the solution there is simple as to just modify the LB to have a persistence. Public Clouds do not support ARP, so forget about Layer 2 and play with the native LB that load balances between the CE 😉 CE on Public Cloud (AWS/Azure/GCP) When deploying on a public cloud the CE can be deployed in two ways one is through the XC GUI and adding the AWS credentials but this way you have not a big freedom to be honest as you can't deploy 2 CE and make a virtual site out of them and add cloud LB in-front of them as it always will be 3-CE cluster with preconfigured cloud LB that will use all 3 LB! Using the newer "clickops" method is much better https://docs.cloud.f5.com/docs-v2/multi-cloud-network-connect/how-to/site-management/deploy-site-aws-clickops or using terraform but with manual mode and aws as a provider (not XC/volterra as it is the same as the XC GUI deployment) https://docs.cloud.f5.com/docs-v2/multi-cloud-network-connect/how-to/site-management/deploy-aws-site-terraform This way you can make the Cloud LB to use just one CE or have some client Persistence or if traffic comes from RE to CE to implement the virtual site 2 CE node! There is no Layer 2 ARP support as I mentioned in public cloud with 3-node cluster but there is NAT policy https://docs.cloud.f5.com/docs-v2/multi-cloud-network-connect/how-tos/networking/nat-policies but I haven't tried it myself to comment on it. Hope you enjoyed this article!F5 XC Distributed Cloud HTTP Header/Cookie manipulations and using the client ip/user headers
1 . F5 XC distributed cloud HTTP Header manipulations In the F5 XC Distributed Cloud some client information is saved to variables that can be inserted in HTTP headers similar to how F5 Big-IP saves some data that can after that be used in a iRule or Local Traffic Policy. By default XC will insert XFF header with the client IP address but what if the end servers want an HTTP header with another name to contain the real client IP. Under the HTTP load balancer under "Other Options" under "More Options" the "Header Options" can be found. Then the the predefined variables can be used for this job like in the example below the $[client_address] is used. A list of the predefined variables for F5 XC: https://docs.cloud.f5.com/docs/how-to/advanced-security/configure-http-header-processing There is $[user] variable and maybe in the future if F5 XC does the authentication of the users this option will be insert the user in a proxy chaining scenario but for now I think that this just manipulates data in the XAU (X-Authenticated-User) HTTP header. 2. Matching of the real client ip HTTP headers You can also match a XFF header if it is inserted by a proxy device before the F5 XC nodes for security bypass/blocking or for logging in the F5 XC. For User logging from the XFF Under "Common Security Controls" create a "User Identification Policy". You can also match a regex that matches the ip address and this is in case there are multiple IP addresses in the XFF header as there could have been many Proxy devices in the data path and we want see if just one is present. For Security bypass or blocking based based on XFF Under "Common Security Controls" create a "Trusted Client Rules" or "Client Blocking Rules". Also if you have "User Identification Policy" then you can just use the "User Identifier" but it can't use regex in this case. I have made separate article about User-Identification F5 XC Session tracking and logging with User Identification Policy | DevCentral To match a regex value in the header that is just a single IP address, even when the header has many ip addresses, use the regex (1\.1\.1\.1) as an example to mach address 1.1.1.1. To use the client IP address as a source Ip address to the backend Origin Servers in the TCP packet after going through the F5 XC (similar to removing the SNAT pool or Automap in F5 Big-IP) use the option below: The same way the XAU (X-Authenticated-User) HTTP header can be used in a proxy chaining topology, when there is a proxy before the F5 XC that has added this header. Edit: Keep in mind that in some cases in the XC Regex for example (1\.1\.1\.1) should be written without () as 1\.1\.1\.1 , so test it as this could be something new and I have seen it in service policy regex matches, when making a new custom signature that was not in WAAP WAF XC policy. I could make a seperate article for this 🙂 XC can even send the client certificate attributes to the backend server if Client Side mTLS is enabled but it is configured at the cert tab. 3. F5 XC distributed cloud HTTP Cookie manipulations. Now you can overwrite the XC cookie by keeping the value but modifying the tags and this is big thing as before this was not possible. When combined with cookies this becomes very powerful thing as you can match on User-Agent header and for Mozilla for example to change the flags as if there is bug with the browser etc. The feature changes cookies returned in the Response Set-Cookie header from the origin server as it should.F5 XC vk8s workload with Open Source Nginx
I have shared the code in the link below under Devcentral code share: F5 XC vk8s open source nginx deployment on RE | DevCentral Here I will desribe the basic steps for creating a workload object that is F5 XC custom kubernetes object that creates in the background kubernetes deployments, pods and Cluster-IP type services. The free unprivileged nginx image nginxinc/docker-nginx-unprivileged: Unprivileged NGINX Dockerfiles (github.com) Create a virtual site that groups your Regional Edges and Customer Edges. After that create the vk8s virtual kubernetes and relate it to the virtual site."Note": Keep in mind for the limitations of kubernetes deployments on Regional Edges mentioned in Create Virtual K8s (vK8s) Object | F5 Distributed Cloud Tech Docs. First create the workload object and select type service that can be related to Regional Edge virtual site or Customer Edge virtual site. After select the container image that will be loaded from a public repository like github or private repo. You will need to configure advertise policy that will expose the pod/container with a kubernetes cluster-ip service. If you are deploying test containers, you will not need to advertise the container . To trigger commands at a container start, you may need to use /bin/bash -c -- and a argument."Note": This is not related for this workload deployment and it is just an example. Select to overwrite the default config file for the opensource nginx unprivileged with a file mount. "Note": the volume name shouldn't have a dot as it will cause issues. For the image options select a repository with no rate limit as otherwise you will see the error under the the events for the pod. You can also configure command and parameters to push to the container that will run on boot up. You can use empty dir on the virtual kubernetes on the Regional Edges for volume mounts like the log directory or the Nginx Cache zone but the unprivileged Nginx by default exports the logs to the XC GUI, so there is no need. "Note": This is not related for this workload deployment and it is just an example. The Logs and events can be seen under the pod dashboard and even the container/pod can accessed. "Note": For some workloads to see the logs from the XC GUI you will need to direct the output to stderr but not for nginx. After that you can reference the auto created kubernetes Cluster-IP service in a origin pool, using the workload name and the XC namespace (for example niki-nginx.default). "Note": Use the same virtual-site where the workload was attached and the same port as in the advertise cluster config. Deployments and Cluster-IP services can be created directly without a workload but better use the workload option. When you modify the config of the nginx actually you are modifying a configmap that the XC workload has created in the background and mounted as volume in the deployment but you will need to trigger deployment recreation as of now not supported by the XC GUI. From the GUI you can scale the workload to 0 pod instances and then back to 1 but a better solution is to use kubectl. You can log into the virtual kubernetes like any other k8s environment using a cert and then you can run the command "kubectl rollout restart deployment/niki-nginx". Just download the SSL/TLS cert. You can automate the entire process using XC API and then you can use normal kubernetes automation to run the restart command F5 Distributed Cloud Services API for ves.io.schema.views.workload | F5 Distributed Cloud API Docs! F5 XC has added proxy_protocol support and now the nginx container can work directly with the real client ip addresses without XFF HTTP headers or non-http services like SMTP that nginx supports and this way XC now can act as layer 7 proxy for email/smpt traffic 😉. You just need to add "proxy_protocol" directive and to log the variable "$proxy_protocol_addr". Related resources: For nginx Plus deployments for advanced functions like SAML or OpenID Connect (OIDC) or the advanced functions of the Nginx Plus dynamic modules like njs that is allowing java scripting (similar to F5 BIG-IP or BIG-IP Next TCL based iRules), see: Enable SAML SP on F5 XC Application Bolt-on Auth with NGINX Plus and F5 Distributed Cloud Dynamic Modules | NGINX Documentation njs scripting language (nginx.org) Accepting the PROXY Protocol | NGINX Documentation
