OSPF Topology Watcher

OSPF Watcher is a monitoring tool of OSPF topology changes for network engineers. It works via passively listening to OSPF control plane messages through either a specially established OSPF adjacency over a GRE tunnel (GRE mode) or by receiving BGP-LS updates from a network router (BGP-LS mode). The tool logs OSPF events and/or export by Logstash to Elastic Stack (ELK), Zabbix, WebHooks and Topolograph monitoring dashboard for keeping the history of events, alerting, instant notification. Components of the solution are wrapped into containers, so it can be increadebly fast to start it. In GRE mode, GRE tunnel setup on the Linux host is configured manually.

Note

Tracking OSPF topology changes via BGP-LS is supported starting from Docker image vadims06/ospf-watcher:v3.1.0. For older images, BGP-LS support is not available.

Logged topology changes:

• OSPF neighbor adjacency Up/Down
• OSPF link cost changes
• OSPF networks appeared/disappeared from the topology
• OSPF TE attributes (via Opaque LSA / BGP-LS):
  • Administrative Group (color, resource class)
  • Maximum Link Bandwidth
  • Maximum Reservable Link Bandwidth
  • Unreserved Bandwidth
  • Traffic Engineering Default Metric

Architecture

Connection Modes

OSPF Watcher supports two connection modes:

GRE Mode (Traditional)

The FRR container is isolated in an individual network namespace and establishes an OSPF adjacency over a GRE tunnel with a network device. The XDP OSPF filter inspects all outgoing OSPF advertisements. It checks if FRR instance advertises only locally connected network (assigned on GRE tunnel) and no more. If it advertises multiple networks, OSPF Database description (DB) or LSUpdate will be dropped. It prevents the network from populating by unexpected network prefixes.

BGP-LS Mode (New)

The watcher connects to a network router via BGP-LS (BGP Link-State) protocol. The router must be configured to advertise OSPF topology information via BGP-LS. The watcher uses GoBGP (bgplswatcher) to establish a BGP session and receive BGP-LS updates, which are then processed by the OSPF watcher component. This mode eliminates the need for GRE tunnels and OSPF adjacencies, making it easier to deploy in environments where GRE tunnels are not feasible.

Each Watcher instance maintains all routes and updates within an isolated network namespace. This isolation ensures efficient monitoring without interference and prevent route leaks.

Note

ospfwatcher:v1.1 is compatible with topolograph:v2.7 , it means that OSPF network changes can be shown on the network graph.

Functional Role

Demo

Instant notifications

Click on the image in order zoom it.

Discovering OSPF logs in Kibana. Examples

OSPF cost changes on links

Logs if OSPF adjacency was Up/Down or any networks appeared/disappeared.

Topolograph OSPF Monitoring. New subnet event shows where the subnet appeared

Topolograph OSPF Monitoring. Filter any subnet-related events, select Change metric event

new and old metric is shown

Topolograph OSPF Monitoring. up/down link events

Red timelines show link (~adjacency) down events, green one - up link (~adjacency).
Timeline 10.1.1.2-10.1.1.3 has been selected.

OSPF topology change notification/alarming via Zabbix. Examples

Zabbix's dashboard with active OSPF alarms detected by OSPFWatcher

Zabbix OSPF neighbor up/down alarm

This alarm tracks all new OSPF adjacencies or when device loses its OSPF neighbor

Zabbix OSPF Cost changed on transit links

Transit links are all links between active OSPF neighbors. If cost on a link was changed it might affect all actual/shortest paths traffic follows

Zabbix alert if OSPF network was stopped announcing from node

If a subnet was removed from OSPF node (the node withdrew it from the announcement) it means the network from this node became unavailable for others, this event will be logged too.

Slack notification

HTTP POST messages can be easily accepted by messengers, which allows to get instant notifications of OSPF topology changes:

Quick lab

Containerlab

Here is a lab for tracking OSPF topology changes placed here containerlab/frr01. Watcher logs:

OSPF topology changes are printed by Watcher in a text file only.

./containerlab/frr01/prepare.sh
sudo clab deploy --topo ./containerlab/frr01/frr01.clab.yml

How to connect OSPF watcher to real network

Table below shows different options of possible setups, starting from the bare minimum in case of running Containerlab for testing and ending with maximum setup size with Watcher, Topolograph and ELK. The following setup describes setup №2.

№	Deployment size	Compose files	Text logs	View on map	Zabbix / HTTP / Messengers	Search events
1	Bare minimum (Containerlab)	0	+	−	−	−
2	Local Topolograph + local compose (ELK disabled)	2	+	+	+	−
3	Local Topolograph + local compose (ELK enabled)	3	+	+	+	+
4	Same as №2 but Fluent Bit instead of Logstash (Zabbix not available)	2	+	+	HTTP/Webhook only	−

Setup №2. Text logs + timeline of network changes on Topolograph

1. Choose a Linux host with Docker installed
2. Run script:

curl -O https://raw.githubusercontent.com/Vadims06/topolograph-docker/master/install.sh
chmod +x install.sh
sudo ./install.sh

It will:

1. Setup Topolograph It's needed for network events visualization on Topolograph UI. Skip if you don't want it.

• launch your own Topolograph on docker using topolograph-docker or make sure you have a connection to the public https://topolograph.com
• create a user for API authentication using Local Registration form on the Topolograph page, add your IP address in API/Authorised source IP ranges. Set variables in .env file:

Note

• TOPOLOGRAPH_HOST - set the IP address of your host, where the docker is hosted (if you run all demo on a single machine), do not put localhost, because ELK, Topolograph and OSPF Watcher run in their private network space
• TOPOLOGRAPH_PORT - by default 8080
• TOPOLOGRAPH_WEB_API_USERNAME_EMAIL - by default ospf@topolograph.com or put your recently created user
• TOPOLOGRAPH_WEB_API_PASSWORD - by default ospf
• TEST_MODE - if mode is True, a demo OSPF events from static file will be uploaded, not from FRR

2. Setup OSPF Watcher

git clone https://github.com/Vadims06/ospfwatcher.git
cd ospfwatcher
cp .env.template .env

Generate configuration files
vadims06/ospf-watcher:v1.7 includes a client for generating configurations for each Watcher for each OSPF area. To generate individual settings - run the client with --action add_watcher

sudo docker run -it --rm --user $UID -v ./:/home/watcher/watcher/ -v /etc/passwd:/etc/passwd:ro -v /etc/group:/etc/group:ro vadims06/ospf-watcher:latest python3 ./client.py --action add_watcher

Output:

+---------------------------+
|  Watcher Host             |                       +-------------------+
|  +------------+           |                       | Network device    |
|  | netns FRR  |           |                       |                   |
|  |            Tunnel [4]  |                       | Tunnel [4]        |
|  |  gre1   [3]TunnelIP----+-----------------------+[2]TunnelIP        |
|  |  eth1------+-vhost1    |       +-----+         | OSPF area num [5] |
|  |            | Host IP[6]+-------+ LAN |--------[1]Device IP         |
|  |            |           |       +-----+         |                   |
|  +------------+           |                       |                   |
|                           |                       +-------------------+
+---------------------------+
[1]Network device IP [x.x.x.x]: 

The script will create:

1. a folder under watcher folder with FRR configuration under router folder
2. a containerlab configuration file with network settings
3. an individual watcher log file in watcher folder.

OSPF routes of each Watcher instance stay isolated in watcher's network namespace. To stop OSPF routes from being installed even in the watcher's network namespace, we the following policy has been applied on the watcher:

# quagga/config/ospfd.conf
route-map TO_KERNEL deny 200
exit
!
ip protocol ospf route-map TO_KERNEL

5. Start OSPF Watcher
   Install containerlab To start the watcher run the following command. clab deploy is like a docker compose up -d command

sudo clab deploy --topo watcher/watcher1-tun1025/config.yml

It will create:

• Individual network namespace for Watcher and FRR
• A pair of tap interfaces to connect the watcher to Linux host
• GRE tunnel in Watcher's namespace
• NAT settings for GRE traffic
• FRR & Watcher instance
• assign XDP OSPF filter on watcher's tap interface

6. Start log export to Topolograph and/or ELK (optionally if you configured Step 2 or 3)

docker-compose build
docker-compose up -d

Compose profiles (Logstash vs Fluent Bit):

• Default (no profile): starts Logstash and the index-creator. Use docker compose up -d (or docker compose up -d ospf-logstash ospf-logstash-index-creator).
• Fluent Bit only: the fluent-bit service has profile fluent-bit. To start only Fluent Bit (no Logstash, no index-creator), run:

  docker compose --profile fluent-bit up -d fluent-bit

  Pipeline definition: fluentbit/fluent-bit.yaml (bind-mounted into the container). Logstash and Fluent Bit send HTTP payloads in different shapes; see HTTP output: Logstash vs Fluent Bit under Minimum version.

Device configuration

Setup GRE tunnel from the network device to the host. An example for Cisco

Note

You can skip this step and run ospfwatcher in test_mode, so test LSDB from the file will be taken and test changes (loss of adjacency and change of OSPF metric) will be posted in ELK.

interface gigabitether0/1
ip address <GRE tunnel ip address>
tunnel mode gre
tunnel source <router-ip>
tunnel destination <host-ip>
ip ospf network type point-to-point

Set GRE tunnel network where is placed to quagga/config/ospfd.conf

Check OSPF neighbor, if there is no OSPF adjacency between network device and OSPF Watcher, check troubleshooting OSPF Watcher <-> Network device connection section below (to run diagnostic script).

Optionally

Setup ELK (skip it, it's only needed for setup № 3)

• if you already have ELK instance running, fill ELASTIC_IP in env file and uncomment Elastic config here ospfwatcher/logstash/pipeline/logstash.conf. Currently additional manual configuration is needed for Index Templates creation, because create.py script doesn't accept the certificate of ELK. It's needed to have one in case of security setting enabled. Required mapping for the Index Template is in ospfwatcher/logstash/index_template/create.py. To create Index Templates, run:

sudo docker run -it --rm --env-file=./.env -v ./logstash/index_template/create.py:/home/watcher/watcher/create.py vadims06/ospf-watcher:latest python3 ./create.py

• if not - boot up a new ELK from docker-elk compose. For demo purporse set license of ELK as basic and turn off security. The setting are in docker-elk/elasticsearch/config/elasticsearch.yml

xpack.license.self_generated.type: basic
xpack.security.enabled: false

Tip

When the Elastic output plugin fails to connect to the ELK host, it blocks all other outputs and ignores EXPORT_TO_ELASTICSEARCH_BOOL value from env file. Regardless of EXPORT_TO_ELASTICSEARCH_BOOL being False, it tries to connect to Elastic host. The solution - uncomment this portion of config in case of having running ELK.

Kibana settings

1. Index Templates Have been already created by ospf-logstash-index-creator container in compose yaml file. Open Management -> Stack Management -> Index Management ->[ Index Templates ] to make sure that the following templates are in the list:

   • ospf-watcher-costs-changes
   • ospf-watcher-updown-events
     

2. Index Pattern Create indices with the same name as index templates Go to: old ELK Stack Management/ Kibana/ Stack Management/ Index Pattern -> Create index pattern new ELK 8.x Management -> Stack Management -> Index Management -> [ Indices ]

   then Create index

   • ospf-watcher-costs-changes
   • ospf-watcher-updown-events

3. Data View Create data view for two event types. Go to Management -> Stack Management -> Data Views then Create data view

   Name: ospf-watcher-costs-changes
   Index pattern: ospf-watcher-costs-changes
   Timestamp field: use watcher time
   

   
   Repeat the same for ospf-watcher-updown-events As a result, there are two data views should be listed

Tip

What time to use @timestamp or watcher?

It's better to use watcher time, because connection between Watcher and Logstash can be lost, but the watcher continues to log all topology changes with the correct time. When the connection is repaired, all logs will be added to ELK and you can check the time of the incident. If you choose @timestamp

• the time of all logs will be the time of their addition to ELK.

4. Additional checks Make sure that:
   • .env has EXPORT_TO_ELASTICSEARCH_BOOL=True
   • ./logstash/pipeline/logstash.conf has ELK uncommented

Browse your topology changes logs

Your logs are here http://localhost:5601/ -> Analytics/Discover watcher-updown-events.

Zabbix settings

Zabbix settings are available here /docs/zabbix-ui. There are 4 hosts and items (host and item inside each host has the same names) are required:

• ospf_neighbor_up_down
• ospf_network_up_down
• ospf_link_cost_change
• ospf_stub_network_cost_change

WebHook setting

1. Create a Slack app
2. Enable Incoming Webhooks
3. Create an Incoming Webhook (generates URL)
4. Uncomment EXPORT_TO_WEBHOOK_URL_BOOL in .env, set the URL to WEBHOOK_URL

Logs sample 1

2023-01-01T00:00:00Z,demo-watcher,host10.10.10.4,down,10.10.10.5,01Jan2023_00h00m00s_7_hosts,0,1234,192.168.145.5

• 2023-01-01T00:00:00Z - event timestamp
• demo-watcher - name of watcher
• host - event name: host, network, metric
• 10.10.10.4 - event object. Watcher detected an event related to 10.10.10.4 host
• down - event status: down, up, changed
• 10.10.10.5 - event detected by this node.
• 01Jan2023_00h00m00s_7_hosts - name of graph in Topolograph dashboard
• 0.0.0.0 - OSPF area ID
• 1234 - AS number where OSPF is working
• 192.168.145.5 - IP address on detected node Summary: 10.10.10.5 detected that 10.10.10.4 host on the interface with 192.168.145.5 IP address in area 0 in AS 1234 went down at 2023-01-01T00:00:00Z

Logs sample 2

2023-01-01T00:00:00Z,demo-watcher,network,192.168.13.0/24,changed,old_cost:10,new_cost:12,10.10.10.1,01Jan2023_00h00m00s_7_hosts,0.0.0.0,1234,internal,0

• 2023-01-01T00:00:00Z - event timestamp
• demo-watcher - name of watcher
• metric - event name: host, network, metric
• 192.168.13.0/24 - event object. Watcher detected an event related to 192.168.13.0/24 subnet
• changed - event status: down, up, changed
• 10 - old cost
• 12 - new cost
• 10.10.10.1 - event detected by this node.
• 01Jan2023_00h00m00s_7_hosts - name of graph in Topolograph dashboard
• 0.0.0.0 - OSPF area ID
• 1234 - AS number where OSPF is working
• internal - type of network: internal or external
• 0 - subtype of network: type-1, type-2 or 0 for internal subnets Summary: 10.10.10.1 detected that metric of 192.168.13.0/24 internal stub network changed from 10 to 12 at 2023-01-01T00:00:00Z in area 0

Listen-only mode. XDP in action.

If for some reason an extra network is advertised from Watcher, this announcement will be dropped.
Lab schema: there are two wireshark sessions on the interfaces before (on the left side) and after (on the right side) XDP filter.
This examples shows that 8.8.8.8 prefix was redistributed on Watcher and added into its announcement, but it was dropped by XDP and eventually didn't reach the network.
The same logic is applied on Database Description messages and for extra stub networks in LSA1 Update To check XDP logs, run

sudo cat /sys/kernel/debug/tracing/trace_pipe

To check whether XDP filter is assigned on the interface, run

ubuntu20:~/ospfwatcher$ ip l show dev it-vhost1025
178: it-vhost1025@if177: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 xdp qdisc noqueue state UP mode DEFAULT group default
    link/ether aa:c1:ab:e3:cb:d9 brd ff:ff:ff:ff:ff:ff link-netnsid 0
    prog/xdp id 153 <-- !!!

Demo video

To enable/disable XDP

sudo docker run -it --rm -v ./:/home/watcher/watcher/ --cap-add=NET_ADMIN -u root --network host vadims06/ospf-watcher:latest python3 ./client.py --action enable_xdp --watcher_num <num>

Support

Currently XDP was tested on Ubuntu 18,20 Kernel 5.4.0-204-generic. If you faced with XDP errors - skip it while generating config file or use --action disable_xdp as it mentioned in the example above.

Troubleshooting

Symptoms

Networks changes are not tracked. Log file ./watcher/logs/watcher...log is empty.

Steps:

1. Run diagnostic script. It will check OSPF Watcher <-> Network device connection (iptables, packets from FRR/network device)

   sudo docker run -it --rm -v ./:/home/watcher/watcher/ --cap-add=NET_ADMIN -u root --network host vadims06/ospf-watcher:latest python3 ./client.py --action diagnostic --watcher_num <num>
   

2. Login on FRR.

   sudo docker exec -it watcher#-gre#-ospf-router vtysh
   

   show ip ospf neighbor should show network device as a neighbor in the output.

For BGP-LS Mode:

1. Check bgplswatcher container logs:

   sudo docker logs watcher<num>-bgpls-ospf-bgplswatcher

   OSPF Watcher posts topology to Topolograph after BGP session establishment via BGP-LS.

2. Verify BGP session is established using gobgp CLI:

   # List all BGP neighbors with session status
   docker exec -it watcher<num>-bgpls-ospf-bgplswatcher gobgp neighbor
   
   # Show detailed status for a specific neighbor
   docker exec -it watcher<num>-bgpls-ospf-bgplswatcher gobgp neighbor <neighbor-ip-address>
   
   # Check routes received from a neighbor (BGP-LS address family)
   docker exec -it watcher<num>-bgpls-ospf-bgplswatcher gobgp neighbor <neighbor-ip> adj-in -a ls
   
   # Check global RIB (accepted routes)
   docker exec -it watcher<num>-bgpls-ospf-bgplswatcher gobgp global rib -a ls

   The gobgp neighbor command shows:

   • Peer: Neighbor IP address
   • AS: Remote AS number
   • Up/Down: Session uptime
   • State: Session state (Establ = Established, Idle, Connect, Active, etc.)
   • #Received: Number of routes received
   • Accepted: Number of routes accepted

   Run sudo sysctl -w net.ipv4.ip_unprivileged_port_start=0 on the Linux host to allow GoBGP on port 179 if needed to start goBGP in passive mode.

Symptoms

Dashboard page is blank. Events are not present on OSPF Monitoring page.

Steps:

OSPF Watcher consists of three services: OSPFd/FRR [1] -> Watcher [2] -> Logstash [3] -> Topolograph & ELK & Zabbix & WebHooks. Let's start checking each component one by one.

1. Check if FRR tracks OSPF changes in ./watcher/logs/watcher...log file (previous case)
   You should see tracked changes of your network, i.e. here we see that 10.0.0.0/29 network went up at 2023-10-27T07:50:24Z on 10.10.1.4 router.

   2024-07-22T20:24:08Z,watcher-local,network,8.8.0.60/30,changed,old_cost:12,new_cost:-1,10.10.10.5,01Jan2023_00h00m00s_7_hosts,0.0.0.0,65001,external,1
   

2. Check that logstash container from docker-compose.yml is running via docker ps command.

   1. Uncomment DEBUG_BOOL="True" in .env and start continuous logs docker logs -f logstash.
   2. Copy and paste the log from the first step in watcher's log file ./watcher/logs/watcher#-gre#-ospf.ospf.log. docker logs -f logstash should print the output. If not - check logstash container.

3. Check if logs are in Topolograph's DB. Connect to mongoDB and run:

   docker exec -it mongodb /bin/bash
   

   Inside container (change):

   mongo mongodb://$MONGO_INITDB_ROOT_USERNAME:$MONGO_INITDB_ROOT_PASSWORD@mongodb:27017/admin?gssapiServiceName=mongodb
   use admin
   

   1. Check the last two/N records in adjacency changes (ospf_neighbor_up_down) or cost changes (ospf_link_cost_change)

   db.ospf_neighbor_up_down.find({}).sort({_id: -1}).limit(2)
   db.ospf_link_cost_change.find({}).sort({_id: -1}).limit(2)
   

   Sample output:

   { "_id" : ObjectId("67a9ecfe112225e8df6000001"), "graph_time" : "01Jan2023_00h00m00s_7_hosts", "path" : "/home/watcher/watcher/logs/watcher1-gre1-ospf.ospf.log", "area_num" : "0.0.0.1", "event_name" : "metric", 
   

Note

If you see a single event in docker logs logstash it means that mongoDB output is blocked, check if you have a connection to MongoDB docker exec -it logstash curl -v mongodb:27017

ii. Check that graph_time is not empty. If so, check that you can login on the Topolograph page [Login/Local Login] using credentials defined in .env and your local network is added in API/Authorised source IP ranges. Usually, 10.0.0.0/8, 172.16.0.0/12 ,192.168.0.0/16 is enought.

Development

Logstash pipeline development. Start logstash container

[ospf-watcher]# docker run -it --rm --network=topolograph_backend --env-file=./.env -v ./logstash/pipeline:/usr/share/logstash/pipeline -v ./logstash/config:/usr/share/logstash/config ospfwatcher_watcher:latest /bin/bash

Inside container run this command:

bin/logstash

It will expect watcher's log file change, so add new log (copy and paste this line) into ./watcher/logs/watcher#-gre#-ospf.ospf.log file

2023-01-01T00:00:00Z,watcher-local,network,10.1.14.0/24,changed,old_cost:10,new_cost:123,10.1.1.4,01Jan2023_00h00m00s_7_hosts,0.0.0.0,12345,internal,0

The output should be:

[INFO ] 2024-05-13 21:15:25.462 [[main]-pipeline-manager] javapipeline - Pipeline started {"pipeline.id"=>"main"}
The stdin plugin is now waiting for input:
[INFO ] 2024-05-13 21:15:25.477 [Agent thread] agent - Pipelines running {:count=>1, :running_pipelines=>[:main], :non_running_pipelines=>[]}
2023-01-01T00:00:00Z,watcher-local,network,10.1.14.0/24,changed,old_cost:10,new_cost:123,10.1.1.4,01Jan2023_00h00m00s_7_hosts,0.0.0.0,12345,internal,0
{
                            "graph_time" => "01Jan2023_00h00m00s_7_hosts",
                     "event_detected_by" => "10.1.1.4",
                           "subnet_type" => "internal",
                        "int_ext_subtype"=> "0",
                                   "asn" => "12345",
                          "watcher_name" => "demo-watcher",
                          "watcher_time" => "2023-01-01T00:00:00Z",
                            "@timestamp" => 2024-05-13T21:15:50.628Z,
                              "old_cost" => "10",
                              "@version" => "1",
                                  "host" => "ba8ff3ab31f8",
                            "event_name" => "network",
                              "new_cost" => "123",
                          "event_object" => "10.1.14.0/24",
                          "event_status" => "changed"
}

Add your changes in ./logstash/pipeline file, stop logstash process via CTRL+C bin/logstash and start it again. Add the same log in the watcher's log file and check how logstash works with your new changes.

Minimum version

Logstash

7.17.21, this version includes bug fix of issues_281, issues_5115

Note

HTTP output: Logstash vs Fluent Bit — the http plugins use different format options and batching rules. Align the receiver (e.g. Topolograph /websocket) with the body shape you send.

• Logstash (http output): format is one of json (default), json_batch, form, or message. With json, each request body is typically one JSON object per event. With json_batch, each request sends one JSON array containing a whole pipeline batch (common for Logstash-to-Logstash). content_type defaults from format (e.g. application/json for json / json_batch).
• Fluent Bit (http output): Format is one of json (default), json_lines, json_stream, msgpack, or gelf. With json, the body is a JSON array of records for a flush chunk. json_lines is newline-delimited JSON (one object per line). How many records appear in one POST is driven mainly by the engine chunk and the global [SERVICE] Flush interval, not by Logstash’s json vs json_batch switch.

Example Fluent Bit pipeline for this repo: fluentbit/fluent-bit.yaml.

scapy

2.5.0 works, 2.6.0 raises an exception

Topolograph suite

• OSPF Watcher link
• IS-IS Watcher link
• Topolograph link
• Topolograph in docker link

Community & feedback

• https://t.me/topolograph
• admin at topolograph.com

License

GPL-3.0 license Elastic search was used with Basic ELK license.