- Notifications
You must be signed in to change notification settings - Fork5
skoef/birdwatcher
Folders and files
Name | Name | Last commit message | Last commit date | |
---|---|---|---|---|
Repository files navigation
healthchecker forBIRD-anycasted services
bird1
branch.
This project is heavily influenced byanycast-healthchecker. If you want to know more about use cases of birdwatcher, please read their excellent documention about anycasted services and how a healthchecker can contribute to a more stable service availability.
In a nutshell: birdwatcher periodically checks a specific service and tells BIRD which prefixes to announce or to withdraw when the service appears to be up or down respectively.
When I found out about anycast-healthchecker (sadly only recently onHaproxyConf 2019), I figured this would solve the missing link between BIRD and whatever anycasted services I have running (mostly haproxy though). Currently however in anycast-healthchecker, it is not possible to specify multiple prefixes to a service. Some machines in these kind of setups are announcingmany/32
and/128
prefixes and I ended up specifiying so many services (one per prefix) that python crashed giving me atoo many open files
error. At first I tried to patch anycast-healthchecker but ended up writing something similar, hence birdwatcher.
It is written in Go because I like it and running multiple threads is easy.
The notation oftimeout
for services changed from int (for number of seconds) to the formattime.ParseDuration
uses, such as "300ms", "-1.5h" or "2h45m".
For example:10
should become"10s"
in your config files.
This simple example configures a single service, runshaproxy_check.sh
every second and manages 2 prefixes based on the exit code of the script:
[services] [services."foo"]command ="/usr/bin/haproxy_check.sh"prefixes = ["192.168.0.0/24","fc00::/7"]
Sample output in/etc/bird/birdwatcher.conf
ifhaproxy_check.sh
checks out would be:
# DO NOT EDIT MANUALLYfunction match_route() -> bool{return net ~ [192.168.0.0/24,fc00::/7];}
As soon as birdwatcher finds out haproxy is down, it will change the content in/etc/bird/birdwatcher.conf
to:
# DO NOT EDIT MANUALLYfunction match_route() -> bool{return false;}
and reconfigures BIRD by givenreloadcommand
. Obviously, if you have multiple services being checked by birdwatcher, only the prefixes of that particular service would be removed from the list inmatch_route
.
Integration in BIRD is a matter of including/etc/bird/birdwatcher.conf
(or whatever you configured atconfigfile
) in the configuration for BIRD and use it in a protocol like this:
protocol bgp my_bgp_proto { local as 12345; neighbor 1.2.3.4 as 23435; ... ipv4 { ... export where match_route(); } ... ipv6 { ... export where match_route(); }}
Configuration section for global options.
key | description |
---|---|
configfile | Path to configuration file that will be generated and should be included in the BIRD configuration. Defaults to/etc/bird/birdwatcher.conf. |
reloadcommand | Command to invoke to signal BIRD the configuration should be reloaded. Defaults to/usr/sbin/birdc configure. |
compatbird213 | To use birdwatcher with BIRD 2.13 or earlier, enable this flag. It will remove the function return types from the output |
Each service under this section can have the following settings:
key | description |
---|---|
command | Command that will be periodically run to check if the service should be considered up or down. The result is based on the exit code: a non-zero exit codes makes birdwatcher decide the service is down, otherwise it's up.Required |
functionname | Specify the name of the function birdwatcher will generate. You can use this function name to use in your protocol export filter in BIRD. Defaults tomatch_route. |
interval | The interval in seconds at which birdwatcher will check the service. Defaults to1 |
timeout | Time in which the check command should complete. Afterwards it will be handled as if the check command failed. Defaults to10s, format following that oftime.ParseDuration . |
fail | The amount of times the check command should fail before the service is considered to be down. Defaults to1 |
rise | The amount of times the check command should succeed before the service is considered to be up. Defaults to1 |
prefixes | Array of prefixes, mixed IPv4 and IPv6. At least 1 prefix isrequired per service |
Configuration for the prometheus exporter
key | description |
---|---|
enabled | Boolean whether you want to export prometheus metrics. Defaults tofalse |
port | Port to export prometheus metrics on. Defaults to9091 |
path | Path to the prometheus metrics. Defaults to/metrics |
About
healthchecker for BIRD-anycasted services
Topics
Resources
Uh oh!
There was an error while loading.Please reload this page.
Stars
Watchers
Forks
Packages0
Uh oh!
There was an error while loading.Please reload this page.
Contributors2
Uh oh!
There was an error while loading.Please reload this page.