Troubleshooting

Start with the page at the root of the server. It shows each service with the endpointsOne IP address and port that a service gives, with its load balancing weight. See HTTP API. it gives right now, or the error of its sourceWhere a service reads its endpoints, one of dns, file, url or kubernetes. See Services and sources. in red. Then look at the logs of the process, on its standard error.

The server doesn’t start

vigie serve prints vigie: error: followed by the cause, and exits.

Message	Cause	Fix
read config: open <path>: no such file or directory	No file at the path of -c, ./config.yml by default.	Give the path with -c.
read config: services is empty	The file has no services.	Add at least one service.
read config: service dns, file, url and kubernetes are empty, you must specify one of them	A service has no source.	Give each service one source.
load services: service <name>: load kubernetes client config: <err>	The kubeconfig or its context of a kubernetes source can’t be loaded, or Vigie runs outside a cluster without a kubeconfig.	Fix kubeconfig and context, see Read Kubernetes Services.
listen tcp :8080: bind: address already in use	Another process listens on server.listen_addr.	Stop it, or change listen_addr.

Configuration lists the error of each key.

A key has no effect

vigie serve and a reloadReading the configuration file again to replace the services, on SIGHUP or POST /reload. See Reload the services. ignore the keys they don’t know, so a misspelled key keeps its default. Run vigie check-config, which refuses them:

Copy commands

$ vigie check-config -c /etc/vigie/config.yml
vigie: error: invalid config /etc/vigie/config.yml: yaml: unmarshal errors:
                line 2: field listen_adr not found in type config.plain

When the key is right, check that it’s one a reload applies. A change to server or log needs a restart, and so does a change to disable_reload_by_request.

The answer of /services is an error

Status and body	Cause	Fix
401 unauthorized	server.basic_auth is set and the request has no username and password, or wrong ones.	Give the username and password of one of its users, such as url.user_info of a Flock backend.
404 service <name> not found	No service has this name. Without service_name, no service is named default.	Check the name, or reload the file that adds the service.
500 resolve service <name> failed: cannot resolve on lookup ...	The DNS lookup failed or found no record.	Compare with dig, see Read endpoints from DNS.
500 resolve service <name> failed: open <path>: ...	The file of a file source is missing or unreadable by the user of Vigie.	Fix the path or its permissions.
500 resolve service <name> failed: invalid port: ... or invalid weight: ...	A line of a file, a remote list or a TXT record has a port or a weight that isn’t a number.	Fix the line, see Endpoint lines.
500 resolve service <name> failed: bad status code: <code>, <body>	The server of a url source answered with another status than 200.	Check the URL and basic_auth.
500 resolve service <name> failed: get service <namespace>/<service>: ...	The Kubernetes Service doesn’t exist, or Vigie can’t reach the cluster or isn’t allowed to read it.	Check the name and the Role.
500 resolve service <name> failed: list endpointslices of service <namespace>/<service>: ...	Vigie isn’t allowed to list the EndpointSlicesThe Kubernetes object that lists the pods behind a Kubernetes Service, read by the kubernetes source. See Read Kubernetes Services. of the Service.	Add the endpointslices rule to the Role.

A failed read isn’t cached, so the next request tries the source again.

The endpoints are old

Vigie keeps each answer for the cache_ttl of the service, 5 seconds by default, so a change shows up to that time later. A client also polls at its own pace, such as the refresh_rate of a Flock backend, which adds to the delay.

A change to the configuration file itself needs a reload. Without one, Vigie keeps serving the services it loaded.

A reload doesn’t apply

POST /reload answers 422 services not reloaded: with the cause, and a SIGHUP logs it with read config or load services. In both cases Vigie keeps the previous services. Fix the file, check it with vigie check-config, and reload again.

POST /reload answers 404 when the file sets disable_reload_by_request: true. Use SIGHUP instead, see Reload the services.

A client refuses the endpoints

The Flock agent refuses an endpoint whose ip isn’t an IP address. Vigie copies the host of the lines of a file, a remote list or a TXT record without resolving it, so a line with a host name gives an ip that is a name. Write IP addresses in those lines, and an IPv6 address in brackets with its port.