inputs.phpfpm exit if unix socket doesn't exist

[antitbone]

Relevant telegraf.conf

[[inputs.phpfpm]]
  urls = ["/usr/local/php-7.4/var/run/7.4-default.sock:fpm_status","/not_exist"]
  fieldpass = [
    "active_processes",
    "idle_processes",
    "total_processes",
    "max_active_processes",
    "accepted_conn",
    "listen_queue",
    "max_listen_queue",
    "slow_requests"
  ]

Logs from Telegraf

2023-11-08T14:39:41Z E! [inputs.phpfpm] Error in plugin: socket doesn't exist "/not_exist"

System info

Telegraf 1.23.4, CentOS Linux release 7.4

Docker

No response

Steps to reproduce

1. start from a configuration with a single unix socket.

[[inputs.phpfpm]]
  urls = ["/usr/local/php-7.4/var/run/7.4-default.sock:fpm_status"]

the metrics are collected correctly

$ telegraf --test /etc/telegraf/telegraf.conf | grep phpfpm
2023-11-08T14:39:56Z I! Using config file: /etc/telegraf/telegraf.conf
> phpfpm,host=tutu.com,pool=7.4-default,url=/usr/local/php-7.4/var/run/7.4-default.sock:fpm_status accepted_conn=622065i,active_processes=4i,idle_processes=32i,listen_queue=0i,max_active_processes=35i,max_listen_queue=0i,slow_requests=0i,total_processes=36i 1699454396000000000

2. add a non-existent unix socket in the socket list of the inputs.phpfpm plugin

- urls = ["/usr/local/php-7.4/var/run/7.4-default.sock:fpm_status"]
+ urls = ["/usr/local/php-7.4/var/run/7.4-default.sock:fpm_status","/not_exist"]

3. the plugin reports an error for the non-existent socket but no longer reports those for the functional socket.

2023-11-08T14:39:41Z E! [inputs.phpfpm] Error in plugin: socket doesn't exist "/not_exist"

$ telegraf --test /etc/telegraf/telegraf.conf | grep phpfpm
> nothing

Expected behavior

The phpfpm plugin should collect the functional socket metrics even if some are inaccessible

Actual behavior

the phpfpm plugin seems to exit at the first non-existent socket

urls = ["/usr/local/php-7.4/var/run/7.4-default.sock:fpm_status","/not_exist","/not_exist2"]

> 2023-11-08T14:55:37Z E! [inputs.phpfpm] Error in plugin: socket doesn't exist "/not_exist"

urls = ["/usr/local/php-7.4/var/run/7.4-default.sock:fpm_status","/not_exist2","/not_exist"]

> 2023-11-08T14:55:13Z E! [inputs.phpfpm] Error in plugin: socket doesn't exist "/not_exis2t"

If the file exists but is not a socket or does not have the correct rights, accessible sockets metrics remains functional.

2023-11-08T14:58:35Z E! [inputs.phpfpm] Error in plugin: dial unix /not_exist: connect: connection refused

2023-11-08T00:51:00Z E! [inputs.phpfpm] Error in plugin: dial unix /not_exist: connect: permission denied

The non-existence of the socket should not completely stop execution on accessible sockets

Additional info

No response

[powersj]

the phpfpm plugin seems to exit at the first non-existent socket

That is correct, inside the Gather function, any error during the call to expandUrls will exit. You are currently hitting the socket does not existing during globUnixSocket. Which causes gather to exit.

In general, if telegraf cannot connect to an external service, whether that is a cloud service, socket, IP address, hostname, etc. then we fail. To some users this may indicate a typo, an incorrect authentication, or a real network issue. Rather than acting that everything is ok, giving the user the false sense that metrics will correctly be collected, we error.

We have a number of feature requests open to allow Telegraf to continue on connection failures like this. I am happy to see a PR that changes the logic, with opt-in via a configuration option.

[srebhan]

@antitbone can you please test the binary in PR #14852 available as soon as CircleCI finished all tests!? Let me know if this fixes the issue!