Collectd: write_prometheus plugin fails with "MHD_start_daemon() failed" (on version 5.7.2.44)

[tcarreira]

• Version of collectd: 5.7.2.44.gbe906f9
• Operating system / distribution: Debian/wheezy

This last update (5.7.2.44.gbe906f9) makes the write_prometheus fail.
On the previous version (I had 5.7.2.37.g45492ba) it worked perfectly.

The logs only show this:

[2017-10-10 18:35:52] plugin_load: plugin "write_prometheus" successfully loaded.
[2017-10-10 18:35:52] write_prometheus plugin: MHD_start_daemon() failed.
[2017-10-10 18:35:52] write_prometheus plugin: MHD_start_daemon() failed.
[2017-10-10 18:35:52] Initialization of plugin `write_prometheus' failed with status -1. Plugin will be unloaded.
[2017-10-10 18:35:52] plugin_unregister_read: No such read function: write_prometheus
[2017-10-10 18:35:52] Error: one or more plugin init callbacks failed.

Steps to reproduce

• activate write_prometheus plugin

I will be glad to help (just don't know what to do)

[octo]

Judging from the commit range, this is a regression introduced in #2466.

[octo]

Hi @tcarreira, thanks for reporting this! Could you please provide the version of libmicrohttpd you're using? E.g. the output of:

apt-cache policy libmicrohttpd-dev

[tcarreira]

Sure!

# apt-cache policy libmicrohttpd-dev
libmicrohttpd-dev:
  Installed: 0.9.20-1+deb7u1

[octo]

Thanks! Could you try running collectd in the foreground (collectd -f) to see if the HTTP library we use logs anything to STDERR?

[tcarreira]

There is nothing relevant in the output:

# collectd -C /etc/collectd/collectd.conf -f
option = Hostname; value = xxxx;
option = FQDNLookup; value = false;
Done parsing `/usr/share/collectd/types.db'
option = Interval; value = 60.000000;
option = Timeout; value = 2.000000;
option = ReadThreads; value = 5.000000;
option = WriteThreads; value = 5.000000;
Created new plugin context.
[2017-10-10 23:07:13] plugin_load: plugin "logfile" successfully loaded.
[2017-10-10 23:07:13] type = logfile, key = LogLevel, value = info

And it is just this.
Even if I change the log level to debug, there is nothing about this error (just what I reported)

[octo]

@tcarreira Thanks for that! It's quite possible that libmicrohttpd only logs anything if the MHD_USE_DEBUG is set. Could you give #2476 a shot to see if that surfaces anything useful?

[tcarreira]

Well, it did change nothing.

(the logfile plugin is sending to STDOUT now, and LogLevel is debug)

# /opt/collectd/sbin/collectd -C /opt/collectd/etc/collectd.conf -f
[2017-10-11 09:38:27] plugin_load: plugin "logfile" successfully loaded.
[2017-10-11 09:38:27] logfile: invalid loglevel [debug] defaulting to 'info'
[2017-10-11 09:38:27] [info] plugin_load: plugin "cpu" successfully loaded.
[2017-10-11 09:38:27] [info] plugin_load: plugin "interface" successfully loaded.
[2017-10-11 09:38:27] [info] plugin_load: plugin "load" successfully loaded.
[2017-10-11 09:38:27] [info] plugin_load: plugin "memory" successfully loaded.
[2017-10-11 09:38:27] [info] plugin_load: plugin "network" successfully loaded.
[2017-10-11 09:38:27] [info] plugin_load: plugin "write_prometheus" successfully loaded.
[2017-10-11 09:38:27] [error] write_prometheus plugin: MHD_start_daemon() failed.
[2017-10-11 09:38:27] [error] write_prometheus plugin: MHD_start_daemon() failed.
[2017-10-11 09:38:27] [error] Initialization of plugin `write_prometheus' failed with status -1. Plugin will be unloaded.
[2017-10-11 09:38:27] [warning] plugin_unregister_read: No such read function: write_prometheus
[2017-10-11 09:38:27] [error] Error: one or more plugin init callbacks failed.
[2017-10-11 09:38:27] [info] Initialization complete, entering read-loop.

[tcarreira]

@octo Would you give this fix a try? https://github.com/collectd/collectd/pull/2481
Looks very simple. Would you be able to hotfix it?

thanks

[octo]

Fixed by #2481, thanks @tcarreira!

[mojili]

It is related to SELinux

[octo]

@mojili can you give us some more context? I was under the impression that #2481 identified and fixed the problem.

Best regards,
—octo