difficulties running cf-execd

[LucVereecken]

I'm having difficulties running cf-execd (CFEngine Core 3.21.0). The logfile it tries to create in /var/cfengine/outputs has too long a filename.
This is what shows up in the output:
error: Couldn't open '/var/cfengine/outputs/cf________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________node0420___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________jukinet_fz_juelich_de__1710172596_Mon_Mar_11_16_56_36_2024_0x7ff5ba697740' - aborting exec. (fopen: File name too long)

there are obviously too many underscores in the part where it should just be the fully qualified node name. I couldn't find anything on google about this, and as far as I can tell this is something that occurs inside the cfengine code, so not something I can affect by any configuration setting.

[olehermanse]

Hello @LucVereecken, how did you install CFEngine? Did you build from source? Or installed one of our prebuilt packages? Enterprise or community?

[LucVereecken]

More relevant info: the domain name is set in the promises.cf file, in body common control { domain => "jukinet.fz-juelich.de"; } so the domain name is probably not derived from e.g. the /etc/hostname file, even when that contains the fqdn. The jukinet.fz-juelich.de is not a "real", registered domain, i.e. it doesn't exist in the DNS records, only in static configuration within my internal network.

[olehermanse]

@LucVereecken Fixing the "surface issue" of the long filename should be easy, we can limit the sys.fqhost variable to 256 characters, and also strip the spaces. However, the underlying issue (why there's so many spaces in the domain name) might be harder, could you try to reproduce it on a fresh machine (cloud VM for instance) and send us some instructions for doing so?

[LucVereecken]

@olehermanse These are fresh machines, installed using autoyast and set up using cfengine (running cf-agent as part of the init scripts in the autoyast file). That's quite an elaborate setup to communicate.
In the machine, the domain name is set in the /etc/hostname file, in the /etc/hosts file, and in the sysctl kernel.domainname parameter.
For CFEngine, the domain name is set in the "body common control' in promises.cf (as already mentioned). Interestingly, when bootstrapping, I send an update.cf file that doesn't have the "domain" directive in the body common control, and for those runs a normal output file name is generated (but obviously without a domainname in the file name). This again points to a problem when concatenating the hostname and the domain, with the latter coming from the promises.cf file.
I'll see if i can narrow things down, e.g. a simpler system setup.

[LucVereecken]

@olehermanse I narrowed down in which version of CFEngine the problem starts:
cfengine-community-3.18.0-1.suse15.x86_64.rpm -> has it
cfengine-community-3.15.7-1.suse15.x86_64.rpm -> doesn't have it
I didn't check the 3.16 and 3.17 versions as there are no suse15 rpms available for these non-LTS distributions.

[olehermanse]

I'll see if i can narrow things down, e.g. a simpler system setup.

@LucVereecken Were you able to set up a simpler system and write some instructions, so that we can reproduce the problem on a cloud VM?

[LucVereecken]

@olehermanse I narrowed down in which version of CFEngine the problem starts:
cfengine-community-3.18.0-1.suse15.x86_64.rpm -> has it
cfengine-community-3.15.7-1.suse15.x86_64.rpm -> doesn't have it
I didn't check the 3.16 and 3.17 versions as there are no suse15 rpms available for these non-LTS distributions.

[LucVereecken]

@olehermanse I finally found time to create a test case, using Opensuse Leap 15.5. It' very simple:
Install Leap 15.5:

• standard installation medium as provided by opensuse. I used the large installation medium, not the network based, but I suspect that is not critical.
• Allow online repositories to be added (the proposal is good, i.e. all non-debug repositories). I did this as I
  assume CF-Engine isn't on the install disk
• Role: Standard desktop with plasma
• When configuration options are given, select software, search for “cfengine” and add packages “cfengine” and “cfengine-masterfiles3”. Alternatively, you can just use zypper to install it after the installation. Confirmation will also show a list of dependencies (all selected packages), just accept them.
• I used timezone Europe -> Germany. though I doubt that matters.
• The remaining few questions (user account,...) as as you want.

Once logged in as root, I first did a "cf-agent --bootstrap localhost", but that is not necessary. I then created skeleton files in /var/cfengine/inputs :

File ; /var/cfengine/inputs/promises.cf

body common control
{

version => "0.1";
domain => "jukinet.fz-juelich.de";

bundlesequence => {
       "update"
       };

inputs => {
       "update.cf"
       };

File /var/cfengine/inputs/update.cf :

bundle agent update {

}

As you can see, this actually does nothing. Running cf-promises gives no errors.

Then running "cf-execd –no-fork –once" yields

node0418:/var/cfengine/inputs # cf-execd --no-fork --once
  error: Couldn't open '/var/cfengine/outputs/cf________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________node0418___________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________jukinet_fz_juelich_de__1713453174_Thu_Apr_18_17_12_54_2024_0x7f52692f6740' - aborting exec. (fopen: File name too long)

Then, comment out the "domain" directive, run the cf-execd command again. This will run without errors.

[nickanderson]

I ran into this myself yesterday and I knew it felt familiar. Didn't recall this discussion but @larsewi mentioned it today.

Anyway, I created a ticket for this here: https://northerntech.atlassian.net/browse/CFE-4389 but then I rejected it in favor of https://northerntech.atlassian.net/browse/CFE-4053 which was more specifically about the problem with setting domain in body common control.