Feature req: have a field/tickbox for "SNMP enabled" on a switch/console-server

[rowanthorpe]

This is not a showstopper, but sometimes people may want to have - for example - terminal servers set up and active but without SNMP enabled (or not reachable by ixp-m). In that case if there is only one peering switch then the snmp-poll cron-job can just specify -p with that switch's name. However it seems that parameter can only be used to specify exactly one switch, or the other option is for all active "switches" (and TSes, etc) to be polled. In the case where there are multiple switches (so -p can't be used) and one or more of the TSes/etc don't respond to SNMP the cron-job has unnecessary waits for timeouts and also spills error output for them which cron ends up emailing to the admin unless ugly ixptool XX 2>&1 | grep -v XX >&2 pipelines are used on the output.

If a field and tickbox could be added along the lines of snmp enabled [x] for each switch/console-server/etc, and the code could grok that before trying to snmp-poll each, that would solve the problem.

Also, allowing multiple switches to be passed to switch-cli via -p option would be great too anyway. If you think this is worth raising as a separate issue, let me know and I will.

[barryo]

Setting the switch type as Console Server should accomplish the same thing if I understand you correctly. I.e. it should solve the underlying issue without the additional feature.

I don't want to add a 'snmp disabled' feature if I can avoid it as the design mindset assumes SNMP is available.

[rowanthorpe]

Our Console-Servers have SNMP-functionality (so presuming console-servers = no SNMP in all cases would be wrong), but in our case we have disabled SNMP because we don't use it on them. Even with their switch type set to Console Server ixpm still tries to access them via SNMP (which would be correct if we had enabled it). So, I guess the design-decision is "do you want IXP-Manager to allow for use-cases like ours, where SNMP is disabled on some hardware?" (and if not we just need to stick with our present workaround of grepping parts of our cron output to /dev/null - which is far from elegant, and doesn't solve the issue of timeout-delays during failed polls).

As for my second point above - it highlights that there would be (at least) two ways to implement it if you decide you want to:

• The tickbox approach, which is more "user-friendly" but would add complexity to the interface
• Adding functionality to the switch-cli command so the -p flag can handle multiple entries, meaning specific switches (plural) can be polled rather than only one. This also solves the main problem but hides it from IXP-Manager's interface.

[nickhilliard]

i feel sure somewhere that a console server isn't a switch, no matter how much it seemed otherwise when we first implemented it like this.

[barryo]

Our Console-Servers have SNMP-functionality (so presuming console-servers = no
SNMP in all cases would be wrong)

Indeed. If you bought such a device with no SNMP, I'd say you'd have been robbed.

What I meant is that IXP Manager, as it stands, doesn't (iirc) care about console servers and SNMP. Or shouldn't. But I was wrong. They will be polled for platform and OS details but not port details.

So, I guess the design-decision is "do you want IXP-Manager to allow for use-cases like
ours, where SNMP is disabled on some hardware?

Let me put it another way: I'm happy to review a pull request that implements this.

[rowanthorpe]

Would you prefer whoever tries to send a PR to code something for the first (interface/tickbox) approach, second (switch-cli) approach, or both approaches? I ask because I think my out-of-practise php-fu would stretch as far as the second approach relatively easily, but the first approach would probably require more free time than I should attempt at the moment unfortunately...

[barryo]

I don't think there's much merit in the second approach to be honest. Everytime a new switch is added, it's an extra manual task to add that switch into the command line cron job. The point of IXP Manager is to stop these manual tasks 😄

It would have to be the second tick box approach. Once you decide to do this, let me know and I'll push a schema change first. You'll have to grep the code for where ever we do SNMP requests.

We may get to it ourselves as it's a fairly easy task. I just have the same issue with time.