Question from the audience at Empower in London:
- Can I have all my third party devices send SNMP traps to one IP address (one DataMiner node), and will DataMiner then forward it from one node to the other, so that it ends up with the proper associated element for each specific trap (irrespective of which node that element is hosted on)?
- When the swarming is released, will this then also work if the above is possible?
- A single destination IP address is very convenient (as this has to be configured on the various involved third party products, and you do not want to change that), but it then also represents a single point of failure. Would it still be possible to use on that node a classic main / backup node (with virtual IP address moving in between them) to deal with that, in combination with the upcoming swarming leveraged across the other nodes?
Or any other recommended way to deal with SNMP traps?
Hi,
Yes, traps will distribute automatically within a DMS to the correct node, so it doesn't matter to which node you send the trap and where the element resides. It is possible to disable this, but it's enabled by default: DataMiner.xml | DataMiner Docs
Trap distribution should also work in combination with swarming when that's being released. This still needs to be validated at this point, but I see no reason why this combination would pose any issues.
Swarming cannot be combined with failover. That is not supported. If you want to have an IP address to send your traps to independent of DataMiner, you could look at some kind of load balancing in front of DataMiner or maybe e.g. a public IP in Azure, where you define the accepted source IPs from where you except traps, and then assign this IP to one or another DMA.
Bert
This has been a question in the back of my head for ages.
Can the traps be sent to multiple nodes to cover redundancy, without causing any type of conflict receiving duplicates?
Hi Philip,
Sending the same trap to multiple DMAs is not something we’ve being testing and validating, therefore I probably need to say it’s officially not supported…
Now, that being said, I would expect DataMiner not to care about this. Whenever a trap comes in, it will look in the DMS to which element this needs to be delivered. One DMA does not know what the other DMAs received already. Then, when the trap is being delivered to the element, the protocol defines what needs to happen with the trap and there are many possibilities… Therefore it probably depends on the protocol if receiving the same trap twice is supported or not. So, again this would need to be validated and this for each protocol receiving traps, but I would assume this can work if the handling of the traps in the protocol is done in a robust way, and the trap content allows it to do that (we would need know somehow that it’s the same trap somehow).
So, I would say it’s possible, but it would need to be tested and validated with the protocol you need.
Bert
If we can send the SNMP trap to any DMA, would this be a use case for Anycast, as SNMP uses UDP and is stateless? Could you add the same /32 IP to each DMA, then use your favourite routing protocol to update the routing tables, Then failover wouldn't be an issue.
Interesting idea! I would indeed expect this to work like you explain it, but I’m not immediately aware of anybody setting this up in that way. But it sounds feasible.
Perfect! Thanks for the feedback.