Ugh, sorry about the formatting.
See it here instead: http://sxi.dk/tmp/provision-code.txt

The custom provisioner used here is really simple. It iterates over my CMDB, fetches hostnames and IP-addresses and provides the following XML output:

<?xml version="1.0" encoding="UTF-8" standalone="yes"?>
<model-import non-ip-snmp-primary="N" non-ip-interfaces="false" last-import="2013-01-28T09:32:47.801+01:00" foreign-source="niacpe" date-stamp="2013-01-28T09:32:47.472+01:00" xmlns="http://xmlns.opennms.org/xsd/config/model-import">
<!-- per node: -->
<node node-label="a.fqdn.hostname" foreign-id="same.fqdn.hostname">
<interface snmp-primary="P" status="1" ip-addr="172.31.xx.yy" descr="Mgmt">
<monitored-service service-name="ICMP"/>
<monitored-service service-name="SNMP"/>
</interface>
<category name="CPE"/>
<category name="C-CUSTOMERNAME"/>
</node>
</model-import>

The relevant XML-building ruby code follows:

builder = Nokogiri::XML::Builder.new do |xml|
xml.send(:"model-import", 'foreign-source' => 'niacpe', 'date-stamp' => Time.now.xmlschema(fraction_digits=3), 'xmlns' => "http://xmlns.opennms.org/xsd/config/model-import") do
pnodes.each do |hostname, values|

1. We try to guess customername
   if hostname =~ /^(cpe|ce)/i
   customername = hostname.split('.')[1]
   else
   customername = hostname.split('.')[0]
   end
   if hostname =~ /(N\w{2}-?\d{6})/
   crid = $1
   else
   crid = nil
   end

xml.node('foreign-id' => hostname, 'node-label' => hostname) do

values[:ips].each do |ip|
xml.interface('status' => '1', 'snmp-primary' => ip[:status], 'ip-addr' => ip[:ip], 'descr' => 'Mgmt') do
xml.send(:"monitored-service", "service-name" => "ICMP")
xml.send(:"monitored-service", "service-name" => "SNMP")
end
end
xml.category('name' => 'CPE')
xml.category('name' => "C-#{customername.upcase}") # Changeme!
if crid
xml.asset('name' => 'circuitId', 'value' => crid)
end

end
end
end
end

Hi there,

I've started to roll out OpenNMS again and I have yet again run in to this issue.

The OpenNMS version is now 1.10.7 and I'm running provisiond with a custom adapter that produces a requisition XML file from my CMDB.

Two steps to reproduce this goes like this:

1) On a node with outages, forcibly unmanage the interface and services. This puts ismanaged = 'F' on the ipinterface table for the interface.
2) Sync the requisition.
3) ismanaged is now 'M' on that interface.

Further observations:
The availability of the node and interface is reset to 100%. No polling seems to happen to the interface, although the ip address is unreachable.

Provisiond reports the following upon scanning the node:
The Node with Id: 4134; ForeignSource: niacpe; ForeignId:<redacted> has aborted for the following reason: Aborting node scan : Agent timed out while scanning the system table.

No outage is reported for the interface or node, although it should report a uei.opennms.org/nodes/nodeDown event as no interfaces are reachable for the given node.

Further analysis:

As the requisition reports the interface as 'P', I could suspect that the re-management behaves as it does by design, however the 100% availability and no further polling issue observed does not strike me as working as designed. This is, perhaps, the fixable issue here.

One possible option for me would be to read the ipinterface table on reprovisioning and simply removing information about interfaces with ismanaged == 'F'. Would this be considered best practice by the OpenNMS designers?

Please do not hesitate to contact me about further information if required.

Best,

Allan

Sorry for the lack of feedback, but I do not work for the company using this feature anymore, so I am unfortunately not able to test it.

No feedback from the user, I'm going to assume that the fixes to https://opennms.atlassian.net/browse/NMS-3906#icft=NMS-3906 have fixed the issue sufficiently.