multihomed? are the packets coming and going on the same interface? RouterOS will reply with src ip from whatever route it takes, so if you are not using the default gateway you will get no snmp responses from the router (they dont come from the requested ip). i wish you could bind snmp to the loopback ip or something.
I have this snmp issue on devices that both wlan and ether1 are in bridge1, bridge1 having the only IP. The snmp requests come in the bridge interface but then then the device does not reply. It is like the service is down, but routeros provides me with no way to check if the service is functioning and no way to restart it…
you can disable and then enable the snmp service to restart it. do you only have a single ip on that interface? have you run packet sniffer to confirm its not replying but with wrong source?
single ip, torch says no reply whatsoever
so in following these steps it is possible to achieve that snmp service is not working:
- clean router w/o ip address
- enable snmp
- set up bridge with 2 itnerfaces
- set up ip on the bridge interface
- do snmp querry that will fail
is that correct?
That was it!
There was an alternate route which was (accidentally) the default route for traffic from the system we run MRTG on, and once I increased the cost on that route via OSPF to make it prefer the expected route, the router began responding to snmpwalk normally.
Thank you for pointing me in the right direction!
I am also having issues with SNMP on an RB1100AHx2 with 5.20!!!
This works for a short time and stops… (no filters, no firewalls) 
This is a BUG, a ooold bug!
do you have any dynamic routing?
SNMP has problems on RoS if the routing isn’t symmetrical.
I saw a clever “work-around” for it the other day, but can’t recall what it was.
Do some searches on SNMP and symmetrical routing - I think you’ll find it. [Perhaps that isn’t your issue, but something to consider.]
-Greg
[Found it: http://forum.mikrotik.com/t/snmp-doesnt-work-with-asymmetric-routes/58545/1]
pppoe tunnels back to the router the snmp server is located. static route the snmp server’s subnet via pppoe tunnel.
I am also having problems with routerboard1100-ahx2 the snmp it does not work. I’m using version 5.20.
Hey,
I got the same problems discribed here:
http://forum.mikrotik.com/t/snmp-problems-with-5-x/60370/1
as it seems it is a bug.
Pls fix!
Thanks
Simon
I just noticed this problem on ROS 5.24.
nmap of working router -sU -p161 says “161/udp open”, nmap of problem router says “161/udp open|filtered”. No other connectivity problems to router management IP, can winbox and access via port 80 and ping and ssh.
Packet sniff on problem router clarifies that all SNMP packets are being received, no reply packets are being sent. (or possibly it’s trying to send them from some unexpected source IP, I didn’t test for that I filtered by the IP I was directing requests to. Filtering only for udp port 161 floods me with all snmp traffic flowing through this router to other destinations, which is quite a lot. 
)
When I got to the part in this thread about asymmetrical routes, I checked and confirmed problem router had asymmetrical routes. I altered the ingress route (eg, no config changes to problem router whatsoever, just to router closer to host originating snmp get request) to match the egress route, and problem miraculously vanished. Now packet sniff at both calling host and router shows snmp get replies sourced from the same management IP address the get requests are addressed to. Said management IP is bound to a loopback (empty bridge affectionately named “lo”).
While superficially fixed, I still find this bug (cannot talk SNMP over an asymmetrical route) highly inconvenient. I rely on dynamic routing and prefer the freedom to establish asymmetric routes as circumstances call for them. 
Is there any workaround for these or roadmap to mikrotik fix it ?