Ethernet Switching
Reply
Trusted Contributor
wsanders
Posts: 88
Registered: ‎08-10-2009
0

EX4200 DHCP service flaky

[ Edited ]

We've set up several virtual chassis of EX-4200s. The largest is 5 FPCs, with about 190 ports active over five VLANS (each with its own routable interface). We are finding that using JunOS as the DHCP server on this big stack is very unreliable. About once a week, the DHCP service gives up handing out IP addresses when leases expire, and no one connected to the VC can get an IP address. Snooping the network shows that the EX is just not responsing to the DHCP discover or request. The requests appear to be valid.

 

Restarting the DHCP service always fixes this problem. (the dhcpd does not die, it just becomes unresponsive to all DHCP discovers and requsts.) The problem is worst on the large VC, but DHCP is flaky on smaller, 1 and 2-FPC routers.

 

We've had a case open for a while but have made no progress. (They keep asking for a "network diagram". well, I have 190 hosts connected to 190 interfaces, that's it. I just escalated the case since progress seems to be stalled.)

 

We have DHCP traceoptions enabled "level info; flag all" (the undocumented "level" command seems to be necessary to get DHCP to log anything.) The only interesting things I see getting logged are clumps of these messages that appear only when the DHCP service is dead:

 

Dec 15 15:04:03 max message size is 1260 bytes
Dec 15 15:04:03 max message size is 1260 bytes
Dec 15 15:04:04 max message size is 1260 bytes
Dec 15 15:04:06 max message size is 1260 bytes
Dec 15 15:04:08 max message size is 1260 bytes
Dec 15 15:04:10 max message size is 1260 bytes

 

Another thing I see is that while I have 147 files in /var/db/leases there are only 10 or 20 leases in the output of "show system serv dhcp binding". In fact, "show system serv dhcp binding" prints nothing after a DHCP restart, even though the dhcpd log says dhcpd is reloading all the leases from files. So maybe dhcp is losing track of the lease files?

 

Does this set off any alarms for anyone reading this?

 

Is anyone else having problems with the JunOS DHCP servcice, and did you do anything to resolve it?

 

 

W Sanders
System and Network Administrator
St Marys College of California
Trusted Contributor
wsanders
Posts: 88
Registered: ‎08-10-2009
0

Re: EX4200 DHCP service flaky

We were seeing these in the DHCP logs:

 

Dec 16 09:30:11 skipping option code 1, not found in scope

 

DHCP Option 1 is the subnet mask, a rather important parameter :-) For some reason DHCP was deciding not to send it out in some cases. Normally, the dhcpd seems to deduce the submet mask from the mask of the interface the packets are exchanged on.

 

As a workaround I explicitly added a submet mask to the HDCP configuration:

 

option 1 ip-address 255.255.255.0;

 

and the log message has not recurred. We are still having problems, though, perhaps due to some bad leases in /var/db/leases made before I performed the workaround.

 

BTW release if 9.5R2.7. We plan to upgrade once the TAC advises us as to which relase might offer the most reliable DHCP service.

 

 

 

 

 

 

 

 

W Sanders
System and Network Administrator
St Marys College of California
Copyright© 1999-2013 Juniper Networks, Inc. All rights reserved.