Hi Fajar, On Tue, Feb 16, 2010 at 1:16 PM, Fajar A. Nugraha <fajar@fajar.net> wrote:
On Tue, Feb 16, 2010 at 6:09 AM, Phil Pierotti <phil.pierotti@gmail.com> wrote:
Tue Feb 16 09:40:25 2010 : Proxy: Marking home server 192.168.147.2 port 1813 as zombie (it looks like it is dead).
There should be other things before that
Yes, I agree, there *should* have been something more than that. I pored over the log, carefully, for a good while. Everything "looked normal" (get a request, process it, proxy it, get reply, send it back, lather/rinse/repeat). Nothing at all looking even slightly like "something is wrong", until that message in the log.
Sending Accounting-Request of id 228 to 192.168.147.2 port 1813 User-Name := "-------------@-------------" Acct-Status-Type := Stop Acct-Session-Id := "00000000" Event-Timestamp := "Feb 16 2010 09:40:25 EST" NAS-Identifier := "Status Check. Are you alive?" Tue Feb 16 09:40:25 2010 : Debug: Waking up in 0.7 seconds. rad_recv: Accounting-Response packet from host 192.168.147.2 port 1813, id=228, length=20 Tue Feb 16 09:40:25 2010 : Proxy: Received response to status check 34 (1 in current sequence)
Like that one. That particular status check was completed immediately. How were other status check responses, do they arrive on time? How
"on time" is subjective, but every status-check I saw came back within the same second. the log has no finer granularity. I would not be surprised if this is a case of "happy" replies are instant, but anything with a problem is lagging. status-check is a known-good condition (at least the user/pass) so it always succeeds, and is always fast.
about actual accounting request, do they get a timely response? It is
It could easily be that the downstream server is lagging in responsiveness , given that it's a db backend. Best-case is snappy, worst-case is abysmal is not at all surprising with a db. But the question is how long before "timely" runs out? One second, ten seconds, half-a-second? Where (other than reading every single line of a debug log for an entire day) can I find how happy (or not) freeradius is about a server it is proxying to? This is a live radius proxy for a small ISP, not just a console auth-server, so we're seeing anything up to ten requests per second - not lots-n-lots, but also not practical to eyeball the entire thing in realtime. Spot-checks are fine, but if nothing broke while you were checking then it's "tree falls in a forest" time. Thanks, Phil P