On Wed, 13 Jun 2001, Joseph Mack wrote:
> OK have nolb working with a client box and 3 of my 4 servers. 3 of
> the servers appear to be executing the different phases just fine.
>
> I don't know what the problem with the 4th server is. The 4
> servers were setup at different times but are approximately
> similar in software (kernel, compiler, glibc versions etc) and
> identical in hardware.
>
> I can polyprobe between the client and the 4th server and get the
> expected throughput. The server when running as a nolb server is
> listening on 8080 (my polygraph port) by looking with netstat -an
> and I can connect to it with telnet from the client. The telnet
> connection hangs (I can't send "quit" ^C, ^D etc).
Keep in mind that you have to use HTTP if you telnet to an HTTP
server. "quit" ^C, ^D are not HTTP commands. The server will do
nothing until it thinks it has HTTP header or times out waiting for
one. You can type a sample request followed by an empty line or do
Cntrl+] followed by "close" to close the session.
> When I fire up the nolb client, it has hundreds of connections in
> the SYN state (by netstat -an) to the 4th server, while the
> connections to the other 3 servers are in more normal states
> (ESTABLISHED etc).
>
> At the bottom is the output of nolb startup on the 4th server if
> you can give me any ideas as to what's happening.
Sorry, cannot see anything wrong with it. Based on your description,
it smells like a routing problem rather than Polygraph
mis-configuration, but cannot say for sure.
Make sure you can telnet *from the client alias* to the server
address. While doing that, You might want to run tcpdump on the server
side to see if the server is getting any packets from the client.
> On a slightly different topic, I'm trying to see if my L4 box
> fits your test.
>
> In the setup I'm running with nolb-14-4.pg, I have one client
> which knows about 4 servers, all on different IPs, and on the
> same network.
>
> My L4 box expects the clients to access the servers through the
> IP on the outside of the L4 box (which I call the VIP). The
> servers behind the L4 box have their eth0 devices on private
> addresses and are either listening on those addresses (different
> for each server) or on the VIP (which is the same for all servers,
> and is bound to a non arping device eg lo:0). In the first
> case the requests are reverse masqueraded into the servers and
> in the 2nd case the requests are routed to the VIP on the server
> through each server's eth0 IP and the replies come back directly
> to the client.
>
> No matter which way I do it, the client sees only 1 IP no matter
> how many servers are behind the L4 box and all the servers are on
> a different network to the client (client comes from a public
> IP, servers are on private IPs, L4 box has the public IP=VIP, that
> a single server would have if there was only a single box as
> server).
>
> I know little about other L4 boxes. To me if 4 servers each have different
> IP's and a client has a table of 4 IP's to connect to, then the
> only thing a box in the middle can do is regular routing.
>
> Is the setup I have testable in the switch-off?
I might be misunderstanding your description, but it seems to me that
you are on the right track. For non-DSR configuration, you have to
tell the clients that VIP is their origin server and then map VIP to
real server addresses using AddrMap. The srvlb-l4-4.pg file does that
(but as you have learned already, you need to set vipMap.addresses to
TheBench.server_side.hosts to avoid aliasing on the server side).
I have not tried a DSR configuration myself yet, but I suspect that
the only difference, from Polygraph point of view, would be that
TheBench.server_side.hosts are set to VIP rather than individual
[unique] server addresses. In other words, from PGL point of view,
there will be only one origin server, residing at VIP.
If this is unclear or confusing, lets share PGL configurations
off-line so that we can help you to build the "right" setup.
Thanks,
Alex.
This archive was generated by hypermail 2b29 : Tue Jul 10 2001 - 12:00:19 MDT