On Thu, Apr 28, 2005 at 11:38:18AM -0700, Phil Mocek wrote:
> After much frustration and reading anything relevant I could
> find short of the source code, I compiled with the
> --with-*portrange options because both the backup server and
> client machine are behind firewalls. I specified a range of 10
> ports because I'm using 10 dumpers (``inparallel=10'' in
> amanda.conf). I've configured both firewalls to allow traffic
> on UDP 850 - 859 and TCP 50000 - 50029.
>
> When I run amdump, it completes backup of the server disks as
> expected, but fails to receive backups from the client. I see
> in the server's firewall logs that the client is attempting to
> connect to the server on UDP ports 591 and 835.
[snip]
Answering my own question:
inparallel should have been set to no larger than half of the
number of ports specified with --with-udpportrange.
Searching amdump's log on the server for mentions of 'port'
turned up the following:
amdump.1:dumper: pid 11850 executable dumper8 version 2.4.4p1, using port
850
amdump.1:dumper: pid 11849 executable dumper7 version 2.4.4p1, using port
859
amdump.1:dumper: pid 11846 executable dumper4 version 2.4.4p1, using port
856
amdump.1:dumper: pid 11845 executable dumper3 version 2.4.4p1, using port
855
amdump.1:dumper: pid 11842 executable dumper0 version 2.4.4p1, using port
852
amdump.1:dumper: pid 11851 executable dumper9 version 2.4.4p1, using port
851
amdump.1:dumper: pid 11848 executable dumper6 version 2.4.4p1, using port
858
amdump.1:dumper: pid 11847 executable dumper5 version 2.4.4p1, using port
857
amdump.1:dumper: pid 11844 executable dumper2 version 2.4.4p1, using port
854
amdump.1:dumper: pid 11843 executable dumper1 version 2.4.4p1, using port
853
amdump.1:planner: time 0.000: bind_portrange: all ports between 850 and 859
busy
I took another look at the PORT.USAGE doc, and saw that I had
misread the following:
Pick the max of (2 * inparallel) and (3 * largest maxdumps).
Allocate at least that many ports in the unprivileged (1024
or larger) range.
Reconfiguring with inparallel set to 5 instead of 10 solved the
problem.
--
Phil Mocek
|