Open GoogleCodeExporter opened 9 years ago
Apologies, I accidentally hit enter before I had typed the message... and
apparently you can't edit these...
If you specify a DNS name in the hosts list, distcc will attempt to resolve it
for _every_ remote compile... TBH, I think it would be fine if it just resolved
it once, but even if you want to be really pedantic about it, shouldn't it at
least wait for the duration of the TTL? Or a configurable time window? Even in
the best case, that's 20ms of latency for every compile...
I get that distcc is executed anew every time it compiles something, but it
doesn't seem as though it would be so difficult to add some sort of DNS
caching. Would you be interested in this? I'd be happy to write up a patch for
you.
`distcc --version`:
distcc 3.1 x86_64-pc-linux-gnu
(protocols 1, 2 and 3) (default port 3632)
built Mar 24 2012 18:14:36
`gcc --version`:
gcc (Gentoo 4.5.3-r2 p1.1, pie-0.4.7) 4.5.3
`uname -a`:
Linux Beethoven 3.4.0-rc2-amd #8 SMP Sun Apr 8 06:30:22 CDT 2012 x86_64 AMD
FX(tm)-8120 Eight-Core Processor AuthenticAMD GNU/Linux
Original comment by JCalvinO...@gmail.com
on 8 Apr 2012 at 5:46
In my opinion, DNS caching is not something each application should implement
considering something like nscd exists.
A latency of 20ms sounds like you're running neither nscd (part of glibc, gives
me times <0.4ms for most names on repeated queries) nor a caching forwarder in
your LAN (e.g. dnsmasq, gives 2-3ms for all names on repeated queries). There's
also `unscd`, a "simple & stable nscd replacement". I don't use it myself
because I have no problem with nscd.
Measurements were done with:
python -c "from socket import gethostbyname; from time import time;
start=time(); gethostbyname('eris'); stop=time(); print stop-start"
using different host names
To try nscd: "eselect rc start nscd"
Original comment by Domo.Sok...@gmail.com
on 12 Aug 2012 at 12:02
I agree with you in general about applications not caching DNS requests, but I
think this is something of an exception, given the frequency with which the
lookups occur. Yes, I could improve my network's DNS latency, but that's not
the point: I think you would find that in the wild, most people don't have
low-latency DNS situations.
Even .4ms is still a lot of CPU time when added up. In a very large project
(like the kernel), that could easily add up to over a second.
Original comment by JCalvinO...@gmail.com
on 12 Aug 2012 at 5:51
If you use pump mode and set DISTCC_POTENTIAL_HOSTS, which is what I would
recommend, then DNS resolution will only happen once per build, not once per
compile; the pump script will run "lsdistcc" with the "-n" flag to set
DISTCC_HOSTS to the numeric IP addresses of the reachable distcc servers in
DISTCC_POTENTIAL_HOSTS.
Original comment by fergus.h...@gmail.com
on 4 Sep 2012 at 4:20
Where is DISTCC_POTENTIAL_HOSTS described?
Original comment by dave@boostpro.com
on 14 Mar 2014 at 1:40
DISTCC_POTENTIAL_HOSTS is documented in "man pump".
Original comment by fer...@google.com
on 14 Mar 2014 at 1:45
Original issue reported on code.google.com by
JCalvinO...@gmail.com
on 8 Apr 2012 at 5:37