[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

Re: svnserve under Linux inetd hangs, burning CPU cycles, under too low TCP-sendbuffer.

From: Neels J Hofmeyr <neels_at_elego.de>
Date: Sat, 27 Feb 2010 00:13:17 +0100

Hi Andreas,

thank you for reporting this problem. I am sorry to see that you have
received no helpful replies yet. That's probably because many of us are tied
up on the road to 1.7.

Let me assure you that your mail has arrived in our inboxes. I hope someone
familiar with svnserve can reproduce and analyse the issue. If you don't get
an answer by next week, please poke this list again!

If you do, it would help if you could provide, ideally, a runnable script
that reproduces the problem.

Thanks again for your efforts, we welcome your help!

Dr. Andreas Krüger wrote:
> Hello,
>> He additionally reports that this is reproducible not only with
>> 1.5.1, but with Debian's build of 1.6.9. (That mail is probably in
>> the moderation queue now.)
> thank you Peter! (To me, this is the first indication that someone
> actually noticed my attempts to tie that bug down where you folks can
> hit it.)
> Peter refers to my attempt to ask the Debian bug tracking system to
> forward my Debian bug report to this list. Nice try on my part... As
> that doesn't look like it's getting anywhere, here comes a manual copy.
> The new information in a nutshell:
> * Reproducible with Debian's unstable "1.6.9dfsg-1" version, as Peter
> mentioned,
> * In all my experiments thus far, the client always has been Ubuntu's
> 1.6.5dfsg-1ubuntu1.
> Below comes the whole story as available in the Debian bug tracking
> system. Basically, this is a summary of what I know at this point.
> Turn to http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=571457 to
> follow Debian developments.
> So far, so good. As for me, I intent to sit back and relax until
> either someone asks me for further information or else until I receive
> clearance to introduce this to Subversion's bug tracking system.
> Regards, and thank you all for providing fine software,
> Andreas
> ======================================================
> Package: subversion
> Version: 1.6.9dfsg-1
> X-Debbugs-CC: dev_at_subversion.apache.org
> Hello, subversionists,
> I have a subversion repository with a single revision, consisting of a
> single roughly 10 MB file with random data.
> I use svnserve to serve that repository.
> I use a client to access the svnserve over the network, via "svn co".
> (The client happens to be Ubuntu's 1.6.5dfsg-1ubuntu1, in case that
> matters.)
> The server is an openvz guest. When, on the corresponding host, I set
> vzctl set NNN --tcpsndbuf 415k:715k
> or lower (my guess is the first number is the one that matters), the
> checkout starts, there is some initial network traffic, but then
> svnserve starts to eat up all CPU cycles it can get and no further
> progress is made, until I kill the svnserve process manually on the
> server. Just killing the client does not seem to stop the waste of CPU
> cycles on the server.
> When I increase the tcpsndbuf above that value, there is no problem.
> However, when I check in a much larger file into the repository (I do
> that directly, using file:/// - access), and try to check out again
> (using svnserve), the problem reappears.
> This is quite reproducibly, with svnserve as a standalone daemon as well
> as svnserve running under inetd. For the latter case, I made no precise
> experiments about the --tcpsndbuf numbers.
> This is the summary. I have written three mails yesterday and today to
> the dev_at_subversion.apache.org mailing list, but have not received any
> answers there yet. There is lots of nitty-gritty background information
> in those mails, which, hopefully, you'll be able to find here:
> http://mail-archives.apache.org/mod_mbox/subversion-dev/201002.mbox/browser
> Regards, and thank you for providing fine software,
> Andreas
> P.S.: I ask the Debian bug tracking system to forward a copy of this
> mail to the SVN dev mailing list. This contains two pieces of
> information not available in my previous mails to that list:
> * The bug is reproducible with the latest subversion software version I
> can easily install, short of compiling myself.
> * The client I'm using is, and always has been, 1.6.5dfsg-1ubuntu1. I
> think I misquoted it to be 1.5 in one of my earlier mails. I apologize
> if this caused any confusion!

Received on 2010-02-27 00:13:57 CET

This is an archived mail posted to the Subversion Dev mailing list.