savannah-hackers-public
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Savannah-hackers-public] Savannah git repos misbehaving


From: Bob Proulx
Subject: Re: [Savannah-hackers-public] Savannah git repos misbehaving
Date: Sun, 15 May 2016 14:45:26 -0600
User-agent: Mutt/1.5.24 (2015-08-30)

Hi Arnold,

address@hidden wrote:
> I'm seeing things working with HTTP urls and failing with git:// URLs.
> Not just the gawk repo.  It's around 5:40 a.m. east coast time as I
> write this.

Thank you for the updated report.  To help explain things let me
ramble a moment.

Imagine a Venn diagram with multiple circles around different areas.
They overlap.  One is networking.  We know there has been network
problems.  Some of those (dropped packets logged at the managed
switches) seem to have been caused by the kernel change.  However even
though reverting the kernel fixed those the reports of global network
connectivity continue to be reported.  So while one problem may have
been the kernel it was not all of the problems.

Other network changes were implemented.  I only have a small peephole
looking into what changed there.  But it was a large change.

The file systems were repackaged into different volumes as part of the
hardware move.

A long term problem is that the git daemons sometimes get stuck for
unknown reasons.  This has seemingly been exacerbated by the current
network connectivity problems.  When the stuck daemons consume all
available slots then git:// stops working due to what *I* think is a
bug in xinetd that simply closes connections instead of queuing them.
This is why git:// may stop but ssh:// and http:// will continue
working.  git:// uses the git daemon and is also the most popular.
ssh:// uses sshd and http:// uses apache.  So alternate protocols use
different daemons each of which has different slot cap limits.

There are many fewer people accessing source through ssh:// and http://
connectivity than the many people using git:// connectivity.  The many
using git:// may overwhelm the system and worse with the stuck git://
process problem.  That is why the git:// sees the problem the most.
However it is normally the fastest protocol to use and most people
will want to continue to use it.

The immediate need is to solve the recent network connectivity
problems.  These are intermittent.  As you say you were able to access
the machine.  But at the same time others report failures.  Then
longer term we need to upgrade all of the software onto a new system
and that should hopefully solve some of these other problems.

Hope this helps!

Bob



reply via email to

[Prev in Thread] Current Thread [Next in Thread]