[This is copied from original page due to some browsers not being able to handle https://.]

Topic: oppless channel fix
Called by: wjr (irc.concentric.net)
Seconded by: riedel (efnet.vuurwerk.nl)
Called on: 04/07/01
Ended on: 04/21/01
Status: PASSED (yes: 14, no: 1, abstain: 3, elig: 29)

Description:


Proposal for EFNet Channel Fixer
William Rockwood, March 28, 2001
Modified April 07, 2001 by William Rockwood



PURPOSE:

Prevent channels which lose ops from remaining in an oppless state.


BACKGROUND:

There have been a number of solutions proposed informally and even
tried on the production EFnet network and each have and have had
their own drawbacks. The common drawbacks among all of the ideas 
involve human resources or the time of operators to actually intervene. 
The reason for this resource drain is different in each case and more 
extreme in two.

The Hybrid 7 implementation of 'vchans' requires an operator to 
create them, due to scaleability concerns if users were to create 
thousands of the same vchan. This creates an unacceptable amount 
of work for the operators and admins who have enough to do already 
with day to day maintenance of their own server, their day jobs, 
family life and what not.

The EFNext (ircd-comstud-based) project implemented OJOIN and specified 
a means of aggregating the channel ops in every channel on the network 
which would be referenced by server operators when requested to assist 
with an oppless channel. The administrative overhead of this would also 
quickly become insane and is unreasonable.

The patch for hybrid (infamously known as "OPME") was released, and run
without a vote on several servers, and technically worked fine, but 
completely overlooked many issues which are raised when restoring ops
to a channel.  Essentially requiring a "leap of faith" on the part of
the oper involved and left no room for accountability should the decision
made in the end have been the wrong one.

Further, there is precedent which shows that operator assistance in any 
of the above ways implies that we as a group of administrators and 
operators sanction the content of any of the channels which opers
"interfere" with. This becomes undesirable when opers are asked to "fix" a 
channel such as #0daywarez or #sexpics (random examples) which 
potentially encourage the trading of illegal material. While opers could 
simply refuse to help these channels, a much simpler, cleaner, fairer 
solution can be managed.

This proposed "channelfix" solution presents a quick fix to the oppless 
channel problem with a "services-like" server which does, on a constant 
basis, what the efnext "snapshot" mechanism does--cull a list of 
chanops, index it and build averages over periods of time. Rather than 
require server operators to be concerned with it, however, channelfix 
operates on the client layer with a "fixerbot" client which it uses to 
restore ops to channels which have become oppless. There are further
potential uses for the channelfix server which are outside the scope
of this document but include "repairing" TS0 channels, DCC functionality
so admins and selected operators could "log in" to the service to watch
chanfix's activities in realtime.


FUNCTIONAL OVERVIEW:

The channelfix server operates as a "services-like" server in that it
links to the network but does not carry clients, or even have the
ability to carry clients.  It merely grabs the 8-10meg burst every
10-30 minutes, distills it to useful information and stores it in
a database to be held for 3-10 days as a historical datagram of
information which it will then self-reference when it detects that
a channel has lost ops.  Once fully functional, channelfix will
join its "chanfix" client to the affected channel, compare the
members of the channel to its stored list of people who have been
opped in that channel and op the 5-most-frequent nicks (by userhost)
and leave the channel, leaving it once again in a state of having
ops.

Here's a mock-up of what might happen:

Pub: #fixtest   stranged log @wjr 
*** #fixtest : End of /NAMES list.
*** Mode change "-o wjr" on channel #fixtest by wjr
*** chanfix (bot@chanfix.efnet) has joined channel #fixtest
*** Mode change "+o chanfix" on channel #fixtest by chanfix.efnet
*** Mode change "+oo wjr stranged" on channel #fixtest by chanfix.efnet
*** chanfix (bot@chanfix.efnet) has left channel #fixtest
Pub: #fixtest   @stranged log @wjr
*** #fixtest : End of /NAMES list.
#fixtest        stranged  G@  ~wjr@armitage.concentric.net (0 *Unknown*)
#fixtest        log       H*  ~wjr@not.real.concentric.net (4 *Unknown*)
#fixtest        wjr       H@  ~wjr@armitage.concentric.net (0 *Unknown*)


As you can see, when wjr deopped himself, chanfix immediately saw it and
remedied the situation. Since this channel was newly-created and wjr was
the only one to ever have ops in the channel, log's userhost was not
included in the op snapshot and thus log was not opped.


TERMS AND LIMITATIONS:

1) The channelfix server will not be able to be used to restore
ops to channels until it has been linked long enough to get enough 
samples to build averages. 3-5 days will be required for this.

2) The channelfix server will not operate when less than 7/8 of
the "full network" is linked--in other words, if 6 servers are split,
and oppless channels exist on the part of the net that the channelfix
server is connected to, it will do nothing, until the servers rejoin.
This is to avoid excessive "fixing" during periods of instability. It
will also not update its "snapshot" database during such times.

3) The channelfix server will log all channels it fixes and who it
opped.  This information will be available (updated nightly) on a 
password-protected ssl webserver and made available to all admins.

4) The channelfix server will only track channels with 8 or more
users in it on a "full net."  Channels with 7 or less users will
need to coordinate the re-creation of said channels on their own.
This will cut down on the database size and the amount of "re-ops"
performed by channelfix tremendously.

5) The channelfix server will only track non-dynamic, idented 
userhosts. A list of dynamic 'keywords' will be used to "weed out" 
hosts to ignore.  This prevents Joe Dialup User from stealing 
John Dialup User's identity in order to take over a channel.

6) The channelfix server will have C/N lines to all hub servers
with full H:.  This allows it to connect anywhere in the event
of a network outage due to attacks or natural catastrophe.


DETAILED FUNCTIONAL SPECIFICATION:

Awaiting coder input. Non-detailed spec should be sufficient for
admins to make a determination whether or not this system is 
appropriate for use on EFnet or not.


Details:

irc.plur.net: YES
irc.emory.edu: YES
irc.lightning.net: ABSTAIN
irc.lagged.org: YES
irc.inter.net.il: ABSTAIN
irc.rt.ru: YES
irc.umn.edu: YES
irc.concentric.net: YES
irc.light.se: YES
irc.du.se: YES
irc.umich.edu: YES
irc.hemmet.chalmers.se: YES
efnet.vuurwerk.nl: YES
irc.ins.net.uk: YES
irc.isdnet.fr: YES
irc.stanford.edu: YES
irc.gigabell.de: ABSTAIN
irc.prison.net: NO


Return to main page.