Grex Oldcoop Conference

Item 315: VOTE PROPOSAL: Remove the idle daemon.

Entered by cross on Mon Feb 20 06:44:57 2006:

I propose that a vote be put to the membership, as suggested by mcnally in
agora, to remove the idle daemon.  The reason being that the idle daemon
serves no purpose anymore, since (a) grex is not resource constrained for
pty's or CPU/memory capacity, and (b) the benefit it has of preventing
preventing obnoxious behavior is easily subverted, and (c) there is no
evidence that it stops anybody from doing anything anyway.
88 responses total.

#1 of 88 by mcnally on Mon Feb 20 07:49:25 2006:

 I'm not a voting member, which means it's probably not my prerogative
 to second cross's proposal that the matter be put to a membership vote,
 but I do think it would be a good idea to settle the matter by a vote.


#2 of 88 by naftee on Mon Feb 20 15:37:54 2006:

i vote yes


#3 of 88 by krj on Mon Feb 20 15:40:40 2006:

What benefit is to be gained by eliminating the idle daemon?


#4 of 88 by nharmon on Mon Feb 20 16:24:12 2006:

Well, some of us like to sit in party waiting for someone to come along
to talk to. It becomes annoying when you have to re-login every 15
minutes or so.


#5 of 88 by cross on Mon Feb 20 16:32:42 2006:

It provides a minimal increase in resources to other user processes, but
that's not terribly significant.  However, I think a better question is,
what benefit is there to retaining it?  I can't see any, and it's kind of
annoying to get kicked off every 20 minutes.  Even if you `touch' your
terminal every few minutes, you still get kicked off after 10 hours,
regardless if whether you're busy or not.


#6 of 88 by nharmon on Mon Feb 20 16:39:30 2006:

Grex should only restrict things insomuch as they are necessary to
keeping Grex safe. IMHO.


#7 of 88 by naftee on Mon Feb 20 16:42:11 2006:

I'm going to argue that a 15 minute limit (which is VERY short ; m-net's used
to be an hour) will actually encourage people to use idle daemon killers. 
these small programmes put a stress on the system load, and would obviously
not be needed for a system with no idle daemon.


#8 of 88 by kingjon on Mon Feb 20 18:38:22 2006:

Another resource that the idle daemon helps conserve is network bandwidth, at
least for users who come in over the net. I've noticed at least once in the
past week a time of definite slowdown, which may come from CPU or network load,
so I think we're not totally past the days of limited resources. 

(One point: I believe it's actually twenty (20) minutes: fifteen (15) before
the first warning, and then disconnect after five more.)


#9 of 88 by nharmon on Mon Feb 20 18:40:46 2006:

How much bandwidth does an inactive login use? Not very much, I would
say. Maybe some packets every few minutes to make sure the connection is
still active?


#10 of 88 by mcnally on Mon Feb 20 18:55:54 2006:

re #8:  Bandwidth and CPU utilization from idle users is so negligible
it wouldn't be noticable if you had 1000 users with idle sessions open.


#11 of 88 by kingjon on Mon Feb 20 19:49:10 2006:

Re #10: Is the bandwidth/CPU usage that would occur from idle users if the
daemon were turned off less or more than the resources the daemon currently
uses?



#12 of 88 by mcnally on Mon Feb 20 20:33:29 2006:

 I don't know.  How many idle users would we have if we turned the daemon
 off?  There's no real way to tell except by trying it.

 It's not like one has to make an irrevocable decision on the issue.  
 It could be turned off with an understanding that in a month or two
 if people aren't happy with the results it could be put to another
 vote..  If people are really concerned that it will really hurt the
 system direction can be given to staff to turn it back on if problems
 start developing. 

 In my opinion, however, the thing to keep in mind, is that ideally
 direction on policy changes should come from the board or from the
 members, not from staff, and that continuing to do something the way
 we've done it just because we've always done it that way and don't
 know what the alternative would be like isn't a sound strategy for
 directing system policy.


#13 of 88 by krj on Mon Feb 20 21:29:09 2006:

Doesn't today's idle timer crudely enforce the policy against 
(outgoing) bots and servers, by periodically logging out users and 
thus killing all their processes?


#14 of 88 by mcnally on Mon Feb 20 21:34:19 2006:

 I'm not sure to what extent that's controlled by "robocop" and to
 what extent by "idled"


#15 of 88 by kingjon on Mon Feb 20 21:53:24 2006:

#14: As I understand it, robocop only kills orphan processes, waiting for idled
to kill the idle parents (the shell).



#16 of 88 by cross on Tue Feb 21 00:03:53 2006:

Mcnally addressed the resource usage issue.  I don't think that kicking off
idle users is really keeping grex safe in anyway, though I concur with nharmon
that grex should have the authority to do what it feels necessary to keep the
system ``safe'' (though by what metric you judge safety is an open question).

As for `bots' and the like; (a) You need network access to, well, talk to the
network, which is only something members have (other users have very limited
network access; but that's not at issue here). (b) m-net turned off its idle
daemon, and doesn't seem to have much of a problem. (c) those users likely to
run annoying bots on grex that are targeted at grex can already circumvent
idled.  Therefore, idled doesn't really buy one much.


#17 of 88 by eprom on Tue Feb 21 00:27:51 2006:

I vote yes.


#18 of 88 by kingjon on Tue Feb 21 02:01:38 2006:

Re #16: You need network access to talk to the network, but that doesn't stop
them from building the thing, starting it up, testing it (it can talk to
localhost, after all), and then going away.



#19 of 88 by nharmon on Tue Feb 21 03:21:22 2006:

Re 18. Robocop will.


#20 of 88 by cross on Tue Feb 21 04:01:24 2006:

(What Nate said.)


#21 of 88 by kingjon on Tue Feb 21 13:40:44 2006:

Re #19: As I understand it -- not without idled.



#22 of 88 by remmers on Tue Feb 21 14:06:04 2006:

I have an opinion on this issue, but before I give I think I should
point out that this is not a "vote proposal" in the strict sense because
according to the bylaws, vote proposals have to be entered by members
and unless there's been a recent change in his status that hasn't been
recorded yet, Dan Cross isn't one.  The bylaws specify timelines such as
a two-week discussion period for proposals, but since this isn't a vote
proposal, the clock hasn't started ticking yet.  (See
http://cyberspace.org/local/grex/bylaws.html and particular "Article 5:
Voting Procedures".  (The copy of the bylaws posted in this conference
isn't up-to-date, so please refer to the web version.))  

In the meantime, this is a discussion item, not a vote proposal.  Which
is fine -  the idle daemon is certainly an issue that's appropriate to
discuss in Coop.

That said, I think this issue is a bit of a silly candidate for a member
vote, as there's no data for members to make an informed choice.  If
even the technical staff can't agree, how can members be expected to
have a reasonable grasp of the consequences of their decision?

So I propose an experiment:  Let's turn off the idle daemon for a trial
period (two weeks? a month?) and see what the effects are.  If serious
problems emerge, we turn it back on.  In any case, after the trial
people will have an idea of the consequences and can cast an informed
vote.  Or the situation might be so clear-cut that there's no need for a
vote.

What do you think?  Let's keep on discussing, but unless somebody comes
up with some really show-stopper reasons for not trying this, I think
I'll turn the daemon off myself two days from now and we can have our trial.


#23 of 88 by jep on Tue Feb 21 14:26:30 2006:

I agree with remmers.  I don't have the information needed to make a 
decision on the idle daemon.  Anyway, it's a technical decision and 
should be left up to the staff.

His suggestion of a temporary test is a good one, provided the staff 
doesn't see a likelihood of catastrophic consequences.

This item did give me another reminder to re-up my Grex membership, as 
a follow-up to aruba's gentle requests.  I finally took care of it.


#24 of 88 by naftee on Tue Feb 21 14:28:03 2006:

You're calling this a discussion item, and yet you say there's "no data for
members to make an informed choice." ??

i think there's lots of data in this fine item, sir.

Also, i like the trial IDea


#25 of 88 by cross on Tue Feb 21 15:39:27 2006:

Sounds good to me.  No, I'm not a member.


#26 of 88 by other on Tue Feb 21 15:52:34 2006:

I support remmers' recommendation.


#27 of 88 by nharmon on Tue Feb 21 17:15:46 2006:

Me too. Remmers' sense of taking charge is something we need more of on
the BoD. Thank you Remmers!


#28 of 88 by remmers on Tue Feb 21 19:21:12 2006:

You're welcome, although I was wearing my staff hat more than my board hat.


#29 of 88 by cross on Tue Feb 21 21:30:32 2006:

You can wear both hats at once.  If one had the bill pointing forward,
and the other backward, then perhaps you'd look something like Sherlock
Holmes.


#30 of 88 by nharmon on Tue Feb 21 23:34:51 2006:

This response has been erased.



#31 of 88 by nharmon on Tue Feb 21 23:37:35 2006:

We could use the "silly hat fund" to buy Remmers a Sherlock Holmes hat.
He would need a Calabash Pipe too, although he would probably just blow
bubbles with it.


#32 of 88 by aruba on Wed Feb 22 04:31:47 2006:

Heh.  I agree with trying it for a while.


#33 of 88 by other on Wed Feb 22 05:01:12 2006:

I actually saw a gentleman walking in town wearing a deerstalker hat the
other day.


#34 of 88 by scholar on Wed Feb 22 05:02:19 2006:

DID YOU SEE WHAT HE WAS WEARING ON HIS OTHER HEAD<

AHAHAHA



WHY ARE YOU PAYING SO MUCH ATTENTION TO WHAT MEN WEAR


#35 of 88 by sholmes on Wed Feb 22 05:12:15 2006:

I can lend my hat.


#36 of 88 by scholar on Wed Feb 22 05:43:28 2006:

OTHER IS A BIG FAG
,.


#37 of 88 by naftee on Wed Feb 22 06:29:27 2006:

OTHER IS A FIG BAG


#38 of 88 by scholar on Wed Feb 22 06:33:14 2006:

AHAHA< MROE LIKE HE DOESN"T HAVE A FORESKIN BECAUSE A DEMONIC RABBI ( WHO WAS
PRESUMABLY DRESSED UP LIKE A DEVIL AND NOT A TROLL BECAUSE IT WOULD BE HARD
TO HIDE THE HORNS) BIT IT OFF AT THE BEHEST OF HIS PARENTS>

AAHAHAH


#39 of 88 by tod on Wed Feb 22 09:13:54 2006:

I can relate


#40 of 88 by scholar on Wed Feb 22 19:12:22 2006:

CHEWED UP AND SHABBAT OUT


#41 of 88 by tod on Wed Feb 22 22:26:09 2006:

FIB BAG HAHAHAH


#42 of 88 by naftee on Thu Feb 23 02:49:10 2006:

AHAHAHA < FULL OF LIES AND USED FORESKINS


#43 of 88 by remmers on Thu Feb 23 15:32:19 2006:

Turned off idled.  Also commented out the code in /etc/rc.local that
invokes it, so that it won't start if the system reboots.  Let's keep
our eyes open and see whether or not problems arise.

I looked at the robocop source code to see if it depends in idled in any
way.  Apparently it does not.


#44 of 88 by sholmes on Thu Feb 23 16:36:20 2006:

 can we then run 'screen' from work, go home and start again ?


#45 of 88 by kingjon on Thu Feb 23 18:58:10 2006:

Re #43: When I, at least, said robocop "depends on" idled, I meant that robocop
wouldn't kill a process with a living parent, and so to prevent someone from
running eggdrop, say, idled would kill their shell, and *then* robocop would
kill the daemon.



#46 of 88 by nharmon on Thu Feb 23 19:33:37 2006:

Except a person trying to run eggdrop isn't likely to stay logged in 
after it doesn't work because of the network restrictions.


#47 of 88 by remmers on Thu Feb 23 21:51:54 2006:

In the absence of idled, people can presumably run daemons that don't
access blocked network services.  But then, they could do that before by
running them in conjunction with a (trivial to implement) idled defeater.


#48 of 88 by cross on Fri Feb 24 02:35:52 2006:

Thanks, John.  This will be an interesting experiment.


#49 of 88 by scholar on Fri Feb 24 05:12:10 2006:

By the way, uh, just for the sake of the item, the following is a shell script
you can run in the background to foil that nasty idle killer:

#!/bin/sh

while :
do
        touch `tty`
        sleep 59
done

I think I stole that from like jp2 or someone a million years ago, but it's
what I've used ever since, when I've had need for such a thing.  :(


#50 of 88 by remmers on Fri Feb 24 16:22:52 2006:

That script has been around forever.  Like I said in #47, an
idled-defeater is trivial to implement.


#51 of 88 by richard on Fri Feb 24 16:59:43 2006:

I think the idle zapper should stay because when a user is logged in and his
computer crashes or he otherwise becomes a ghost, that ghost login won't go
away unless its zapped.  you could run a !who and see a zillion ghost logins
because nobody on staff logged in to manually clean up the mess.


#52 of 88 by cross on Fri Feb 24 17:13:47 2006:

Normally, those `ghost' logins go away by themselves once the network notices
that the connection is gone.  Let's try this and see what happens.  If it's
a problem, we can act accordingly.


#53 of 88 by scholar on Fri Feb 24 17:26:15 2006:

Re. 50:  :(


#54 of 88 by remmers on Sun Feb 26 02:44:28 2006:

Re ghost logins - they don't seem to be accumulating.  I ran a small
experiment:  Logged in, put my computer to sleep (thus losing the
network connection), came back a few hours later.  I was no longer
connected.  Logged in on another terminal; my previous login was gone. 
So the system seems to be taking care of them.


#55 of 88 by cross on Thu Mar 2 21:49:42 2006:

Well, Idled has been off for a week now, and so far, nothing particularly
bad seems to be happening.  At my count a few minutes ago, about 13 logins
that would have been zapped are logged in.  Most of those have been idle
for less than 3 hours.  The ability to be logged in from more than one
terminal is convenient.


#56 of 88 by cross on Thu Mar 23 17:06:33 2006:

Today marks four weeks since John Remmers turned off idled.  I don't think
there have been any major system problems; one person complained about lack
of a TTY device on login, but I note that the upper limit for tty's seems
pretty low right now (somewhere in the 40's).  It seems that that limit
should be increased, rather than bring back idled.

As of right now, grex claims to have 45 users logged in.  At least two of
those are doing something with FTP; the rest are doing whatever it is that
people do on grex.  Of those, around half have been idle long enough that
they would have been zapped.  But, only about 7 have been idle longer than
3 hours.  If we assume that some people go to sleep, wake up, and do their
grex thing but just leave an idle shell open most of the time, that doesn't
seem like a bad number.

In my opinion, this experiment has been a success.  Other comments?


#57 of 88 by marcvh on Thu Mar 23 17:35:37 2006:

For my own experience, I sometimes grex from work, where we have an
application proxy that is very aggressive about timing out idle
connections.  The net result is that if I am distracted even briefly my
connection gets terminated there, but it seems to take Grex a long time
before noticing and removing the session.

The upshot is that it's pretty easy for me to, if I'm not paying
attention, be logged in several times "at once" because the other
connections are no longer open on my end.  This isn't a big deal for
me; when I notice I can clean it up.  I'm only mentioning it in case
there might be others like me.


#58 of 88 by nharmon on Thu Mar 23 18:43:59 2006:

If it was a big deal, you could kill your other login processes Marc.


#59 of 88 by marcvh on Thu Mar 23 19:05:03 2006:

Um, yes, I'm aware of that, that's why I said I "can clean it up."
Thanks.


#60 of 88 by nharmon on Thu Mar 23 21:08:23 2006:

Do you run screen?


#61 of 88 by tod on Thu Mar 23 21:45:29 2006:

I wasn't able to log in on Tuesday night due to used up tty's.


#62 of 88 by naftee on Thu Mar 23 22:41:22 2006:

isn't tuesday night family time ?!


#63 of 88 by cross on Fri Mar 24 00:00:07 2006:

I'd like to bump up the tty limit and see what happens.  It should be higher
anyway....


#64 of 88 by naftee on Fri Mar 24 03:55:46 2006:

i'm sure you do a lot of bumping in those friday night dance clubs, dan


#65 of 88 by cross on Fri Mar 24 06:21:12 2006:

What can I say?  Your mom is a freak like that, son.


#66 of 88 by naftee on Fri Mar 24 23:34:28 2006:

you're.... my mom ?!


#67 of 88 by tod on Fri Mar 24 23:56:40 2006:

And that wasn't breast milk, soup


#68 of 88 by naftee on Sat Mar 25 04:37:08 2006:

oh my.

that's why they told me milk is yellow


#69 of 88 by kingjon on Wed Mar 29 00:43:59 2006:

I'm getting "all network ports in use" errors when I try to telnet in.
(Ssh was just sitting there, so I tried telnet to confirm my
suspicion; this was posted via backtalk.)


#70 of 88 by cross on Wed Mar 29 03:47:39 2006:

All network ports, or all terminals?  The two are rather different.  At
present, there are only about 42 people logged in, so I don't think it's
the latter.

Possibly, if grex is serving a lot of network requests (mail, the web
server, etc) it can run out of endpoints for TCP connections, which would
make telnet not work (nor SSH, for that matter) though that's rather
different than running out of pseudo-terminals, which would be the case
if there were too many idle logins, as you seem to be implying.


#71 of 88 by spooked on Wed Mar 29 09:04:32 2006:

And, it is near impossible for grex to run out of the former (unless we 
are being DoS attacked).


#72 of 88 by kingjon on Wed Mar 29 11:15:18 2006:

"All network ports in use" was what telnet said. 



#73 of 88 by cross on Wed Mar 29 14:57:59 2006:

Regarding #71; Oh, I don't know about that.  I've gotten machines to run
out of TCP PCB's pretty easily, usually by running a benchmark program or
something against it.

Regarding #72; I doubt it had much to do with idle logins, then.


#74 of 88 by kingjon on Wed Mar 29 18:51:03 2006:

I just assumed it was the same problem -- sorry.



#75 of 88 by cross on Wed Mar 29 23:35:01 2006:

That's understandable; no need to apologize.


#76 of 88 by fuzzball on Sun Apr 2 05:09:50 2006:

i say kill it, aperantly the overall system idle killer is down due to 
i forgot about a grex session a few days back and nearly 24 hours 
later i was still here.

the party thing is annoying as hell though. every 20 minutes? kill it 
or extend it to an hour or something.


#77 of 88 by remmers on Wed Apr 12 15:34:24 2006:

Right, idle daemon has been turned off.

If I recall correctly, one reason for implementing the party idle killer
was to make it (marginally) more difficult to defeat the system idle
daemon.  Since the latter has been disabled, maybe the party idle killer
should go too.  What do other people think?


#78 of 88 by nharmon on Wed Apr 12 15:37:20 2006:

I have no problem with turning off the party idle killer. But I do 
think it would be crappy to see 15 people in party and be the only one 
talking.


#79 of 88 by remmers on Fri Apr 14 18:48:03 2006:

That's the main reason I can think of for keeping the party idle killer.


#80 of 88 by fuzzball on Mon May 1 17:51:17 2006:

would it be possibel to create some sore of idle notification?

like 
User                             Started          Channel
fuzzball                         May  1 13:17:38  party
nharmon                          May  1 00:01:19  party 
remmers                          May  1 13:48:58  party

would be what you see and this is what would happen if say I was idle 
for a set amount of time:

User                             Started          Channel
fuzzball                         May  1 13:17:38  party (IDLE)
nharmon                          May  1 00:01:19  party 
remmers                          May  1 13:48:58  party


#81 of 88 by remmers on Mon May 1 17:57:40 2006:

How would that help?


#82 of 88 by naftee on Mon May 1 20:06:42 2006:

lets people know if they're going to respond or not


#83 of 88 by fuzzball on Tue May 2 17:42:20 2006:

right. so if you popped into party and everyone was idle, its knowing 
they arent gonna chat right offhand. if you killed the idle zapper and 
had no notifications its the luck of the draw as to if yor gonna get a 
chat.


#84 of 88 by cross on Tue May 2 19:20:44 2006:

Party has its own idle timeout.


#85 of 88 by fuzzball on Tue May 2 23:37:20 2006:

re:84

yea, i thought this was about getting rid of it.


#86 of 88 by cross on Wed May 3 02:42:06 2006:

No, this item was about getting rid of the *system* idle daemon.


#87 of 88 by fuzzball on Wed May 3 03:53:55 2006:

oh, oh yea i guess it is.....

still since its gone anyways, maby this needs to become a party idle 
zapper conversation


#88 of 88 by jesuit on Wed May 17 02:16:04 2006:

TROGG IS DAVID BLAINE


There are no more items selected.

You have several choices: