Comments on: Period Pain

By: diamond

diamond — Wed, 03 Feb 2010 19:59:09 +0000

Nice writeup, i must bear this in mind.

Btw, in the interests of simplicity, line 14 of your first code snippet could be simplified from:

if sorted(numbers)[-1] >= 30:

to:

if max(numbers) >= 30:

Steve

By: warpedvisions.org :: Link: The problem of periodic processes

warpedvisions.org :: Link: The problem of periodic processes — Sun, 03 Jan 2010 00:54:28 +0000

[...] smart writeup of the problem of periodic processes, or why to consider gossipy protocols. window.onload = [...]

By: /~colmmacc/ » Period Pain 3

/~colmmacc/ » Period Pain 3 — Thu, 26 Nov 2009 21:44:55 +0000

[...] promised, though it’s been a while coming, I wrote that there’d be a followup on scheduling [...]

By: /~colmmacc/ » Period Pain part 2

/~colmmacc/ » Period Pain part 2 — Sun, 27 Sep 2009 12:37:40 +0000

[...] week I wrote about problems with periodicity but it was only half of the problem. But before moving on to the second half, it seems like a good [...]

By: brady

brady — Sun, 20 Sep 2009 21:11:15 +0000

Here is an example of something that uses the md5sum of the hostname using modern bashisms. Its not the best code in the world but it works. =)

(the last line is a cheap abs() function as the mod function in bash returns negative numbers)

SYS_HASH=$(echo $HOSTNAME | md5sum | awk ‘{print $1}’)
SLEEP_TIME=$(( (16#$SYS_HASH)%3600 ))
SLEEP_TIME=${SLEEP_TIME/-/}

By: Brian

Brian — Thu, 17 Sep 2009 10:49:33 +0000

The cron.daily for spamassassin in ubuntu seems to have this feature, if you’re volunteering to keep a list…

By: Foobot

Foobot — Wed, 16 Sep 2009 18:41:47 +0000

1. Taking the *daily* apt script as the example, and then extrapolating it to a nominally hourly event is a bit off. A 60 min wobble factor is perfectly reasonable for a daily event. Especially if it’s something like this which is kicked off across the globe at the same *local* time.
A 60 min wobble for an hourly event is bonkers, because 60 mins is a stupid value to use, not because randomising the wait period is necessarily bad.

2. If we’re talking about worst-cases then “Pick a random number once” still has the same worst-case as “random every time”; everybody might randomly pick the same number, only this way they’ll never pick again and the thundering herd
will be a permanent feature of that time-of-day. Predictability is nice but I’m not sure that’s solving the right problem :)

By: gavinmc

gavinmc — Tue, 15 Sep 2009 12:22:04 +0000

It might be useful if cron could have option #1 built into it, so you could do something like:

R 4 * * * root test -x /usr/sbin/cron-apt && /usr/sbin/cron-apt

and cron could make up a random valid number for the R and remember it either permanently for that line or until it was next restarted.

By: colmmacc

colmmacc — Tue, 15 Sep 2009 09:02:53 +0000

@cliff , @jcorn – the easiest and fastest way is to use the mod operator.

$md5sum % 60

will turn it into an evenly distributed number in the 0:59 range.

By: jcorn

jcorn — Tue, 15 Sep 2009 07:18:33 +0000

You know, I changed my mind: multiply by 59 instead. I always get all mixed up by this procedure.