Bug#462268: [Popcon-developers] Bug#462268: i dont understand this...
[Bill Allombert] > The number of stable subscription has been raising again before we > change unstable. This is true, and I can not explain it. > The number of machine running unstable is so small that it cannot do > a difference. I believe this is false. As almost all machines in a given time zone will try to report at the same time (if they got a correct clock), I believe the 5000 machines reported to run unstable is enough to affect the result. As more and more unstable machines was upgraded and stopped competing for the small time slot used by the other popcon.debian.org submitters, the number of submissions grew. And > If a DOS did occur, it would take down every web service hosted onnn > gluck, not just popcon and we would probably hear from it. Actually, only services being contacted at the same time as popcon results are submitted, sundays 06:47 for all time zones. So I believe it is unlikely we would hear about it, as the DOS situation would be over very soon, and thus a reload would hide the issue for those seeing it while accessing other web pages on gluck.debian.org. As can be seen from http://popcon.debian.org/ >, the submission growth has stopped. After sunday, we should see if the number of submissions have gone down since I uploaded the package without the random submission time patch to unstable. So far the increase since 2008-03-18 has been 794 submissions. The number of stable submissions have dropped by 13, and the increase is mostly with installations of testing. I'm very curious what these numbers will be on sunday, after the popcon.debian.org web pages are updated that day. :) Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#462268: [Popcon-developers] Bug#462268: i dont understand this...
Hi, On Thursday 20 March 2008 00:11, you wrote: > Must admit I am surprised to see some decrease already, with 77 > machines upgraded to version 1.44 in unstable, and down 21 > submissions. I must say I'm not surprised. We basically *know* (*) this is because the clients are DOSing the server at the same time, so why be surprised if this comes back, when the code, which prevented this (poorly), was remmoved? (*) I forgot why we knew this, maybe its in this bug log even, but there was a very plausible explanation, like >500 machines submitting per hour, all at the full hour, and the apache config allows 500 connections. Something like this. The solution definitly involves making the clients submit their data at random time. > I'm on vacation and offline until sunday, so I will not do anything > more with this until then. I suspect we should wait for 20 days > before we know, which is the period a submission is kept before it is > removed, if it isn't updated by the client. I suggest not to wait 20 days to take action. popcon-stats are widely used, and currently people think debian has way less (popcon) users, because our popcon system is broken. regards, Holger pgp0MlZVeFJKN.pgp Description: PGP signature
Bug#462268: [Popcon-developers] Bug#462268: i dont understand this...
[Petter Reinholdtsen] > A decrease is a lot harder to explain as anything else than a DOS > issue, so I decided to test if that will happen first. This is the way I plan to extract the measurements: ssh popcon.debian.org (cd /org/popcon.debian.org/popcon-mail/all-popcon-results; \ for f in popcon-2008-03-*.gz; do \ echo -n "$f: "; zcat $f |egrep 'Submi|Release: 1.44'; \ done) The current result (to early to conclude, I belive) is popcon-2008-03-01.gz: Submissions:76046 popcon-2008-03-02.gz: Submissions:76043 popcon-2008-03-03.gz: Submissions:76122 popcon-2008-03-04.gz: Submissions:76136 popcon-2008-03-05.gz: Submissions:76062 popcon-2008-03-06.gz: Submissions:76098 popcon-2008-03-07.gz: Submissions:76144 popcon-2008-03-08.gz: Submissions:76193 popcon-2008-03-09.gz: Submissions:76404 popcon-2008-03-18.gz: Submissions:77416 popcon-2008-03-19.gz: Submissions:77395 Release: 1.44 77 Must admit I am surprised to see some decrease already, with 77 machines upgraded to version 1.44 in unstable, and down 21 submissions. I'm on vacation and offline until sunday, so I will not do anything more with this until then. I suspect we should wait for 20 days before we know, which is the period a submission is kept before it is removed, if it isn't updated by the client. Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
Bug#462268: [Popcon-developers] Bug#462268: i dont understand this...
Hi, On Wednesday 19 March 2008 12:03, you wrote: > Note that I decided to test the hypothesis of random sleep not > affecting number of submissions by disabling the workaround and see if > the amount of submissions go down, wouldn't it have been better to let this version go to testing and see, if the number of submissions goes up, as suddenly the machines running testing can submit data? :) > So closing this bug do not really help any at the moment, so I suggest > we keep it open. Then please downgrade it to important. RC bugs are looked at by many people working on fixing RC bugs and this bug would just waste their time. regards, Holger pgpuFGtMwULMw.pgp Description: PGP signature
Bug#462268: [Popcon-developers] Bug#462268: i dont understand this...
[Holger Levsen] > Can you please close this bug, so that the release team has one bug less to > worry about? Note that I decided to test the hypothesis of random sleep not affecting number of submissions by disabling the workaround and see if the amount of submissions go down, so version 1.44 is in unstable without the random sleep. If the number of submissions go down as the unstable machines are upgraded to use the new version, we will try the suggestion from Bill on spreading the submission on different days. So closing this bug do not really help any at the moment, so I suggest we keep it open. Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
