Re: Stepping the clock during boot
[John Hasler] Would it be unacceptable for Chrony to step the clock (particularly back) during boot? If you are going to do this, it might be a good idea to try to do it before the syslog collector starts, as most daemons depend on $syslog and would thus start after the clock is correct. Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/2flaap45dpx@login2.uio.no
Re: Stepping the clock during boot
On Tue, 3 Aug 2010, John Hasler wrote: I wrote: Well, it's a reason. But note that such backward stepping would only happen when your clock is really screwed up. brian writes: It actually used to happen every reboot of my server, which is why I'm aware of the dovecot problem. When ntp (or ntpdate, I'm not sure which) would correct the time, the clock would move backwards 30 seconds or so and dovecot would crash. I can add an X-Starts-Before: dovecot line to protect dovecot. I think that ideally chrony would replace hwclock as provider of $time when present but I don't know how to arrange that... You could append +chrony to the $time line in /etc/insserv.conf as a local fix. You might be able to ship a /etc/insserv.conf.d/chrony that to that effect but I'm unsure whether that would replace or add to the existing line. I couldn't find documentation on exactly how the /etc/insserv.conf.d files work. -- Edward Allcutt -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/alpine.deb.2.00.1008031008260.2...@jago.allcutt.me.uk
Re: Stepping the clock during boot
[Edward Allcutt] You could append +chrony to the $time line in /etc/insserv.conf as a local fix. You might be able to ship a /etc/insserv.conf.d/chrony that to that effect but I'm unsure whether that would replace or add to the existing line. I couldn't find documentation on exactly how the /etc/insserv.conf.d files work. It will add it to the existing line. Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/2flpqy03upz@login2.uio.no
Re: Stepping the clock during boot
Jaldhar H. Vyas writes: I haven't really looked into the new dependency-based stuff so this might be a naive question but wouldn't X-Starts-After: chrony in dovecots init script be a better idea? There is no such thing as far as I know (and I should have written X-Start-Before: no s). A Should-Start might suffice, though. -- John Hasler -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/87y6cnao4i@thumper.dhh.gt.org
Re: Stepping the clock during boot
Edward Allcutt writes: You could append +chrony to the $time line in /etc/insserv.conf as a local fix. You might be able to ship a /etc/insserv.conf.d/chrony that to that effect but I'm unsure whether that would replace or add to the existing line. As Peter says, it would add it to the existing line. That wouldn't really help as hwclock is always present and would satisfy the $time requirements before chrony started. -- John Hasler -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/87tynbamn7@thumper.dhh.gt.org
Re: Stepping the clock during boot
Petter Reinholdtsen writes: If you are going to do this, it might be a good idea to try to do it before the syslog collector starts, as most daemons depend on $syslog and would thus start after the clock is correct. Unfortunately, chrony also wants $syslog. -- John Hasler -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/87pqxzamgf@thumper.dhh.gt.org
Re: Stepping the clock during boot
[John Hasler] As Peter says, it would add it to the existing line. That wouldn't really help as hwclock is always present and would satisfy the $time requirements before chrony started. Actually, it would help, because all the parts making up $time need to be satisfied before those depending on $time will start. PS: Peter Reinholdtsen is my cousin. :) Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/2fl7hk7g5ae@login1.uio.no
Re: Stepping the clock during boot
Petter Reinholdtsen writes: Actually, [a chrony file in insserv.conf.d] would help, because all the parts making up $time need to be satisfied before those depending on $time will start. Then that solves my problem (should apply to ntp too) if adding a file to insserv.conf.d containing $time chrony guarantees that no script depending on $time will start before chrony does. No, wait. Chrony depends on things that depend indirectly on $time (so does ntp, IIRC). -- John Hasler -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/87y6cn9070@thumper.dhh.gt.org
Re: Stepping the clock during boot
[John Hasler] No, wait. Chrony depends on things that depend indirectly on $time (so does ntp, IIRC). Yes, such change need to be done very carefully and in stages, to avoid dependency loops. See #542602 for a discussion on the ntp order relative to $syslog. :) Happy hacking, -- Petter Reinholdtsen -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/2fl1vaffvth@login1.uio.no
Re: Stepping the clock during boot
Petter Reinholdtsen writes: Yes, such change need to be done very carefully and in stages, to avoid dependency loops. See #542602 for a discussion on the ntp order relative to $syslog. After some research I'm tending to think that many (if not most) of the scripts that require $time, shouldn't (or at least should only require hwclockfirst). I also don't see that we still need two hwclock scripts. Then ntp, chrony, and hwclock could provide $time and only those packages that really need exact time should require it. If the clock gets stepped at bootup that fact will be quite obvious in the logs so $syslog should be able live with the rare stepping. I think I'll have chrony require hwclockfirst and add X-Start-Befores for packages such as dovecot that stepping might break (so if you know of such a package tell me). -- John Hasler -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/87lj8n8pp5@thumper.dhh.gt.org
Re: Stepping the clock during boot
I wrote: After some research I'm tending to think that many (if not most) of the scripts that require $time, shouldn't (or at least should only require hwclockfirst). I also don't see that we still need two hwclock scripts. BTW hwclockfirst.sh and hwclock.sh add about a 1.5 second delay. -- John Hasler -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/87d3tz8dm6@thumper.dhh.gt.org
Re: Stepping the clock during boot
On Mon, Aug 02, 2010 at 04:24:52PM -0500, John Hasler wrote: Would it be unacceptable for Chrony to step the clock (particularly back) during boot? This would only happen when the clock was at least 30 seconds off but I'm concerned that it might screw some of the daemons that might already have started. If Chrony can't step the clock it would have to slew it (which can take hours for large errors) or give up. This would only happen on rare occasions. dovecot kills itself when time moves backwards too much. See http://wiki.dovecot.org/TimeMovedBackwards. I don't think it would be okay unless you can either make sure that dovecot handles this gracefully (and that involves not needing to be restarted or disconnecting users) or change the time only before normal (non-time-related) services start. I think the latter is prudent anyway. Other than this, I don't know any reason why you shouldn't. -- brian m. carlson / brian with sandals: Houston, Texas, US +1 832 623 2791 | http://www.crustytoothpaste.net/~bmc | My opinion only OpenPGP: RSA v4 4096b: 88AC E9B2 9196 305B A994 7552 F1BA 225C 0223 B187 signature.asc Description: Digital signature
Re: Stepping the clock during boot
On Mon, Aug 02, 2010 at 04:24:52PM -0500, John Hasler wrote: Would it be unacceptable for Chrony to step the clock (particularly back) during boot? This would only happen when the clock was at least 30 seconds off but I'm concerned that it might screw some of the daemons that might already have started. If Chrony can't step the clock it would have to slew it (which can take hours for large errors) or give up. This would only happen on rare occasions. I think it's better to risk tripping buggy daemons rather than accept the clock being set to 2007. And most of those which are likely to be run early during bootup have already been fixed. The most likely source of problems is ext3 in its default settings -- but hey, those default settings the last time I looked still included gems like let's randomly DoS the machine for 15 mins during boot just in case or let's throw out 50GB of space because having 1MB reserved had been beneficial in the 1970s on massively multiuser systems. I think it's be worthy to try stepping to see if something breaks down badly, so we can fix it. Meow! -- 1KB // Microsoft corollary to Hanlon's razor: // Never attribute to stupidity what can be // adequately explained by malice. -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/20100802230856.ga30...@angband.pl
Re: Stepping the clock during boot
brian m. carlson writes: I don't think it would be okay unless you can either make sure that dovecot handles this gracefully (and that involves not needing to be restarted or disconnecting users)... I don't see how I can do that. ...or change the time only before normal (non-time-related) services start. I think the latter is prudent anyway. Of course, but that is very hard to ascertain with dependency-based booting. Chrony requires $remote_fs so it's starting fairly late. I suppose I could put in a bunch of X-Start-Before lines, but I'd have to know what all could be adversely affected and booting would be slowed. Other than this, I don't know any reason why you shouldn't. Well, it's a reason. But note that such backward stepping would only happen when your clock is really screwed up. -- John Hasler -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/87iq3sbnla@thumper.dhh.gt.org
Re: Stepping the clock during boot
On Mon, Aug 02, 2010 at 06:29:21PM -0500, John Hasler wrote: Well, it's a reason. But note that such backward stepping would only happen when your clock is really screwed up. It actually used to happen every reboot of my server, which is why I'm aware of the dovecot problem. When ntp (or ntpdate, I'm not sure which) would correct the time, the clock would move backwards 30 seconds or so and dovecot would crash. I've since replaced the hardware with something newer and this problem does not occur anymore. Just be aware that some motherboards for Core 2 Duos do have this problem and that you may very well run into it. -- brian m. carlson / brian with sandals: Houston, Texas, US +1 832 623 2791 | http://www.crustytoothpaste.net/~bmc | My opinion only OpenPGP: RSA v4 4096b: 88AC E9B2 9196 305B A994 7552 F1BA 225C 0223 B187 signature.asc Description: Digital signature
Re: Stepping the clock during boot
I wrote: Well, it's a reason. But note that such backward stepping would only happen when your clock is really screwed up. brian writes: It actually used to happen every reboot of my server, which is why I'm aware of the dovecot problem. When ntp (or ntpdate, I'm not sure which) would correct the time, the clock would move backwards 30 seconds or so and dovecot would crash. I can add an X-Starts-Before: dovecot line to protect dovecot. I think that ideally chrony would replace hwclock as provider of $time when present but I don't know how to arrange that... -- John Hasler -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/8739uwbeup@thumper.dhh.gt.org
Re: Stepping the clock during boot
John Hasler jhasler at debian.org writes: I wrote: Well, it's a reason. But note that such backward stepping would only happen when your clock is really screwed up. brian writes: It actually used to happen every reboot of my server, which is why I'm aware of the dovecot problem. When ntp (or ntpdate, I'm not sure which) would correct the time, the clock would move backwards 30 seconds or so and dovecot would crash. I can add an X-Starts-Before: dovecot line to protect dovecot. I haven't really looked into the new dependency-based stuff so this might be a naive question but wouldn't X-Starts-After: chrony in dovecots init script be a better idea? -- Jaldhar H. Vyas jald...@debian.org -- To UNSUBSCRIBE, email to debian-devel-requ...@lists.debian.org with a subject of unsubscribe. Trouble? Contact listmas...@lists.debian.org Archive: http://lists.debian.org/loom.20100803t051259-...@post.gmane.org