Piotr Engelking <[email protected]> wrote: Hi,
> The trigger mode would be also nice, I guess, but it requires kernel > 2.6.21. Which kernels does squeeze support on i386 and amd64? Hmm. I think I'm going to use the trigger mode. I think it's the best solution, all things considered. It gets rid of the cronjob (yay!), a clean instance of mcelog is started when the problem occurs (so you're not relying on the in-memory image of the daemon that could have been corrupted), and I don't have to care about mce support in the kernel. If the relevant sysfs entry doesn't exist, I'm done for the day. It all works with a really simple initscript, which is quite an improvement over the current situation (double-yay!). Also, the log goes to syslog instead of /var/log/mcelog, which will be a very good thing too (network logging). The downside could be that if the machine is screwed enough it may not be able to spawn mcelog. But in that case, the daemon would not be able to write to disk or send out the data over the network anyway. Note that when running the daemon, if the daemon is swapped out (which is likely) you're screwed in just the same way in this case. Seems like a winner to me. If I am missing something, now is the perfect time to tell me about it :) JB. -- Julien BLACHE - Debian & GNU/Linux Developer - <[email protected]> Public key available on <http://www.jblache.org> - KeyID: F5D6 5169 GPG Fingerprint : 935A 79F1 C8B3 3521 FD62 7CC7 CD61 4FD7 F5D6 5169 -- To UNSUBSCRIBE, email to [email protected] with a subject of "unsubscribe". Trouble? Contact [email protected]
