On Wed, 31 May 2006 10:44:27 +0200, Florent Daigni?re wrote: > * Jusa Saari <jargonautti at hotmail.com> > [2006-05-31 11:31:08]: > >> Well, I updated the node, added the LD_ASSUME_KERNEL environmental >> variable, restarted the node, and then went to doing other things. Heavy >> CPU usage things. No problem, since I've niced Freenet to 19. Well, >> guess what ? When I came back, Freenet had restarted itself. >> >> I guess that the deadlock protection had detected that whatever it was >> watching wasn't responding in time, and rebooted the node, which - since >> the CPU was loaded - took a loooong time to come back online, and even >> once it had, would simply restart again. > > It should appear in your logfiles.
Where (wrapper log or the logs in the log directory) and what is the wording of the message, exactly speaking ? I can't grep it without knowing that. > >> So: would you please turn off deadlock detection if LD_ASSUME_KERNEL is >> set ? You can detect it, I know, since the warning on the proxy homepage >> is not displayed if you do. > > You're misleading : > > LD_ASSUME_KERNEL is a WORKAROUND against a problem: the watchdog is trying > to circumvent the problem if it occurs... If the LD_ASSUME_KERNEL works around the problem, then the problem shouldn't happen, so there's no point in using the watchdog, no ? > You're basically saying that the workaround doesn't work for you, and > you're suggesting that we remove the watchdog in order to let YOUR node > harm the network, that's it ? No. I suspect that the watchdog is having unintended side effects. And please explain how my node FREEZING harms the network more than it rebooting every few minutes ? > >> Oh, and I have 10 nodes that have updated to 758. They are all backed >> up. ot good, not good at all :(. > > No, in fact it works : your node is misbehaving and knowing that, it's > backing off in order to prevent the spreading of your local overload, due > to the real problem and the non-working workaround. What overload ? The whole LD_ASSUME_KERNEL and watchdog thing was supposed to protect against the node freezing due to the JVM getting confused about synchronization states of objects. How, exactly speaking, does my node spread any load anywhere if it is frozen solid ?
