We started seeing this problem when we started running Netscape's FastTrack
server on www.caldera.com. It was configured to cache aggressively and soaked
up lotsa inodes apparently. We used the same work around that has already
been mentioned (echo into /proc/sys/kernel/inode-max). But this leaves
me with several questions:
1) Why can't the kernel handle this condition more gracefully? Is it a "minor
matter of coding" or is there a fundamental reason the kernel must fail
so abruptly when out of inodes...
2) What exactly does "contact Linus" mean? Should this never happen? It does.
(And how does the kernel know that Linus has been hard to find lately?)
3) Has anyone started work on documentation for system tuning and configuration
via /proc/sys? I see references to some /proc entries under "Documentation"
in the kernel source, but it's far from comprehensive. (Did I just volunteer
myself to write a HOWTO?)
Ron
-- Ron Holt <ron@caldera.com> Caldera, Inc.