On Sat, May 25, 2002 at 04:09:45PM +0900, Stephen J. Turnbull wrote:
> I was getting rvm_malloc asserts (this is weird because I've already
> got a 90M DATA file, and only 900MB in /coda -- comments?)
> 
> I used norton-reinit to dump my server RVM (in files on ext2), and
> reinitialized the LOG at the same size (2M) and extended the DATA from
> 90M to 130M using vice-setup-rvm.  Then I reloaded, all successfully,
> using coda 5.3.18 utilities.
> 
> I tried to update to current CVS.  The build failed and I was kinda in
> a hurry, so I'll try again and report on that later.

5.3.19 changed the stride of vnode allocation, a small step towards
allowing dynamically growing and shrinking of the replication factor of
a volume. However due to this greater stride some problems in the vnode
lookup became visible. They are stored in an array and with the greater
stride we simply use up RVM far too quickly. Current cvs contains a one
line change to bring the stride back. One of the things I'm working on
is moving from an array to a hashtable for vnode lookup.

> The dpkg installs succeeded (except that I had to use force-overwrite
> because of the codaconfedit conflict between server and client pkgs).

Ehh, I thought I had that fixed with dpkg-divert.

> [15:51:40]Main Process: "rpc2b.c", line 120:    RPC2_Init(): Couldn't create socket
> 15:51:40 fatal error -- CommInit: RPC2_Init failed

Typically caused by another venus already/still running, perhaps a
zombie? killall -9 venus should fix that.

Jan