From: Jim Garlick <garlick.jim@gm...> - 2011-05-06 04:25:59
diod 1.0-pre27 is tagged.
I hope this release is a major step closer to the 1.0 release.
It does away with the diodctl super-server, a strategy that was
just not working out, or at least not converging fast enough for
my liking. Now there is one daemon, one config file, and mount
helper, and some other gunk if you want to integrate with the
In cluster testing on the last release we were seeing strange
failures of getpw*/getgr*/getgrouplist functions that was causing
some mounts in a parallel job launch to fail. In this release, all
those calls are serialized, and a small cache of lookup results should
minimize the number of times we need to call them. We also
don't trust the sysconf() values to give us reasonable buffer sizes
for the reentrant getpw*/getgr* functions (use a minimum of 4K).
Error checking/logging in this area of the code has been greatly
enhanced. Hopefully we will have solved that problem or at least
created enough detailed logging that we'll get it next time.
This release has only been tested on a couple of machines.
It should see some cluster testing early next week (I hope).
Pre-releases are available as tarballs here:
Please report any issues to the google code issue tracker here:
Updating from pre26:
- Chkconfig diod on. Service is now called 'diod' not 'diodctl'.
- In diod.conf, rename 'diodctllisten' to 'listen' and use port 564.
Serialize password/group lookups, log all errors, and use big buffers.
Minimize password/group lookups by briefly caching lookup results.
Perform a late term abortion on diodctl. Now we just start diod.
Added diodcat utility and updated auto.diod to use it not diodctl
Reworked the libnpclient library and used it to improve test coverage.
Add better error handling for setfsuid/getfsuid.
Fix a regression in 1.0-pre25 in the gid was never being reset.