GitLab

have that information in the widearea_nodeinfo table.

(mostly because of a new bridge device in FBSD).

gracefully handle linux kernels that do not support vers=2.

found in the database, use the stack recorded there.

instead of "Experimental" if the stack cannot be determinted.

stack unless no ports are given. This allows it to be used on other
stacks without passing the -S option. There are probably other commands
that should get this treatment, but being conservative for now.

Also improved the debugging output from commands that complain about
multiple stacks.

Fixed bug in Dave's hacky solution. He was only ever adding the constant 10 rather than the correct parameter value.

handle numbers to be hex strings, but the new delay agent only gives
the netlink pipe class the first qdisc handle, from which the class
infers the rest (we always increment handles by decimal 10 -- i.e.,
190, 200, 210).  So, we have to add 10 decimal in the string, then
convert the string to hex.  Whee!

correct.

in the root context.  Also, give containers CAP_NET_ADMIN so they can
set the per-device txqueuelen via ifconfig.  This may be a bad idea...
but for now it's helpful!

like shared host nodes. These are removed when the experiment swapped
or modified, just like the rest of the virtual state.

like shared host nodes. These are removed when the experiment swapped
or modified, just like the rest of the virtual state.

on devices they contain.

The xmlrpc logfile was getting left open in children and things were not
being properly background-ed as a result.

Also fix error reported by Mike where pool_daemon fails
if shared-nodes experiment is NEW. Now, daemon exits
silently if the experiment is NEW or SWAPPED.

kernel patch's problem), adjust param units on Linux.

  Force a new session, should be used by daemons before performing some
  sort of swap related activity.  The orignal session information will
  be lost.

FreeBSD network devices are named after the device driver used, whereas
Linux network devices are named after the interface type, such as 'eth0'
for the first ethernet interface.  The newclient script would send each
interface name to the server via the 'ifacenameX' parameter, and
newnodecheckin.php would use it to get the driver type for the
interface.  This works with the FreeBSD MFS, but not with the Linux one.
Without this fix, all reported interfaces would be of type 'eth', which
isn't helpful.

To fix this, a new parameter was added to newnodecheckin.php called
'ifacedriverX' (where X is the interface number).  This parameter is
optional for backward compatibility with existing newnode MFS images.
If present, it is used to specify the interface type.  If not, the
ifacenameX parameter is munged as before to extract the interface type.

The newclient script reports this parameter under both FreeBSD and
Linux.  On FreeBSD, the interface name is munged to extract the type.
On Linux, the driver name is extracted from sysfs.

so lands in the email message.

Up the timeout for waiting on a child node to 3000 seconds to accomodate estimated time for OpenVZ w/ 100 containers.