GitLab

multiple interfaces. The multiple interfaces issue is going to take
some thought and work to get right.

geni code ready for primetime.

now.

competing traffic - stemmed the packet losses in rcv buffers. Completely(& unfairly) dominates a competing IPerf TCP flow though.

database is only used for Utah's emulab.  In the process many other
changes where made to dumperrorlog to fix bugs and make it more
correct/robust.  I also avoided the need to explicitly lock any tables.

This is only used to determine if we can skip reloading the disk
(as is, or rather was, the case with some osids).  If we cannot
get the osinfo, we just force the disk to be reloaded.

on widearea nodes (if node is in reloading, it aggressively blows away
partition tables because we don't care then; if not in reloading, looks
for a spare unused disk.  The latter will fail if all disks are in use, of
course.

don't light up right away.

via the API instead of ssh. Did some cleanup (more conversion to
objects) while I was in there.

instead of ssh. Did some cleanup (more conversion to objects) while I
was in there.

Also add -s (silent mode) option.

I'm sure there was a reason for this at one time, but I don't recall it...

Don't update lastlog_seqnum until _all_ the data is copied, not
just the data from the log data.

Protect dumperrorlog with a lock to keep multiple instances form
running at the same time.

Modified patch to not use the binary (system call) sysctl interface.  As far as I can tell, we don't need it anyway as we only access the ipod settings via /proc/sys.  See Documentation/sysctl/ctl_unnumbered.txt in the kernel source for the current policy regarding binary sysctls.

not in use.

to remote nodes.

* Hacky tmcd redirection. If the reserved table slot tmcd_redirect is
  set, return redirect spec that tells the client tmcc to drop the
  connection and retry the server at the new location, using the vnode
  id that is part of the redirection string. Note that tmcd_redirect
  is set on the remote emulab by the Geni startsliver code.

* Neuter the privkey stuff that we require of remote nodes. In fact,
  its already only required for RON nodes, and rather then yet another
  exception, just kill it. It offers us nothing.

* Neuter the ssl client verification. This is where we verify the
  client certificate has a CN field with the type of the node equal to
  type the DB says it is. This is also a pointless check since is
  offers us nothing additional; the client certificate already had to
  be signed by us. Tired of adding special cases to the code for each
  new node type.

* Temporary neutering of the requirement that all remote nodes use ssl
  to talk to tmcd. The problem here is that remote nodes on other
  testbeds will not have the proper certificate on their images, and
  so they will not be able to talk to our tmcd. Since we do not return
  anything sensitive via tmcd, I have relaxed this requirement for
  now, and changed the check so that functions with newly added flag
  F_REMREQSSL will not be allowed unless it is ssl. For the protogeni
  code this will do since I only need a few things.

* For tmcd on the remote testbeds, there is new code in doaccounts
  that will return accounts and ssh keys from the nonlocal users
  table.  This table is set up by the Geni libraries during sliver
  creation (from the registry entry for the slice).

from the default doaccounts widearea cases.  A widearea node is called
dedicated if the dedicated_widearea node_type_attribute for a type is set
to 1.  In this case, remote nodes look exactly like local nodes, of
course.

We need it to make sure that we make writeable mounts for /var, which
the emulab version of netif writes into.

osconfig scripts.
  * fixserial: fixes securetty, inittab, grub.conf according to args
  * cpwa.sh: simply copies widearea files from boot medium to disk
(the scripts can be run without osconfig, but then you need to set some
env vars that osconfig sets before it runs them.)

locking the "log" table (and hence preventing swap activity) for too
long.