GitLab

SLOTHD -> CANARYD

   No longer rely on looking at kernel boot time messages and extracting
   a hardware signature to determine the nodetype to then determine the
   control net.  Now we just DHCP on all interfaces and decree that the
   interface that answers is our control net interface.

   An extraordinary number of sleezy tricks were needed to get FBSD4,
   FBSD5, and RHL to DHCP on all interfaces without changing any standard
   scripts.

   For now, the nodetype/cpuspeed/chipset scripts still exist for the
   benefit of healthd, which uses the output of nodetype to determine
   what kernel module to load.  We should fix this.

   Side-effect: pump, the old RHL DHCP client, is history!  For older
   RHL releases, you will need a version of dhclient.

   Side-effect: in Linux, all non-control net interfaces are left up
   but without a legit IP address.  This is a consequence of dhclient.
   In FBSD, it was trivial to clean this up, RHL will take a little
   more work.  Up or down, it shouldn't matter.

2. Add an mfs-install make target, a scaled-down version of the client
   install.  Added a mandatory DESTDIR check so you don't accidentally
   install in the wrong place on boss.

Oops, forgot to commit the update to configure.

Split off the feedback metric gathering, alerting, and reporting code from
slothd and slammed it into a new daemon: canaryd.  There is currently quite
a bit of code duplication between slothd and canaryd.  Sometime in the future
I need to go back and extract the common portions into a module.

(replaced by dhclient version)

(only problem is, tmcd doesn't allow them...)

1. tmcc (the perl version) has -D flag to support "direct" unproxied call
2. watchdog has a -n flag to support "non-proxied" operation
3. libtmcc has a "noproxy" option, can be set globally or per-call

Oh, and made a per-call version of the "nocache" option for libtmcc:
cleaned up a hack in the watchdog.

2. Support timeout in tmcc proxy, so proxy doesn't hang when it doesn't
   get a UDP reply from the master tmcd.  Unfortunately, the fast path
   here involved introducing a setjmp/longjmp, which I am not too thrilled
   about (worried about Windows support...)  There are other ways (e.g.,
   socket timeouts) to do this, but would have required significant overhaul
   of the code.

3. Clean up some minor, unimportant logic errors

by other means.

 "tmcctest noreply" - does not send an reply back
 "tmcctest delayreply <N>" - delay reply by <N> seconds
 "tmcctest splitreply" - TCP only, send reply back in pieces with delay between

the last two can only be done in debug mode.  This is a quicky hack for
testing tmcc timeouts.

from the DB when the routertype is "static" since those images are
likely to not have ddijk support on them. Newer images will have
ddijk, and so will compute the routes themselves; do not send them
back. Note that static-old implies that nodes will get the routes no
matter what, although old images will probably barf if gets that; I
don't view this as a problem since very people are going to request
static-old.

experimental nodes from the topomap.

"static" routing now implies ddijk routing, but since old images won't
have any of that support, we still have to generate routes and store
them in the DB. However, only doing this when the number of nodes in
the experiment is less than 250; otherwise print a warning and refuse
to generate the routes.

Also add "static-old" so that people can request/force old static
routing no matter how big the experiment.

static-old to use original centralized static routing.

ipassign_wrapper no longer does that. Revert back to always invoking
the static route calculator (which figures out what to do). Remove the
stiff about generating a linkmap, which was an interim measure for
ddijk routing; now done with gentopfile.

	For new distributed routing support, must first run this script as
        root to generate a data file for each existing experiment.

		sudo sql/gentopmaps.pl

client to wait for an experiment to reach a desired state (active,
swapped, etc). Use the optional timeout parameter to terminate the
wait early. See the documentation for a full descriptione of the
interface.

This is implemented using the just added "poll" interface to the
python event interface library

will return this error code instead of RESPONSE_ERROR.

is currently only for use by the RPC server.

between the experiment swapping and the RPC server.

single event at a time, and process that event inline, rather then
within a callback.

this field but forgot to removed it from the create file (it never
went into the migrate file).

interface_capability, which is a comma-separated list of speeds other
than the default which an interface supports.

our own. Also, leave the 'dhcpd_enable' option in rc.conf off so that
the old startup script will stay disabled.