GitLab

Exec summary: after this checkin, the infrastructure exists (once enabled)
to create swapout-time "delta" images for all machines in experiments.
There is only a single, cumulative swap image per node (i.e., all diffs
are from the base image, not from the previous swap).

What doesn't yet exist, is the mechanism for reloading the delta at
swapin time.  That is Phase III.

The nitty-gritty:

1. Keep disk image signature files for all nodes in an experiment.

   New fields in the DB to track, for each disk partition, what image the
   partition was loaded from.  This enables us at swapin or os_load time to
   create signature files in /proj/<pid>/exp/<eid>/swapinfo for the current
   contents of a node disk/partition.  All nodes with the same image loaded
   will share (via symlink) the same signature file.  TODO: no longer
   referenced signature files should be removed.

   Signature info is only collected in the swapinfo directory if the
   experiment is set to have disk state saving enabled (see #5 below).
   Info consists of the <vname>.sig file, which is the file created
   by imagehash, and <vname>.part which says what the root disk is
   for the node and whether to look at the whole disk or just a single
   partition when crafting the delta image.

2. Swapout-time hook for creating swapout image.

   If the experiment is marked as allowing disk state saving, tbswap
   will arrange to run and then monitor the create-swapimage command
   on each node.  This script will run the modified version of imagezip
   which uses the signature file to create a delta image.

   The command to run and maximum timeout are specified via sitevars
   (previously checked in).  Note that the tbswap script currently has
   special knowledge of /usr/local/bin/create-swapimage as a swapout
   time script.  If the swap/swapout_command sitevar is set to that,
   Magic Stuff shall occur (i.e. it will monitor the command and make
   periodic reports of progress).  The sitevars are a total hack and
   will disappear at some point.

3. Client-side script for creating swapout image.

   os/create-swapimage, very similar to create-image.  Uses the info
   stashed in /proj/..blahblah../swapinfo to create a delta image.

   XXX fer now hack: the script first looks in /proj/<pid>/bin for an
   imagezip binary to use.  Failing that, it uses the one in the MFS.
   This allows for easier development of the imagezip changes (i.e.,
   don't have to update the MFS every time.

4. Auto creation of signature files for new images.

   The create_image script (the one that runs on boss when creating images
   for users) has been modified to automatically create a signature via
   imagehash.  The .sig file winds up in /usr/testbed/images/sigs or
   in /proj/<pid>/images/sigs.  From there it will be copied at swapin/os_load
   time to the per-expt swapinfo directory for any node that uses the images.

   The process for creating standard system images (aka, "Mike") has not
   yet been modified.  When the image creation/installation procedure
   is formalized into a script, this will be done.

5. Web changes to set/clear saving of disk state at swapout time.

   Add a checkbox to the experiment create page to allow setting "save
   swap state".  Also added to the experiment modify page, but currently
   "if (0)"ed out as it will need some additional support.  The showstuff
   page will show it.

   Taking a page from Leigh's hack book, if EXPOSESTATESAVE in defs.php3
   is set to zero (as it is now), then the checkbox doesn't appear in the
   create experiment page except for STUDLY users.

I fixed a couple of minor problems, but mostly this worked fine. Note that
I have tested this with the installed perl, *NOT* perl 5.8. I am just
making sure this stuff gets committed before too much more bitrot sets in.

approach is that the person is already overquota and that the initial
mkexpdir script is what fails. Eventually, might need to be more
proactive by checking to see if there some minimal amount of room
before going overquota.

solution, but no time today.

state machine (state). All of the stuff that was previously handled by
using batchstate is now embedded into the one state machine. Of
course, these mostly overlapped, so its not that much of a change,
except that we also redid the machine, adding more states (for
example, modify phases are now explicit. To get a picture of the
actual state machine, on boss:

		stategraph -o newstates EXPTSTATE
		gv newstates.ps

Things to note:

* The "batchstate" slot of the experiments table is now used solely to
  provide a lock for batch daemon. A secondary change will be to
  change the slot name to something more appropriate, but it can
  happen anytime after this new stuff is installed.

* I have left expt_locked for now, but another later change will be to remove
  expt_locked, and change it to active_busy or some such new state name in
  the state machine. I have removed most uses of expt_locked, except those
  that were necessary until there is a new state to replace it.

* These new changes are an implementation of the new state machine,
  but I have not done anything fancy. Most of the code is the same as
  it was before.

* I suspect that there are races with the batch daemon now, but they
  are going to be rare, and the end result is probably that a
  cancelation is delayed a little bit.

in /proj/$pid/exp/$eid, since this would seem to violate group
privacy (/proj exported to all experiments). Minor changes to owner
and mode to allow for non group members to swap/terminate and not get
copy error (of logs).

experiment. Here is mail to tbops:

* Moved the working directory for experiment setup/swap/end to a new
  directory located on boss instead of over NFS to /proj/$pid/$eid. This
  new location is /usr/testbed/expwork/$pid/$eid.

* Changed the name of the directories we create in /usr/testbed/expinfo to
  $pid-$eid.$index where $index is a new autoincrement field in the DB
  table. I really hated the names that were created before.

* Changed where logs are written from /tmp to the new location in
  /usr/testbed/expwork/$pid/$eid.

Okay, why.

* We no longer operate on NFS mounted directories that might hang. Its
  easier to catch the situation where a copy of the log file over at the
  end of experiment creation fails cause of an NFS problem.

* We no longer have user writable files that are inputs to other parts of
  the system (like top and ptop files).  Not that a user would be bad, but
  it closes a hole.

* We no longer copy user writable files from /proj to boss where we might
  fill up an important filesystem cause the user put a .ndz file in the the
  working directory. Not that a user would be bad, but it closes a hole.

* Its easier to save all the log files this way, for each swap in and
  out.

* Removing a directory over NFS is a royal irritant when someone is CD'ed
  into that directory or looking at a file on the other side (the astute
  observer will peg this as the reason I went down this idiotic path in the
  first place!).

* About 6 other reasons that I can no longer remember. Seriously, I really
  had more reasons I can no longer remember! :-)

the testbed list.

via the create-os directive in the NS file.
tbsetup/ir/handle_os.tcl - Do a valid check for the image given with
	set-node-os in the NS file, and propogate that information through
        to the IR file. Nothing is added to the DB.
tbsetup/mkexpdir - Add a tftpboot to the list of experiment directories. The
	tftpd daemon now allows kernels from /proj.
tbsetup/os_setup - Very hacky changes to allow for multiboot kernels. Read
	local images table and cross check against that for nodeos spec.
        Hardwire in "mb" as a special partition tag that says to not try
	and do too much with it. This should be changed to a DB check of
	some kind. On reboot, do not wait for these nodes to come alive since
	there is no way to determine if an oskit kernel (or any foreign)
	kernel is running.