GitLab

Some libcs might not include a sysctl interface (I'm looking at you,
musl!); so only pull this in for the __FreeBSD__ case, which is the only
place we use it anyway.

We've reached the point where it's no longer worth carrying around extra
patch baggage to help iperf 2.0.2 build on newer libcs and compilers.
One reason for that is, when iperf modernized to handle gcc6, they did
it at the same time as a nontrivial reorganization.  So we cannot just
trivially lift their bits, at all.

Second problem: iperf uses a binary packet header.  Our patch used
header bitspace that modern iperfs have since begun using for an
extended header.  So, the 2.0.10 patch in this commit attempts to
maintain binary compat with our hacked 2.0.2 iperf -- but throws out
compat with modern, regular iperfs.

Anyway, this commit just makes the iperf version choice optional, and
it's only 2.0.10 on Alpine Linux; all other build envs will still try to
build 2.0.2 (unless they are ISBRROKEN, of course).

(I know Mike^H^H^H^Hsomeone plans to revisit whether or not our patch is
still even necessary (my bet is that it is, after reading the code), but
I need this commit to get the clientside going on Alpine Linux.)

a unix exitcode, the whole point was to be more flexible!

Previously, we just added to the old end_time which could result in a
time that was still in the past!

the midst of a full cleanup in my devel tree.

munge it, and made it too long.

on a regular basis, which is odd.

not workable longterm); if the slice is no longer there, return a timed
out error. The user will not know what happened, need to work in this.

being used by the cached objects.

Avoids spurious reconnects to the event system.

longer need, expecially XML objects.

1. Do not run the prerender in big experiments. Sucks up a CPU for 15
   minutes, just for a jpeg image on the classic interface, that no one
   is looking at.

2. Do not run vtopgen; this takes a long time too, and since we
   immediately run the mapper right after, why bother. A small change to
   mapper will make sure the node counts are set even in alloconly mode,
   since thats the only reason we run vtopgen in the first place.

this is the cause of the perl core dumps in the py-xml code (according
to the core file). Seems to help but so might a different butterfly. The
core dump was happening in the child we fork to run snmpit in parallel
with os_setup.

10000 lines in the log file.

typically happens when a VM screws up initial boot, and is generally
solvable with a reboot. I had this on the Geni path, but when we
switched to using os_setup directly we lost that functionality.

A small tweak for feeding in an existing assign solution file.

email does not land in my inbox.

including program decide to include it. Also get rid of tblog session
stuff in a couple of libraries, just broken stuff.

CreateSliver() RPC is going to take a really long time as is going to
time out. This would cause the experiment to fail at the Portal. Now we
continue on to WaitForSlivers(), and wait for the manifest to appear,
which signals the CreateSliver() has finally finished, and then we go
into normal waitmode.

* When called from CreateSliver() we can save the assign solution file
  and feed it back into the second mapper run in RedeemTicket(). This
  avoids running assign a second time.

* On a lan, record the link properties before the loop to avoid a O(n^2)
  loop that can take a long time.