We need to use sudo: required, and dist: trusty to avoid old versions of
bash in travis which don't support the -n argument to the `wait` shell
built-in.
We had to disable the -e checks in etcd.sh since the killall || killall
parts were causing those to trigger in travis.
Every graph needs each vertex to have a change to run initially (after
it has started up) so that initial state detection can be applied to
fix anything that happened while the program was not running. We used to
poke every vertex which was unnecessary, when in fact we only need to
poke the set of vertices that are the minimum set of ultimate
pre-requisites for every other vertex in the graph. That way, you're
either poked directly, or poked by someone who was, etc...
It turns out we don't need Dilworth's theorem, and that looking at
vertices with an indegree of 0 is enough (I think it is a special case
when we have a DAG).
This also fixes a goroutine start scheduling race by ensuring the
initial pokes are received!
This required a change in the event system to add an "activity" field.
This is meant to be generic in the case that there is more than one need
for it, but at the moment, allows a poke to tell that it is a poke in
response to an apply that just finished, instead of a regular poke or
backpoke in which all that matters is timestamp updates, because there
wasn't any actual work done (since that state was okay).
* Fixup graph state readability
* Rename original SetState() to SetConvergedState() and friends...
* Add type state management for proper BackPoke() commands...
* Add better DEBUG logging
This is an important optimization that prevents running a BackPoke on a
parent which is in the process of running and will most certainly poke
the caller back in a moment. This avoids unnecessary roundtrips.
Unfortunately, there are still other algorithms required so that races
can't cause the graph to run for longer than necessary.
* Fix Process() object calling
* Add PokeParent() to poke upwards
* Break linear exec chains :(
This was the issue where in a graph f1 -> f2, if you were to rm f2 &&
cat f2, then f2 would not come back because we didn't poke upwards to
refresh the timestamp. Unfortunately this adds another bug which we
solve in a later patch.
* Add exec type
* Switch erroneous use of fmt to log instead
* Check for edge existence for safety before using
* Avoid recalling etcd channel maker
* Clean up logging output
This is the third main feature of this system. The code needs a bunch of
polish, but it actually all works :)
I've tested this briefly with N <= 3.
Currently you have to build your own etcd cluster. It's quite easy, just
run `etcd` and it will be ready. I usually run it in a throw away /tmp/
dir so that I can blow away the stored data easily.
This requires graphviz to be installed on your machine. If you run the
command with sudo, it will create the files with the original user
ownership to make it easier to remove them without root.
This is still a dirty prototype, so please excuse the mess. Please
excuse the fact that this is a mega patch. Once things settle down this
won't happen any more.
Some of the changes squashed into here include:
* Merge vertex loop with type loop
(The file watcher seems to cache events anyways)
* Improve pgraph library
* Add indegree, outdegree, and topological sort with tests
* Add reverse function for vertex list
* Tons of additional cleanup!
Amazingly, on my first successful compile, this seemed to run!
A special thanks to Ira Cooper who helped me talk through some of the
algorithmic decisions and for his help in finding better ones!
