The engine core had some unfortunate bugs that were the result of some
early design errors when I wasn't as familiar with channels. I've
finally rewritten most of the bad parts, and I think it's much more
logical and stable now.
This also simplifies the resource API, since more of the work is done
completely in the engine, and hidden from view.
Lastly, this adds a few new metaparameters and associated code.
There are still some open problems left to solve, but hopefully this
brings us one step closer.
This enables imports in mcl code, and is one of last remaining blockers
to using mgmt. Now we can start writing standalone modules, and adding
standard library functions as needed. There's still lots to do, but this
was a big missing piece. It was much harder to get right than I had
expected, but I think it's solid!
This unfortunately large commit is the result of some wild hacking I've
been doing for the past little while. It's the result of a rebase that
broke many "wip" commits that tracked my private progress, into
something that's not gratuitously messy for our git logs. Since this was
a learning and discovery process for me, I've "erased" the confusing git
history that wouldn't have helped. I'm happy to discuss the dead-ends,
and a small portion of that code was even left in for possible future
use.
This patch includes:
* A change to the cli interface:
You now specify the front-end explicitly, instead of leaving it up to
the front-end to decide when to "activate". For example, instead of:
mgmt run --lang code.mcl
we now do:
mgmt run lang --lang code.mcl
We might rename the --lang flag in the future to avoid the awkward word
repetition. Suggestions welcome, but I'm considering "input". One
side-effect of this change, is that flags which are "engine" specific
now must be specified with "run" before the front-end name. Eg:
mgmt run --tmp-prefix lang --lang code.mcl
instead of putting --tmp-prefix at the end. We also changed the GAPI
slightly, but I've patched all code that used it. This also makes things
consistent with the "deploy" command.
* The deploys are more robust and let you deploy after a run
This has been vastly improved and let's mgmt really run as a smart
engine that can handle different workloads. If you don't want to deploy
when you've started with `run` or if one comes in, you can use the
--no-watch-deploy option to block new deploys.
* The import statement exists and works!
We now have a working `import` statement. Read the docs, and try it out.
I think it's quite elegant how it fits in with `SetScope`. Have a look.
As a result, we now have some built-in functions available in modules.
This also adds the metadata.yaml entry-point for all modules. Have a
look at the examples or the tests. The bulk of the patch is to support
this.
* Improved lang input parsing code:
I re-wrote the parsing that determined what ran when we passed different
things to --lang. Deciding between running an mcl file or raw code is
now handled in a more intelligent, and re-usable way. See the inputs.go
file if you want to have a look. One casualty is that you can't stream
code from stdin *directly* to the front-end, it's encapsulated into a
deploy first. You can still use stdin though! I doubt anyone will notice
this change.
* The scope was extended to include functions and classes:
Go forth and import lovely code. All these exist in scopes now, and can
be re-used!
* Function calls actually use the scope now. Glad I got this sorted out.
* There is import cycle detection for modules!
Yes, this is another dag. I think that's #4. I guess they're useful.
* A ton of tests and new test infra was added!
This should make it much easier to add new tests that run mcl code. Have
a look at TestAstFunc1 to see how to add more of these.
As usual, I'll try to keep these commits smaller in the future!
This allows golang tests to be marked as root or !root using build tags.
The matching tests are then run as expected using our test runner.
This also disables test caching which is unfriendly to repeated test
running and is an absurd golang default to add.
Lastly this hooks up the testing verbose flag to tests that accept a
debug variable.
These tests aren't enabled on travis yet because of how it installs
golang.
This giant patch makes some much needed improvements to the code base.
* The engine has been rewritten and lives within engine/graph/
* All of the common interfaces and code now live in engine/
* All of the resources are in one package called engine/resources/
* The Res API can use different "traits" from engine/traits/
* The Res API has been simplified to hide many of the old internals
* The Watch & Process loops were previously inverted, but is now fixed
* The likelihood of package cycles has been reduced drastically
* And much, much more...
Unfortunately, some code had to be temporarily removed. The remote code
had to be taken out, as did the prometheus code. We hope to have these
back in new forms as soon as possible.
This is an initial implementation of the mgmt language. It is a
declarative (immutable) functional, reactive, domain specific
programming language. It is intended to be a language that is:
* safe
* powerful
* easy to reason about
With these properties, we hope this language, and the mgmt engine will
allow you to model the real-time systems that you'd like to automate.
This also includes a number of other associated changes. Sorry for the
large size of this patch.
Since the pgraph graph can store arbitrary pointers, we don't need a
special method to create the vertices or edges as long as they implement
the String() string method. This cleans up the library and some of the
examples which I let rot previously.
These are helper functions to merge in existing graphs into a main graph
with or without adding an edge relationship between a vertex and the new
graph. These are particularly useful if using mgmt as a lib to break
apart units of work into functions that create sub graphs, which are
then added to the main graph when they're returned.
The graph of dependencies in golang is a DAG, and as such doesn't allow
cycles. Clean up this lib so that it eventually doesn't import our
resources module or anything else which might want to import it.
This patch makes adjacency private, and adds a generalized key store to
the graph struct.
This is required if we're going to have out of package resources. In
particular for third party packages, and also for if we decide to split
out each resource into a separate sub package.
This causes a graph to actually stop processing part way through, even
if there are poke's that want to continue on. This is so that the user
experience of pressing ^C actually causes a shutdown without finishing
the graph execution. It might be preferred to have this be a user
defined setting at some point in the future, such as if the user presses
^C twice. As well, we might want to implement an interrupt API so that
individual resource execution can be asked to bail out early if
requested. This could happen on a third ^C press.
I can't think of a reason we should grab a semaphore before backpoking.
The semaphore is intended to block around the actual work in CheckApply,
not the dependency resolution of the correct vertex.
I forgot about the `concurrent map write` race, but now it's fixed. I
suppose we could probably pre-create all semaphores in the graph at once
before Start, and remove this lock, but that's an optimization for a
later day.
This prevents some nasty races where a BackPoke could arrive on a paused
vertex either during a resume or pause operation. Previously we might
also have poked an excessive number of resources on resume.
The solution was to discard BackPokes during pause or resume. On pause,
they can be discarded because we've asked the graph to quiesce, and any
further work can be done on resume, and on resume we ignore them because
this should only happen during the unrolling (reverse topological resume
of the graph) and at the end of this the indegree == 0 vertices will
initiate a series of pokes which should deal with any BackPoke that was
possibly discarded.
One other aspect of this which is important: if an indegree == 0 vertex
is poked (Process runs) but it's already in the correct state, it should
still transmit the Poke through itself so that subsequent vertices know
to run. Currently this is done correctly in Process().
I'm a bit ashamed that this wasn't done properly in the engine earlier,
but I suppose that's what comes out of running fancier graphs and really
thinking in detail about what's truly correct. Hopefully I got it right
this time!
This prevents a nasty race that can happen in a graph with more than one
resource. If a resource has someone that it can BackPoke, and then
suppose an event comes in. It runs the obj.Event() method (from inside
its Watch loop) and then *before* the resulting Process method can run
it receives a pause event and pauses. Then the parent resource pauses as
well. Finally (it's a race) the Process gets around to running, and
decides it needs to BackPoke. At this point since the parent resource is
paused, it receives the BackPoke at a time when it can't handle
receiving one, and it panics!
As a result, we now track the number of running Process possibilities
via a WaitGroup which gets incremented from the obj.Event() and we don't
finish our pause or exit operations until it has quiesced and our
WaitGroup lets us know via Wait(). Lastly in order to prevent repeated
replays, we detect when we're quiescing and suspend replaying until post
pause. We don't need to save the replay (playback variable) explicitly
because its state remains during pause, and on exit it would get
re-checked anyways.
If two resources are grouped, then the result should contain the
semaphores of both resources. This is because the user is expecting
(independently) resource A and resource B to have a limiting choke
point. If when combined those choke points aren't preserved, then we
have broken an important promise to the user.
This adds a P/V style semaphore mechanism to the resource graph. This
enables the user to specify a number of "id:count" tags associated with
each resource which will reduce the parallelism of the CheckApply
operation to that maximum count.
This is particularly interesting because (assuming I'm not mistaken) the
implementation is dead-lock free assuming that no individual resource
permanently ever blocks during execution! I don't have a formal proof of
this, but I was able to convince myself on paper that it was the case.
An actual proof that N P/V counting semaphores in a DAG won't ever
dead-lock would be particularly welcome! Hint: the trick is to acquire
them in alphabetical order while respecting the DAG flow. Disclaimer,
this assumes that the lock count is always > 0 of course.
