Merge pull request #232 from symbiont-io/runtime-system-prototype

add: prototype runtime system

src/runtime-prototype/.stylish-haskell.yaml

@@ -0,0 +1,25 @@
+steps:
+  - imports:
+      align: none
+      list_align: with_module_name
+      pad_module_names: false
+      long_list_align: new_line_multiline
+      empty_list_align: inherit
+      list_padding: 7 # length "import "
+      separate_lists: false
+      space_surround: false
+  - language_pragmas:
+      style: vertical
+      align: false
+      remove_redundant: true
+  - simple_align:
+      cases: false
+      top_level_patterns: false
+      records: false
+  - trailing_whitespace: {}
+
+# You need to put any language extensions that's enabled for the entire project
+# here.
+language_extensions: []
+
+columns: 72

src/runtime-prototype/CHANGELOG.md

@@ -0,0 +1,5 @@
+# Revision history for stunt-double
+
+## 0.1.0.0 -- YYYY-mm-dd
+
+* First version. Released on an unsuspecting world.

src/runtime-prototype/LICENSE

@@ -0,0 +1,3 @@
+Copyright (c) 2021 Symbiont Inc.
+
+All rights reserved.

src/runtime-prototype/README.md

@@ -0,0 +1,107 @@
+### stunt-double
+
+`stunt-double` is an actor implementation where the real actors can easily and
+faithfully be swapped out for fakes, to make testing of distributed systems
+easier.
+
+#### Actors
+
+* Erlang is perhaps the most commonly associated programming language when one
+  thinks of implementations of actor systems today
+
+* Many other programming languages have tried to copy Erlang, e.g. Scala, Cloud
+  Haskell, etc.
+
+* The concept is, of course, older: Smalltalk (1972), Hewitt, Carl; Bishop,
+  Peter; Steiger, Richard (1973). "A Universal Modular Actor Formalism for
+  Artificial Intelligence". IJCAI.
+
+* Erlang was influenced by Smalltalk, which many people associated with OOP but
+  as per Alan Key's
+  [email](http://lists.squeakfoundation.org/pipermail/squeak-dev/1998-October/017019.html)
+  the key idea is actually message passing.
+
+#### Interfaces (a.k.a. "behaviours")
+
+* Joe Armstrong's PhD thesis "Making reliable distributed systems in the
+  presence of software errors" written in 2003 when he was 53 years old, he
+  started working on telecom systems in 1981. Quite different from a typical PhD
+  thesis written by a 20-something year old person with no or very little
+  experience from industry...
+
+* Anyway, the thesis can perhaps best be summarised by "the big idea in Erlang
+  isn't messaging/actors for concurrency, but using interfaces (a.k.a.
+  "behaviours") to *abstract away the concurrency*" (these are not Joe's words,
+  but rather my interpretation)
+
+* "network normal form"
+
+* Examples of interfaces: gen_server, gen_supervisor, gen_fsm, gen_event. The
+  idea is that an application can be written by largely combining several
+  instances of these.
+
+* Concurrent application becomes sequential, easier to write (especially for
+  less experience programmers) according to Joe
+
+* easier to maintain, if an interface is improved or a bug fixed in it, then all
+  application written against the interface will benefit
+
+* easier to test/prove according to Joe
+
+* In the future work section of the thesis: "How can we program components - the
+  idea of a set of communicating processes leads naturally to the idea of
+  writing the components in different languages. How can we do this?"
+
+#### Deployments
+
+* Infrastructure is a map from machine id (ip address) to set of Vat ids
+
+* Deployment is a supervisor tree (each actor and supervisor has a child id) and
+  a map from child id to machine id and Vat id
+
+* Deploy git commit hashes or hashes of the AST (Unison-style)?
+
+* Join two deployments (make assumptions about existing deployments)?
+
+#### Live debugging
+
+* Connect to (remote) event loop
+* REPL
+* Check crashes that happened by checking the supervisors' logs
+* Check the source code of each actor
+* Check the state of each actor
+
+* Keep a ring buffer for each actor with the last X messages it received so we
+  can step back and forth through time and see how the state changed
+
+* Keep a ring buffer with the latest Y crashes together with a ring buffer with
+  the last Y messages before that crash
+
+#### Upgrades
+
+* Hot-code swapping
+
+* Automatic rollbacks
+
+#### Automatic scaling
+
+* E.g. kubernetes detects high resource usage on some event loop, provisions a new machine, writes to some global register that the new machine is available
+
+#### Protocols
+
+* Type system? Dynamic session types!
+
+
+#### Capabilities
+
+* E programming language
+* [Goblins](https://spritelyproject.org/#goblins)
+* OpenBSD's [pledge](https://man.openbsd.org/pledge.2) and
+  [unveil](https://man.openbsd.org/unveil.2)
+
+#### Deterministic testing
+
+#### See also
+
+* Erlang's [supervisors](https://erlang.org/doc/man/supervisor.html);
+* https://capnproto.org/

src/runtime-prototype/Setup.hs

@@ -0,0 +1,2 @@
+import Distribution.Simple
+main = defaultMain

src/runtime-prototype/app/Main.hs

@@ -0,0 +1,6 @@
+module Main where
+
+import Control.Concurrent
+
+main :: IO ()
+main = print =<< getNumCapabilities

src/runtime-prototype/cabal.project

@@ -0,0 +1,11 @@
+packages: .
+
+with-compiler: ghc-8.10.4
+
+reject-unconstrained-dependencies: all
+
+package stunt-double
+  ghc-options: -Wall
+
+allow-older: *
+allow-newer: *