Read more at attributes.md.
Throughout documentation (and commit messages), attributes are suffixed with a colon, like in vars: while procedures are surrounded by double quotes, like in "sequence", simply "sequence".
Most likely, "branch of execution". As flor parses the workflow definitions to trees, those trees get executed.
concurrence
sequence 'blue customers'
alice 'gather feedback'
brad 'prepare meeting'
sequence 'green customers'
carla 'gather feedback'
david 'prepare meeting'In the code above, any subtree might be a "branch of execution". The "concurrence" has two branch of executions running concurrently, 'blue customers' and 'green customers'.
The flor component that reads hook or tasker configurations and runs Ruby classes or external scripts with the incoming messages (hooked messages or task messages).
Core contains everything that makes the basic flor language interpreter. It's always executed in a single run - always transient. As soon as there is a timer or a trap - something that would not be present in a vanilla interpreter - it goes into unit. As such, the storage is not a "core" thing, it belongs to "unit".
A domain is a dot separated list of names [a-zA-Z0-9]+.
An execution lives in a domain. An execution may only affect itself and other executions in the same domain branch.
Examples of domains: org.example, org.example.accounting, acme.executive, test.red, test.blue, ...
org.example.accounting is a sub-domain of org.example.
A domain and all its sub-domains.
TODO
An execution is an instance of a flow or in other words a live occurence of a process/workflow. It's identified by an execution id (exid).
When it receives a launch message thus calling for a new execution (think of it as a new instance) or a message for a currently sleeping execution, the scheduler will create an executor and hand it the messages for the execution to start/resume.
An executor will die (stop working) when there are no more messages for the execution (did it just terminate?). In addition, under certain conditions, it can die after a certain number of messages has been processed. Without this limit, an execution might monopolize heavily the resources available to the scheduler. Well, at least that's the reasoning behind the design decision.
If there is already an executor processing messages for the execution, the scheduler will not create a new one, it will simply queue the message for the executor to pick it up, sooner or later.
A unique identifier for a flor execution. For example "test-u-20161204.2144.kawajabachu" identifies an instance of a certain flow. It uses the following format: {fully.qualified.domain}-{unitname}-{yyyymmdd.hhmm.mnemo}.
The combination {exid}-{nid}, designates a single node within a flow execution. (The fei concept was more important in ruote).
An entry in a payload.
Fields in the flor language are accessed with the f. or the field. prefix.
sequence
set f.f0 'Lev Tolstoy'
f.f0 # ==> places 'Lev Tolstoy' in the `ret` field
There is a special ret field, it carries the result of a procedure.
A [business] process definition or a flow for short. In flor, they are written in the flor language.
A function groups a subtree in a reusable unit, generally available behind a name.
A function is defined with define. An anonymous function is defined with def.
define f x #
x + 3 #
f 5 # yields 8 in f.ret
find [ 1, 2, 3 ] #
def elt # anonymous function as argument to "find"
(elt % 2) == 1 #
#
# this "find" yields [ 1, 3 ] in f.ret
The backslash helps when making a function more compact:
find [ 1, 2, 3 ] (def elt \ (elt % 2) == 1)
set f
def x
x * 3
is equivalent to:
define f x
x * 3
("def" sets the reference of the new fucntion in f.ret, it gets picked by "set" and placed in the variable 'f')
Flor wants functions, procedures, and taskers calls to look similar. A single way to call.
The ganger is a component of the scheduler. It's the middleman between flor and the taskers. British definition of the word: "the foreman of a gang of labourers."
The ganger receives the tasks from the flor executor, decides what tasker will be invoked and hands it the task.
A hook is a pair criteria/code. The hooker sees each message emitted/consumed in the executor and triggers the matching hooks.
The code is usually Ruby code but it could be an external script (more costly to run).
Hooks can thus be used for specialised logging purposes, FIXME (continue me)
The hooker is a scheduler component that gets notified for each message treated in the current executor.
The hooker tracks a list of hooks. Upon receiving a message, the hooker hands the message to each matching hook. The matching criteria are the message point, the execution domain, the tags, etc...
Hooks are added programmatically to the unit, or the hooker directly, or they are set via the environment with their code and configuration.
The hooker is also responsible for triggering traps (hooks and traps simply share the same hooker infrastructure).
Read more at hooks.md.
Starting a workflow execution is called a "launch". The unit/scheduler method to start an execution is #launch.
exid = flor.launch(
%q{
sequence
alice 'hello' times: 2
bob 'world'
},
payload: { log: [ "started at #{Time.now}" ] })
# launch a new execution, one that chains alice and bob work(taken from quickstart0).
The loader is a scheduler component helping other components load their codes and libraries. It follows the domain and sub-domain system to load only the codes accessible to a given domain.
Read more at services/loader.md.
Flor is merely about emitting and processing messages within an execution.
An initial "execute" message will launch a new execution and further "execute" message will expand it by adding new nodes to it. When a node is done it sends a "receive" message back to its parent node.
Read more at messages.md.
A node is the data representation of an active procedure of a flor execution.
It has a node id.
Read more at node.md and nodes.md.
Or "node id".
A key for the node in its execution. Is of the form /^\d+(_\d+)*$/, the root node having the "0" nid.
A nid is implicitely suffixed with -0. Nids with a -\d+ suffix may appear for loops and branches executing multiple times.
See subnid.
Most of the messages and the "execute" and "receive" messages in particular have a payload.
A payload is a dictionary / hash. It corresponds to a Javascript object (keys are strings, values are JSON values).
Payloads entries are named fields.
Flor maintains a :flor_pointers table. This table is used behind the Flor::Execution#by_tag(tag_name), #by_var(var_name, value=:no), and #by_tasker(tasker_name, taskname=:no).
The table is populated automatically each time an execution enters or leaves a tag, sets or unsets a top-level variable (top level, is the root node level), and each time a task reaches a tasker or is returned by a tasker.
Querying the :flor_pointers table helps looking up executions in a given state, or, in a reverse way, determine the state of a given execution.
Procedures are the basic building block of the flor language. Those were called "expressions" in ruote.
sequence
task 'alice' 'look for a caterpillar'
sleep '1w'
In this example, "sequence", "task" and "sleep" are procedures.
Throughout documentation (and commit messages), attributes are suffixed with a colon, like in vars: while procedures are surrounded by double quotes, like in "sequence", simply "sequence".
Each time an executor is instantiated to run one or more messages for an execution, it is called an "execution run" or a "run" for short. The lifecycle of an execution might comprise one or more runs. For example a simple execution containing only a sleep call will span two runs.
- The first run reaches the sleep and sets a timer for it;
- Then when the timer triggers, a new executor is instantiated and it performs the post-sleep portion which is the final run ending with a "terminated" message ending the execution.
(renamed to caller, since "run" is an executor concept)
The scheduler is the flor component (daemon) that watches for incoming messages and instantiates executors to act upon them. It's named after its main responsibility which is to schedule flor executions by instantiating executors to run them. A scheduler is composed of a storage, a hooker, a loader and a ganger.
The flor component behind the scheduler that stores messages, executions, timers and traps.
It is mostly a wrapper around a Sequel dataset.
org.example.shop is a sub-domain to org.example which is a sub-domain to org.
See domain.
A nid like 0_0_1-1 is composed of the pure nid (0_0_1) and of subnid (-1). A nide like 0_0_1 only has a pure nid, but in reality, it has an implicit sub of -0.
Subnids bigger than 0 are useful in case of loops and function invocations.
loop
bob 'gather information'
break _ if f.enough_info
The node 0_0 points a bob, but in execution, each pass to bob gets a different subnid.
See nid.
A work item, as handed by flor to a tasker.
sequence
concurrence
task 'technician 1' 'check right motor'
task 'technician 2' 'check left motor'
task 'technician 1' 'check reservoir'
The "task" procedure is used to emit tasks from flor executions.
A piece of (usually Ruby) code in charge of taking an action based on a flor task handed to it.
Taskers or pointers to them are found (by the loader in the lib/ directory of a flor environment. There is actually a ganger between the loader and the tasksers, it ultimately decides which tasker gets handed the task.
A timer is the data structure used by flor to keep track of time triggers. It corresponds to a "flor_timers" SQL table. Basically, flor keeps there a message serialized as JSON and the time at which this message has to be queued for execution.
Flor uses timers for procedures like schedule, cron, and for the timeout system.
TODO
TODO
TODO
It could have been named an "engine", it got named an "unit". It's the set of flor services that work together to run flow executions.
It's usually instantatied via something like
FLOR = Flor::Unit.new('envs/test/etc/conf.json')that is pointing to the configuration file for a given flor environment.
Multiple units may share the same database and thus work together on making the executions progress.
Flor has fields in the payload found in the messages it passes from tasker to tasker (from procedure to procedure).
It also has variables. Whereas fields are seen by taskers, variables stay in the execution.
sequence
set a 'alpha'
set f.a 'allegro'
task 'alice'
# the task handled to Alice will mostly look like { 'a' => 'allegro' }
Variables are mostly used as information for flow control and as function parameters/arguments. It's easy to copy values from and back variables and fields with set.
Note that a tasker may be presented the variables as seen in its execution branch if its configuration sports include_vars: true.
Read more about variables.