Class | Ruote::Exp::ConcurrenceExpression |
In: |
lib/ruote/exp/fe_concurrence.rb
|
Parent: | FlowExpression |
The ‘concurrence’ expression applies its child branches in parallel (well it makes a best effort to make them run in parallel).
concurrence do alpha bravo end
The concurrence expression takes a number of attributes that allow for sophisticated control (especially at merge time).
concurrence :count => 1 do alpha bravo end
in that example, the concurrence will terminate as soon as 1 (count) of the branches replies. The other branch will get cancelled.
:count and :wait_for may point to a negative integer, meaning "all but x".
concurrence :count => -2 do # all the branches replied but 2 # ... end
:count can be shortened to :c.
This attribute accepts either an integer, either a list of tags.
When used with the integer, it‘s equivalent to the :count attribute:
concurrence :wait_for => 1 do # ... end
It waits for 1 branch to respond and then moves on (concurrence over).
When used with a string (or an array), it extracts a list of tags and waits for the branches with those tags. Once all the tags have replied, the concurrence is over.
concurrence :wait_for => 'alpha, bravo' do sequence :tag => 'alpha' do # ... end sequence :tag => 'bravo' do # ... end sequence :tag => 'charly' do # ... end end
This concurrence will be over when the branches alpha and bravo have replied. The charly branch may have replied or not, it doesn‘t matter.
:wait_for can be shortened to :wf.
Like the :count attribute controls how many branches have to reply before a concurrence ends, the :over attribute is used to specify a condition upon which the concurrence will [prematurely] end.
concurrence :over_if => '${f:over}' alpha bravo charly end
will end the concurrence as soon as one of the branches replies with a workitem whose field ‘over’ is set to true. (the remaining branches will get cancelled unless :remaining => :forget is set).
:over_unless needs no explanation.
As said for :count, the remaining branches get cancelled. By setting :remaining to :forget (or ‘forget’), the remaining branches will continue their execution, forgotten.
concurrence :count => 1, :remaining => :forget do alpha bravo end
:remaining can be shortened to :rem or :r.
The default is ‘cancel’, where all the remaining branches are cancelled while the hand is given back to the main flow.
There is a third setting, ‘wait’. It behaves like ‘cancel’, but the concurrence waits for the cancelled children to reply. The workitems from cancelled branches are merged in as well.
By default, the workitems override each others. By default, the first workitem to reply will win.
sequence do concurrence do alpha bravo end charly end
In that example, if ‘alpha’ replied first, the workitem that reaches ‘charly’ once ‘bravo’ replied will have the payload as seen/modified by ‘alpha’.
The :merge attribute determines which branch wins the merge.
highest and lowest refer to the position in the list of branch. It‘s useful to set a fixed winner.
concurrence :merge => :highest do alpha bravo end
makes sure that alpha‘s version of the workitem wins.
:merge can be shortened to :m.
By default, the merge type is set to ‘override’, which means that the ‘winning’ workitem‘s payload supplants all other workitems’ payloads.
Setting :merge_type to :mix, will actually attempt to merge field by field, making sure that the field value of the winner(s) are used.
:isolate will rearrange the resulting workitem payload so that there is a new field for each branch. The name of each field is the index of the branch from ‘0’ to …
:stack will stack the workitems coming back from the concurrence branches in an array whose order is determined by the :merge attributes. The array is placed in the ‘stack’ field of the resulting workitem. Note that the :stack merge_type also creates a ‘stack_attributes’ field and populates it with the expanded attributes of the concurrence.
Thus
sequence do concurrence :merge => :highest, :merge_type => :stack do reviewer1 reviewer2 end editor end
will see the ‘editor’ receive a workitem whose fields look like :
{ 'stack' => [{ ... reviewer1 fields ... }, { ... reviewer2 fields ... }], 'stack_attributes' => { 'merge'=> 'highest', 'merge_type' => 'stack' } }
This could prove useful for participant having to deal with multiple merge strategy results.
(Available from ruote 2.3.0)
Will override atomic fields, concat arrays and merge hashes…
The union of those two workitems
{ 'a' => 0, 'b' => [ 'x', 'y' ], 'c' => { 'aa' => 'bb' } { 'a' => 1, 'b' => [ 'y', 'z' ], 'c' => { 'cc' => 'dd' }
will be
{ 'a' => 1, 'b' => [ 'x', 'y', 'z' ], 'c' => { 'aa' => 'bb', 'cc' => 'dd' } }
Warning: duplicates in arrays present before the merge will be removed as well.
(Available from ruote 2.3.0)
Much like :union, but duplicates are not removed. Thus
{ 'a' => 0, 'b' => [ 'x', 'y' ], 'c' => { 'aa' => 'bb' } { 'a' => 1, 'b' => [ 'y', 'z' ], 'c' => { 'cc' => 'dd' }
will be
{ 'a' => 1, 'b' => [ 'x', 'y', 'y', 'z' ], 'c' => { 'aa' => 'bb', 'cc' => 'dd' } }
(Available from ruote 2.3.0)
Identical to :concat but hashes are merged with deep_merge (ActiveSupport flavour).
(Available from ruote 2.3.0)
A very simple merge type, the workitems given back by the branches are simply discarded and the workitem as passed to the concurrence expression is used to reply to the parent expression (of the concurrence expression).
:merge_type can be shortened to :mt.
COUNT_R | = | /^-?\d+$/ |
This method is used by some walking routines when analyzsing execution trees. Returns true for concurrence (and concurrent iterator).
Called by reply_to_parent, returns the unique, merged, workitem that will be fed back to the parent expression.
Given a list of workitems and a merge_type, will merge according to the merge type.
The return value is the merged workitem.
(Still used when dealing with highest/lowest merge_type and legacy concurrence/citerator expressions)