copyright 2020 craigphicks ISC license
TaskSerializer
Outline
The TaskSerializer
module can serialize tasks/promises for integrated control - Tasks/promises can be added immediately as they are produced and then be made available to a consumer when they have resolved and the consumer is ready to read them.
Optionally, the number of concurrently running tasks are limited to a user parameter. In that special case, only functions (and their args) may be added, and function will be executed when a space is available. Trying to add promises will throw an Error.
All rejected tasks/promises are managed so that they don't throw unhandled rejections.
The are 4 different classes exported from the module:
AsyncIter
NextSymbol
WaitAll
Callbacks
Each of those classes has these input functions:addTask(func,...args)
/addTask(promise)
to add tasks/promises.addEnd()
to indicate that no more tasks/promises will be added, thus allowing exit after the pipeline has drained.
The output interface of each of those classes differ, and are suitable for different usage cases. The following table compares some properties of those classes to help decide which is suitable for a given usage case:
property | AsyncIter | NextSymbol | WaitAll | Callbacks |
---|
read buffered | yes | yes | yes | no |
continuous vs. batch | cont | cont | batch | cont |
control loop | yes | yes | no | no |
select style | no | yes | N/A | N/A |
where 'property' are as follows:
- 'read buffered':
- Whether the class has an internal buffer storing the outcomes of finished tasks/promises until they are read by the consumer.
- 'continuous vs. batch':
- Batch indicates either:
- no consumer read until all tasks/promises have resolved, or until any tak/poromise has rejected (
WaitAll.waitAll
) - no consumer read until all tasks/promises have either resolved or rejected (
WaitAll.waitAllSettled
)
- Continous indicates the internal read buffer is intended to be read by consumers before all taks/promises have resolved/rejected.
- 'control loop'
- The output may be easily read in an asynchrous control loop
- 'select style'
- The control loop condition informs an output is 'ready' without actually reading it. This style is useful for a top level control loop integrating 'ready' conditions from many unrelated sources. (See
NextSymbol
usage example.)
Read-buffered classes prioritize rejected-values over resolved-values, and pass the rejected-values first whenever both are availabe. The exception to this rule is WaitAll.waitAllSettled()
, which transforms rejected-values into resolved-values.
Usage examples
Note on shared demo functions
To make the examples more readable, some shared demo functions for a file demo.lib
are used, and those shared functions are listed at the end of the examples. One of those functions is the async function producer(ts)
. It inputs the tasks by calling addTask(...)
staggered over time, followed by addEnd()
. Some of those tasks throw Errors
, other resolve normally.
All the below example code is availalable in the example-usages
subdirectory of the installed node module, e.g., node_modules/task-serializer/usage-examples
.
AsyncIter
usage example
API
'use strict';
const {AsyncIter}=require('task-serializer');
const {exitOnBeforeExit,producer}=require('./demo-lib.js');
async function consumer(ai){
do{
try{
for await(const res of ai){
console.log(' '+JSON.stringify(res));
}
break;
}catch(e){
console.log(' '+'error: '+e.message);
}
}while(true);
}
async function main(){
let ai=new AsyncIter(2);
await Promise.all([producer(ai),consumer(ai)]);
}
main()
.then(()=>{console.log('success');process.exitCode=0;})
.catch((e)=>{console.log('failure '+e.message);process.exitCode=1;});
NextSymbol
usage example
API
'use strict';
const {NextSymbol}=require('task-serializer');
const {makepr,exitOnBeforeExit,producer}=require('./demo-lib.js');
var somethingElse=makepr();
var iv=setInterval(()=>{somethingElse.resolve("somethingElse");},300);
async function consumer(ts){
let emptied=false;
while(!emptied){
let next = await Promise.race([
somethingElse.promise,
ts.nextSymbol(),
]);
switch(next){
case "somethingElse":
console.log(next);
somethingElse=makepr();
break;
case ts.symbolTaskResolved():{
console.log();
let res=ts.getTaskResolvedValue();
console.log("symbolTaskResolved, result="+res);
break;}
case ts.symbolTaskRejected():{
let e=ts.getTaskRejectedValue();
console.log("symbolTaskRejected, message="+e.message);
break;}
case ts.symbolEmpty():{
console.log("symbolEmpty");
emptied=true;
clearInterval(iv);
break;}
}
}
}
async function main(){
let ts=new NextSymbol({concurrentTaskLimit:2});
await Promise.all([consumer(ts),producer(ts)]);
}
main()
.then(()=>{console.log('success');process.exitCode=0;})
.catch((e)=>{console.log('failure: '+e.message);process.exitCode=1;})
;
WaitAll
usage examples
API
'use strict';
const {WaitAll}=require('task-serializer');
const {exitOnBeforeExit,producer}=require('./demo-lib.js');
async function consumer_waitAll(ts){
try{
let r=await ts.waitAll();
console.log(`ts.waitAll() returned`);
console.log(JSON.stringify(r,0,2));
}catch(e){
console.log(`ts.waitAll() caught ${e.message}`);
}
}
async function consumer_waitAllSettled(ts){
let r=await ts.waitAllSettled();
console.log(`ts.waitAllSettled() returned`);
console.log(JSON.stringify(r,0,2));
console.log('consumer finished');
}
async function main(){
let waitAll=new WaitAll({concurrentLimit:2});
await Promise.all([
consumer_waitAll(waitAll),
producer(waitAll),
]);
waitAll=new WaitAll({concurrentLimit:2});
await Promise.all([
consumer_waitAllSettled(waitAll),
producer(waitAll),
]);
}
main()
.then(()=>{console.log('success');process.exitCode=0;})
.catch((e)=>{console.log('failure '+e.message);process.exitCode=1;});
Callbacks
usage example
API
'use strict';
const {Callbacks}=require('task-serializer');
const {exitOnBeforeExit,producer}=require('./demo-lib.js');
async function consumer(ts){
await new Promise((resolve)=>{
ts.onTaskResolved((resolvedValue)=>{
console.log(`onTaskResolved ${resolvedValue}`);
});
ts.onTaskRejected((rejectedValue)=>{
console.log(`onTaskRejected ${rejectedValue}`);
});
ts.onEmpty(()=>{
console.log(`onEmpty`);
resolve();
});
});
console.log('consumer finished');
}
async function main(){
let ts=new Callbacks({concurrentLimit:2});
await Promise.all([
consumer(ts),
producer(ts)
]);
}
main()
.then(()=>{console.log('success');process.exitCode=0;})
.catch((e)=>{console.log('failure '+e.message);process.exitCode=1;});
exitOnBeforeExit(2);
demo-lib.js
function snooze(ms){return new Promise(r=>setTimeout(r,ms));}
function range(len){return [...Array(len).keys()];}
function exitOnBeforeExit(exitCode){
process.on('beforeExit',async()=>{
if (typeof process.exitCode=='undefined'){
console.error('unexpected "beforeExit" event');
process.exit(exitCode);
} else
process.exit(process.exitCode);
});
}
function makepr(){
let pr={};
pr.promise=new Promise((r)=>{pr.resolve=r;});
return pr;
}
async function task(id,ms,err=false){
console.log(`-->enter ${id}`);
if (err)
throw new Error(`task failed id=${id}`);
await snooze(ms);
console.log(`<--leave ${id}`);
return `task ${id}, took ${ms}ms`;
}
async function producer(ts){
for (let i=0; i<6; i++){
ts.addTask(task,i,2**(10-i),(i+1)%3==0);
await snooze(100);
}
ts.addEnd();
console.log('producer finished');
}
APIs
Terminology
- Each task/promise after being added will go through all of the following milestones in order:
- added
- The task has been added with
addTask
.
- started
- If the construtor parameter
concurrentTaskLimit
>0, then a task may be forced to wait before start. - If the construtor parameter
concurrentTaskLimit
<=0, that all task/promises are (considered) started when added.
- finished
- Task/promise has reached an outcome, either one of the following two categories:
- resolved-value
- resulting from
resolve(<resolved-value>)
or return <resolved-value>
- rejected-value
- resulting from ``reject()
or
throw ` - The actual values are determined by the task/promise, not by the
TaskSerializer
module. A rejected-value typically satisfies (<rejected-value> instanceof Error)
, but that is not mandatory.
- read
- Task/promise outcome has been read by the consumer. This state might not be reached of reading is abandoned, e.g. due to a rejected-value.
- The class instance passed through the following milestones, in order:
- started-processing
- First task/promise has been added.
- ended
addEnd
has been called to guarantee no more tasks/promises will be added.
- finished-processing
addEnd
has been called and all added tasks have reached finished.
- empty
addEnd
has been called and all added tasks have reached read.
API shared by all classes
instance=new <Classname>({concurrentTaskLimit=0}={})
- where
<Classname>
is one of AsyncIter
,Callbacks
,NextSymbol
,or WaitAll
. concurrentTaskLimit
is the integer number of task allowed to run concurrently, unless it is <=0
, in which case their is no limit.
instance.addTask(func,...args)
where (func instanceof Function)
, or addTask(promise)
where (promise instanceof Promise)
- in the case of constructor arg
concurrentTaskLimit>0
,
addTask
will allow only the first form, and passing a promise will throw.func(...args)
will be called in the order passed, when the concurrent task limit constraint allows.
- in the case of constructor arg
concurrentTaskLimit<=0
, either form is allowed. - in either case, the tasks/promises are managed by the instance of
<Classname>
until reaching miletone read. The tasks/promises may reject, and those rejections are handled to prevent unhandled promise rejections.
instance.addEnd()
- this a guarantee from the caller that
addTask
will not be called again. It is required so the instance knows that when all tasks/promises have reached the finished milestone, the instance has reached the finished-processing milestone.
- Generally, when possible, rejected-values are passed to the consumer before resolved-values.
WaitAll.waitAllSettled()
is one exception to that rule.
AsyncIter
only API
- see
AsyncIter
usage example for example. const {AsyncIter}=require('task-serializer')
instance=new AsyncIter({concurrentTaskLimit=0}={})
- explicit async
instance.next()
or implicit async for await (iter of instance)
- There are 3 possible outcome categories: resolved-value, rejected-value, and instance-empty, where instance-empty indicated the instance has reached the empty milestone.
- case: explicit
- case: resolved-value
- returns
{done:false,value:<resolved-value>}
- case: rejected-value
- case: instance-empty
- case: implicit
- case: resolved-value
iter
will be the <resolved-value>
- case: rejected-value
- case: instance-empty
NextSymbol
only API
- see
NextSymbol
usage example for example. const {NextSymbol}=require('task-serializer')
instance=new NextSymbol({concurrentTaskLimit=0}={})
- async
instance.nextSymbol()
Returns a value strictly equal to one of instance.symbolTaskResolved()
, instance.symbolTaskRejected()
, or instance.symbolEnd()
.
- case
instance.symbolTaskResolved()
: indicates a task/promise resolved-value is ready to be read. - case
instance.symbolTaskRejected()
: indicates a task/promise rejected-value is ready to be read. - case
instance.symbolTaskResolved()
: indicates the instance milestone empty has been reached.
instance.getTaskResolvedValue()
- This is a sync function intended to be called immediately after
async instance.nextSymbol()
has returned a value equal to instance.symbolTaskResolved()
- It returns the next resolved-value of some task/promise.
instance.getTaskRejectedValue()
- This is a sync function intended to be called immediately after
async instance.nextSymbol()
has returned a value equal to instance.symbolTaskRejected()
- It returns the next rejected-value of some task/promise.
WaitAll
only API
- see
WaitAll
usage examples for example. const {WaitAll}=require('task-serializer')
instance=new WaitAll({concurrentTaskLimit=0}={})
- async
instance.waitAll()
- There is no timing constraint on calling
waitAll
, i.e. no requirement to call before after any instance milestone, although obviously milestone empty is not reached until waitAll
is called. - If any tasks/promises have terminated with an error before the call to
addTask
, then that first one will be returned immediately when waitAll
is called. - If any tasks/promises have terminated with an error after
addTask
is called, waitAll
will immediately return with that error. - Otherise
waitAll
will not return before all the tasks/promises added with addTask
have terminated, and the array of resolved-values will be the order they were added, not the order they were resolved.
- async
instance.waitAllSettled()
- There is no timing constraint on calling
waitAllSettled()
. Any time from before the first addTask
to after all tasks/promises have terminated (i.e., end-of-processing) is allowed. Measures have been taken to prevent unhandled rejections. waitAllSettled
will return no sooner than end-of-processing. It will return the same value as Promise.waitAllSettled()
would return on an array of all tasks/promises added via addTask
in the order the were added.
Callbacks
only API
- see
Callbacks
usage example for example. const {Callbacks}=require('task-serializer')
instance=new Callbacks({concurrentTaskLimit=0}={})
instance.onTaskResolved(callback)
- add the unique callback to be called every time a task/promise resolved-result is ready
instance.onTaskRejected(callback)
- add the unique callback to be called every time a task/promise rejected-result is ready
instance.onEmpty(callback)
- add the unique callback to be called when the instance reaches the empty milestone.
- NOTES:
- Each
instance.on<*>
function should be called only once per instance. Only one callback per function is actually registered. - Each
instance.on<*>
function must be called before the instance has reached the processing milestone, i.e., before the first call to addTask
.
TODO:
- Add in informational functions
instance.getCountAddedNotStarted()
instance.getCountStarterNotFinished()
instance.getCountFinishedNotRead()
instance.getCountRead()