feat: semaphore added

[helio-frota]

Also using blue-tape to fix a specific test and
Connects to #60

[coveralls]

Coverage increased (+0.01%) to 99.288% when pulling e62a569 on semaphore into fd0fee2 on master.

[coveralls]

Coverage increased (+0.01%) to 99.288% when pulling 0faa766 on semaphore into fd0fee2 on master.

[lance]

I have some questions about this that I think we need to figure out the answers to, and possibly have tests for.

• Does it make sense to always use execution isolation with semaphores? What if there's not really a need for it? Why have it be always on? Reading the Netflix documentation here https://github.com/Netflix/Hystrix/wiki/How-it-Works#semaphores it really only makes sense to use semaphore isolation in certain use cases where the latency is very low and throughput requirements very high. Given that, we have no real way of implementing thread pool isolation/rate limiting, so maybe it does make sense to have it always on. I'm not 100% sure, but I think we should discuss it.
• What happens to the semaphore when a circuit fires, but then times out? Is it released? We need to know that it is.
• What happens when the semaphore has reached its limit and semaphore.acquire() is called? Does it return a rejected promise? Immediately? Or something else? We should test for this.
• The docs should be updated.

[helio-frota]

Does it make sense to always use execution isolation with semaphores? What if there's not really a need for it? Why have it be always on? Reading the Netflix documentation here https://github.com/Netflix/Hystrix/wiki/How-it-Works#semaphores it really only makes sense to use semaphore isolation in certain use cases where the latency is very low and throughput requirements very high. Given that, we have no real way of implementing thread pool isolation/rate limiting, so maybe it does make sense to have it always on. I'm not 100% sure, but I think we should discuss it.

Indeed, also I need to research more about it.

What happens to the semaphore when a circuit fires, but then times out? Is it released? We need to know that it is.

Good question, I'm going to investigate it.

What happens when the semaphore has reached its limit and semaphore.acquire() is called? Does it return a rejected promise? Immediately? Or something else?

We should test for this.

Agree.

The docs should be updated.

yup

[helio-frota]

What happens to the semaphore when a circuit fires, but then times out? Is it released? We need to know that it is.

seems like It is not releasing

 Circuit Breaker timeout with semaphore
    ✖ should be equal
    ------------------
      operator: equal
      expected: 10
      actual:   9

update: yes it is released semaphore.acquire() and calling release();
https://travis-ci.org/bucharest-gold/opossum/jobs/243761512#L458

[helio-frota]

@lanceball: not release when CB timeout is a blocker right ?
if yes (I think it is) then maybe we need to test another semaphore lib, because to call the release() we need first the the aquire().then(release => release()) ...

Also, I'm going try to change sem.use() to sem.aquire() and try to call the release() in some place around that timeout code

[coveralls]

Coverage increased (+0.01%) to 99.291% when pulling e181d43 on semaphore into fd0fee2 on master.

[coveralls]

Coverage increased (+0.01%) to 99.291% when pulling 66096e2 on semaphore into fd0fee2 on master.

[helio-frota]

I'm doing another test, but I missed to add more released() calls.
I'll push the changes soon.

[helio-frota]

What happens when the semaphore has reached its limit and semaphore.acquire() is called? Does it return a rejected promise? Immediately? Or something else?

We should test for this.

So after change the code , and adding the release() calls, I noticed that the counter is always zeroed after release () and subsequent fire calls are not affected.

I will send the test code because I may be testing the wrong way. So it's good to have a new view on this part

[coveralls]

Coverage increased (+0.02%) to 99.298% when pulling 31b6d64 on semaphore into fd0fee2 on master.

[helio-frota]

Now I'm using this lib to run the code in parallel :
https://www.npmjs.com/package/run-parallel

interesting because the code "runs" without error.

[helio-frota]

@lance when running in parallel we have this :

  Semaphore capacity limit in parallel

    ✔ semaphore count is: 0 and initial capacity is: 1
    ✔ failures: 0
    ✔ fallbacks: 0
    ✔ successes: 2
    ✔ rejects: 0
    ✔ fires: 4
    ✔ semaphore count is: 0 and initial capacity is: 1
    ✔ failures: 0
    ✔ fallbacks: 0
    ✔ successes: 3
    ✔ rejects: 0
    ✔ fires: 4
    ✔ semaphore count is: 1 and initial capacity is: 1
    ✔ failures: 0
    ✔ fallbacks: 0
    ✔ successes: 4
    ✔ rejects: 0
    ✔ fires: 4
    ✔ This is not in parallel -> semaphore count is: 1 and initial capacity is: 1
    ✔ failures: 0
    ✔ fallbacks: 0
    ✔ successes: 4
    ✔ rejects: 0
    ✔ fires: 4

[coveralls]

Coverage increased (+0.02%) to 99.298% when pulling 3d990c3 on semaphore into fd0fee2 on master.

[helio-frota]

Given that, we have no real way of implementing thread pool isolation/rate limiting, so maybe it does make sense to have it always on. I'm not 100% sure, but I think we should discuss it.

true, with the current code we can see some real changes only when we run the code in parallel.
otherwise, looks like a serial code - like running without semaphore. Since it will release() every time btw.

[coveralls]

Coverage increased (+0.02%) to 99.298% when pulling b3d7b6f on semaphore into fd0fee2 on master.

[coveralls]

Coverage increased (+0.1%) to 99.375% when pulling 4d0c057 on semaphore into fd0fee2 on master.

[coveralls]

Coverage increased (+0.1%) to 99.375% when pulling 4faf0ca on semaphore into fd0fee2 on master.

[coveralls]

Coverage increased (+0.1%) to 99.375% when pulling 80decd2 on semaphore into fd0fee2 on master.

[coveralls]

Coverage increased (+0.1%) to 99.375% when pulling 2c326cc on semaphore into fd0fee2 on master.

[coveralls]

Coverage increased (+0.1%) to 99.375% when pulling 9feb539 on semaphore into fd0fee2 on master.

[coveralls]

Coverage increased (+0.1%) to 99.375% when pulling 5b00bdc on semaphore into fd0fee2 on master.