[RFC 216/ PushStream] Inconsistent time unit of PushbackPolicyOption.getPolicy() between JavaDoc and implementation about design HOT 4 OPEN

Comment author: Clément Delgrange <[email protected]>

Overview:

The JavaDoc of the PushbackPolicyOption.getPolicy method indicates that the time unit is nanoseconds, however, in the reference implementation org.osgi:org.osgi.util.pushstream:1.0.0, when a tasks is scheduled (eg; BufferedPushStreamImpl.startWorker), the time unit used is miliseconds.

The time unit of the time parameter is also missing in the JavaDoc of BufferBuilder.withPushbackPolicy(PushbackPolicyOption pushbackPolicyOption, long time)

Actual Results:

The user cannot decide which of the specification or the implementation is right. This may lead to bugs in the client code.

Expected Results:

Consistence between JavaDoc and implementation.

from design.

Comment author: Clément Delgrange <[email protected]>

Not sure if it is when a task is scheduled, but when a default PushBackPolicy is set in PushStreamProvider.createStream:

if (pushbackPolicy == null) {
   pushbackPolicy = LINEAR.getPolicy(1000);
}

The provided time does not seem to be nanosecondes.

from design.

Comment author: Castro B <[email protected]>

(In reply to Clément Delgrange from comment BZ#1)

The provided time does not seem to be nanosecondes.

Hi Clement i really didn't get what you mean,

Castro B,
https://sparpedia.no

from design.

Comment author: Clément Delgrange <[email protected]>

Hi,

Sorry, I will try to be more explicit.

First, the abstract method PushbackPolicyOption.getPolicy(long value) indicates in its JavaDoc that this method create an instance configured with a base back pressure time in nanoseconds. So, I have assumed that the value parameter must be a time in nanoseconds. As well as for each implementation LINEAR, ON_FULL_EXPONENTIAL, FIXED, ...

Then, from a Tim Ward's answer on stackoverflow (https://stackoverflow.com/questions/53692861/osgi-pushstream-is-slow), he said:

When you create a stream using psp.createStream(source) it is set up with a buffer of 32 elements and a linear back pressure policy based on the size of the buffer, returning one second when full...

Now, when you look at the implementation of PushStreamProvider.createStream:

return createStream(eventSource, 1, null, null,
				new ArrayBlockingQueue<>(32),
				FAIL.getPolicy(), LINEAR.getPolicy(1000));

if the value provided to LINEAR.getPolicy is a time in milliseconds, when the buffer is full we will get a back pressure of 1 second ( 1000*32/(32+0)=1000ms=1s ) as Tim Ward said but not if it is nanoseconds.

Am I right, or I miss something?

from design.