RFC: update retry plugin backoff default to be more sensible

[bendavies]

Q	A
Bug fix?	no
New feature?	no
BC breaks?	no
Deprecations?	no
Related tickets	#79
Documentation
License	MIT

As discussed in #79, the default delay function is not very delaying at all. With 10 retries, the delays (in seconds) of the old impl are:

0.001
0.002
0.004
0.008
0.016
0.032
0.064
0.128
0.256
0.512
1.024

I've simply multiplied by 100 here, so the new delays are:

0.1
0.2
0.4
0.8
1.6
3.2
6.4
12.8
25.6
51.2
102.4

Open to other suggests for a better delay.

[Nyholm]

The documentation for the constructor says:

/**
* @var callable $delay A callback that gets a request, an exception and the number of retries 
* and returns how many milliseconds we should wait before trying again.
*/

So the callable should return milliseconds. But we use in in usleep which takes microseconds. I think we must either multiply the delay with 1000 before we use usleep or change the documentation for the callable.

[bendavies]

Yep, good point. What do you think of my proposed default delay now?

[Nyholm]

I think the backoff times are more reasonable. But it would not hurt to multiply them (yours) with 5 or maybe 10. =)

[bendavies]

* 5

0.005
0.01
0.02
0.04
0.08
0.16
0.32
0.64
1.28
2.56
5.12

* 10

0.01
0.02
0.04
0.08
0.16
0.32
0.64
1.28
2.56
5.12
10.24

[Nyholm]

No, sorry, I meant x500 or x1000.

[bendavies]

Sure, the the relays would get very long quite quickly though:
* 500

0.5
1
2
4
8
16
32
64
128
256
512

[Nyholm]

Yeah, but doing more than 5 will rarely be necessary.

[Nyholm]

Could you make these updates to the PR? Both better defaults and the doc/multiply thing discussed?

[bendavies]

@Nyholm updated

[Nyholm]

Good. Dont forget to update the tests.

[Nyholm]

Lets precalculate this. Write 500000 instead.

[bendavies]

I actually did this intentionally so it's more obvious what we are doing, but happy to oblige .

[bendavies]

spec updated

[Nyholm]

Thank you.