provider: refactor to only maintain one batched implementation and add throughput callback

[Jorropo]

This code does a few things:

1. Remove the simple provider to avoid duplicating features.
2. Add the support for single providing on the batched provider.
3. Fix a bugs in the batched provider.
4. Add support for a throughputCallback in the batched provider.
5. Add support for an offline mode of the batched provider (stuff is exclusively pushed onto the queue).
6. Move the batched provider to be the only provider and make the queue implementation private.

[hacdias]

@Jorropo can you update this once you have time so that the tests are passing? I can take a look after (also add some motivation on why this is beneficial for more context, please).

[codecov]

Codecov Report

Merging #273 (07350aa) into main (c23df38) will increase coverage by 1.08%.
The diff coverage is 50.54%.

@@            Coverage Diff             @@
##             main     #273      +/-   ##
==========================================
+ Coverage   49.36%   50.44%   +1.08%     
==========================================
  Files         281      279       -2     
  Lines       33708    33586     -122     
==========================================
+ Hits        16640    16944     +304     
+ Misses      15329    14886     -443     
- Partials     1739     1756      +17     

Impacted Files	Coverage Δ
provider/noop.go	0.00% <0.00%> (ø)
provider/provider.go	0.00% <0.00%> (ø)
provider/reprovider.go	70.41% <76.19%> (ø)
provider/internal/queue/queue.go	76.76% <100.00%> (ø)

... and 10 files with indirect coverage changes

[BigLep]

Yes, agreed that the context for where this is coming from and why we're prioritizing it now would be great. It's not currently on the 0.21 iteration. (I'll add it now.)

[hacdias]

Is this ready to be reviewed @Jorropo?

[BigLep]

Related to ipfs/kubo#9703

[BigLep]

2023-05-23 maintainer conversation:

Context: we know this will get rewritten again when it moves into the DHT. As a result, this doesn't have to be perfect.

@guillaumemichel : can you please look at this (given this is relevant to future code for the DHT)? We want to make sure you have context here when future DHT rewriting happens.

@aschmahmann will look at the test code.

@Jorropo : add a changelog.md entry

[guillaumemichel]

The function below is not deterministic and only has a 50% chance to return an error when the context is done.

boxo/provider/internal/queue/queue.go

Lines 57 to 65 in 04cf63e

	// Enqueue puts a cid in the queue
	func (q *Queue) Enqueue(cid cid.Cid) error {
	select {
	case q.enqueue <- cid:
	return nil
	case <-q.ctx.Done():
	return fmt.Errorf("failed to enqueue CID: shutting down")
	}
	}

I rewrote it, feel free to update if required

boxo/provider/internal/queue/queue.go

Lines 57 to 71 in 08e0680

	// Enqueue puts a cid in the queue
	func (q *Queue) Enqueue(cid cid.Cid) error {
	select {
	case <-q.ctx.Done():
	return fmt.Errorf("failed to enqueue CID: shutting down")
	default:
	}
	select {
	case q.enqueue <- cid:
	return nil
	case <-q.ctx.Done():
	return fmt.Errorf("failed to enqueue CID: shutting down")
	}
	}

[guillaumemichel]

I relocated some magic numbers in reprovider.go

boxo/provider/reprovider.go

Lines 21 to 37 in c1d114c

	const (
	// MAGIC: how long we wait before reproviding a key
	DefaultReproviderInterval = time.Hour * 22 // https://github.com/ipfs/kubo/pull/9326
	// MAGIC: If the reprovide ticker is larger than a minute (likely), provide
	// once after we've been up a minute. Don't provide _immediately_ as we
	// might be just about to stop.
	defaultInitialReprovideDelay = time.Minute
	// MAGIC: how long we wait between the first provider we hear about and
	// batching up the provides to send out
	pauseDetectionThreshold = time.Millisecond * 500
	// MAGIC: how long we are willing to collect providers for the batch after
	// we receive the first one
	maxCollectionDuration = time.Minute * 10
	)

Note that I updated the DefaultReprovideInterval from 24h to 22h (see ipfs/kubo#9326 and https://github.com/ipfs/kubo/blob/da28fbc65a2e0f1ce59f9923823326ae2bc4f713/config/reprovider.go#L5)

[BigLep]

@Jorropo : this is good to merge right?

[Jorropo]

@BigLep I'll submit changelog update first