Use len(Allocator.poolIPsInUse) for addresses_in_use metric

[uablrek]

The Allocator.poolServices counter was removed and instead len(Allocator.poolIPsInUse) is used for the "addresses_in_use" metric. It handles shared IP's and seem to be intended for this purpose from start, see comment;

metallb/internal/allocator/allocator.go

Lines 204 to 208 in a9a6f31

	if a.poolIPsInUse[al.pool][al.ip.String()] == 0 {
	// Explicitly delete unused IPs from the pool, so that len()
	// is an accurate count of IPs in use.
	delete(a.poolIPsInUse[al.pool], al.ip.String())
	}

Since this would conflict with PR #624 that fix is included in this PR and #624 will be closed.

Fixes #625
Fixed #622
Fixes #626

Signed-off-by: Lars Ekman lars.g.ekman@est.tech

[russellb]

Thanks! This change looks good to me, though I haven't tested it myself.

Bonus points if you add test coverage, though :)

(I'm not a maintainer, so I can't merge this myself)

[andersmarkendahl]

Hi!

I can confirm that #625 and #622 are both solved by this PR (as expected since now covered by unittest as well).

I built commit 722078a and tested the following:

kubectl apply -f service1.yaml
kubectl apply -f service2.yaml
test-1       LoadBalancer   12.0.245.5   10.0.0.0      5001:32235/TCP   9s
test-2       LoadBalancer   12.0.1.31    10.0.0.1      5001:30399/TCP   4s
wget -q -O- http://$controllerpod:7472/metrics | grep metallb_allocator_addresses_in
_use_total | grep -v 'HELP\|TYPE'
metallb_allocator_addresses_in_use_total{pool="default"} 2

kubectl delete -f service1.yaml
kubectl delete -f service2.yaml
wget -q -O- http://$controllerpod:7472/metrics | grep metallb_allocator_addresses_in
_use_total | grep -v 'HELP\|TYPE'
metallb_allocator_addresses_in_use_total{pool="default"} 0

kubectl apply -f services-shared.yaml
test-1       LoadBalancer   12.0.62.191   10.0.0.0      5001:32469/TCP   4s
test-2       LoadBalancer   12.0.44.248   10.0.0.0      5002:30582/TCP   4s
wget -q -O- http://$controllerpod:7472/metrics | grep metallb_allocator_addresses_in
_use_total | grep -v 'HELP\|TYPE'
metallb_allocator_addresses_in_use_total{pool="default"} 1

Both the empty system and the sharedvip system now shows correct metrics.

@rata Is it possible to get a review from a maintainer on this PR?

Thanks in advance
Anders

[russellb]

Thanks for adding a test!

[rata]

Thanks! Had a quick look and looks good. Will review later this week in case I'm missing anything in the rush :)

[russellb]

We just enabled CircleCI to run tests on open PRs, but that didn't make it run against PRs that were already open. I'm leaving this comment to see if a new comment triggers it to run.

[rata]

@uablrek Hi! Thanks a lot for the PR. The actual fix looks good to me :)

I had a look at the test cases, and it seems prometheus has a testutil package to use for tests. Using that, we can actually test that is what we want and not use the protocol transform + strings.Contains(). I tested it locally and works fine.

Added some suggestions to give you a clear idea of what I just tested, but feel free to modify it if you think it's better with some other changes.

Can you please update it to use the prometheus testutil package for testing?

[rata]

Hi!

I can confirm that #625 and #622 are both solved by this PR (as expected since now covered by unittest as well).

I built commit 722078a and tested the following:

Thanks a lot for testing, is really useful! :-)

@rata Is it possible to get a review from a maintainer on this PR?

You got it! :)

[uablrek]

I have added a commit that fixes #626 also and moved the poolCapacity check to before the test-loop in the stats test.

[andersmarkendahl]

Hi!

I tested again, building the newest commit 9111d10 and I can confirm that #626 is as well solved by this PR.

With a configured, but empty (no LB services yet), system the metrics are working as I would expect.
Allocator metrics shows from start with proper values.

## Address pool with ipv4 /28 subnet only
kubectl apply -f metallb.yaml 

wget -q -O- http://$controllerpod:7472/metrics | grep metallb_allocator
# HELP metallb_allocator_addresses_in_use_total Number of IP addresses in use, per pool
# TYPE metallb_allocator_addresses_in_use_total gauge
metallb_allocator_addresses_in_use_total{pool="default"} 0
# HELP metallb_allocator_addresses_total Number of usable IP addresses, per pool
# TYPE metallb_allocator_addresses_total gauge
metallb_allocator_addresses_total{pool="default"} 16

I also retested #625 and #622 to make sure they still work as expected.

Thank you very much!
Anders

[rata]

@uablrek thanks, LGTM! Added one simple comment to avoid type conversions and, while you are there, can you please squash Updated tests after review commit with Added unit-test for Pool metrics?

If not, no problem, I can do it locally and then push :)

@Aoana thanks a lot for the tests, it really helps! :)

[rata]

Can you use float64 here, so no type conversion is needed later?

[uablrek]

@rata Updated, please re-review

[rata]

Thanks again, LGTM! :)

If you don't mind, I might fixup two commits locally before pushing (the "corrected type" and the comit that added the "incorrect type") and maybe switch the log test prints statements to use %v if you don't mind (seems simpler, unless you used %f for some reason :))

[uablrek]

@rata Please adjust the commits as you wich. And no, I just used %f because of my C-heritage 😃

[rata]

Please adjust the commits as you wich

Cool, will do in a sec! :)

And no, I just used %f because of my C-heritage

haha, it happens to me too 😂 . I think I miss programming in C from time to time, until I actually program in C :-P

[rata]

Thanks a lot for the PR @uablrek and thanks a lot for the tests @Aoana ! I just pushed these, with a simple ammend on this commit.

Closing as it is already merged :)

[andersmarkendahl]

This is really great progress!
Big thanks to @uablrek and (@rata for reviewing). :)