Implement random resolver option

[dontcrash]

Implements the option for a random resolver to go alongside this PR AdguardTeam/AdGuardHome#7950

Hopefully resolves AdguardTeam/AdGuardHome#4756

[windsurf-bot]

Other comments (1)

• proxy/upstreammode.go (27-27) The comment for `UpstreamModeRandom` is missing a verb. Consider changing to "makes server to use a random upstream" or "makes server to select a random upstream".

💡 To request another review, post a new comment with "/windsurf-review".

[windsurf-bot]

The PR adds UpstreamModeRandom to the switch statement, but this constant isn't defined anywhere in the visible code. You need to define this constant alongside the other UpstreamMode constants (like UpstreamModeFastestAddr, UpstreamModeLoadBalance, etc.).

[dontcrash]

This one I believe is fine as it is inherited from AdGuardHome?
Although I realised the switch case is wrong, fixed it up

UpstreamModeFastestAddr,
UpstreamModeLoadBalance,
UpstreamModeRandom,
UpstreamModeParallel:
// Go on.

[windsurf-bot]

There's a typo in the variable name: UpstreamModeRandomm should be UpstreamModeRandom.

[dontcrash]

Facepalm.

[ainar-g]

Thanks for the contribution! We'll review it shortly.

[ainar-g]

Please update README.md as well.

[ainar-g]

This doesn't build:

proxy/exchange.go:28:44: p.randSrc.Intn undefined (type "math/rand/v2".Source has no field or method Intn)

See go doc math/rand/v2.

[EugeneOne1]

Please, consider using gofumpt to format the code in this repository:

1. make go-tools

2. ./bin/gofumpt --extra -w .

[Sil3ntVip3r]

I ported this feature onto current dnsproxy master (acf2b30e05202e171bbbaf4a3176ea07edb964b3) in an isolated worktree and completed the missing implementation and test coverage. No commit or push was made.

The smallest current-master implementation is:

• add UpstreamModeRandom = "random" to parsing and configuration validation;
• reuse the existing sampleuv.Weighted sampler with equal weights in random mode; this produces a uniform random permutation and retries failed upstreams without replacement;
• retain RTT-derived weights unchanged for load_balance;
• preserve fallback and query-statistics accounting;
• document the new CLI value in both internal/cmd/args.go and generated README.md.

Using the existing sampler is important here: it avoids the reported rand/v2.Source.Intn build failure and preserves the established retry path rather than selecting one upstream and discarding the rest.

I added deterministic tests for enum encoding, configuration validation, uniform choice despite six-orders-of-magnitude RTT differences, retrying every failed upstream exactly once, fallback, and statistics. Verification on the patched current-master tree passed focused and repeated race tests, the full race suite, go vet, make go-check, build, Markdown/text lint, formatting/diff checks, and Darwin, FreeBSD, OpenBSD, Linux, and Windows Go checks. An independent adversarial review found no remaining source blocker.

This also addresses the existing review requests to update README.md, make the code build with math/rand/v2, and format it with the repository toolchain. The companion AdGuard Home port still needs an accepted dnsproxy commit or release before its go.mod/go.sum can be finalized.