Inspired by https://en.wikipedia.org/wiki/Happy_Eyeballs this should
improve latency and fixes the long standing behavior where a single
resolve attempt could take 2x the timeout.
UpstreamResolver.Resolve can still take more than the configured timeout
so maybe that can be improved by splitting the retry algorithm into its
own resolver type.
Replaces `startVerifyUpstream` and behaves just like
`blocking.loading.strategy`.
We use the bootstrap resolver for any requests that arrive before the
upstreams are initialized.
Move `startVerifyUpstream` to `upstreams.startVerify` so it's accessible
via `UpstreamGroup` and we don't need to pass `startVerify` to all
resolver constructors that call `NewUpstreamResolver`.
Also has the nice benefit of greatly reducing the usage of `GetConfig`.
- `CacheControl.FlushCaches`
- `Querier.Query`
- `Resolver.Resolve`
Besides all the API churn, this leads to `ParallelBestResolver`,
`StrictResolver` and `UpstreamResolver` simplification: timeouts only
need to be setup in one place, `UpstreamResolver`.
We also benefit from using HTTP request contexts, so if the client
closes the connection we stop processing on our side.
* refactor: make `config.Duration` a struct with `time.Duration` embed
Allows directly calling `time.Duration` methods.
* refactor(HostsFileResolver): don't copy individual config items
The idea is to make adding configuration options easier, and searching
for references straight forward.
* refactor: move config printing to struct and use a logger
Using a logger allows using multiple levels so the whole configuration
can be printed in trace/verbose mode, but only important parts are
shown by default.
* squash: rename `Cast` to `ToDuration`
* squash: revert `Duration` to a simple wrapper ("new type" pattern)
* squash: `Duration.IsZero` tests
* squash: refactor resolvers to rely on their config directly if possible
* squash: implement `IsEnabled` and `LogValues` for all resolvers
* refactor: use go-enum `--values` to simplify getting all log fields
* refactor: simplify `QType` unmarshaling
* squash: rename `ValueLogger` to `Configurable`
* squash: rename `UpstreamConfig` to `ParallelBestConfig`
* squash: rename `RewriteConfig` to `RewriterConfig`
* squash: config tests
* squash: resolver tests
* squash: add `ForEach` test and improve `Chain` ones
* squash: simplify implementing `config.Configurable`
* squash: minor changes for better coverage
* squash: more `UnmarshalYAML` -> `UnmarshalText`
* refactor: move `config.Upstream` into own file
* refactor: add `Resolver.Type` method
* squash: add `log` method to `typed` to use `Resolover.Type` as prefix
* squash: tweak startup config logging
* squash: add `LogResolverConfig` tests
* squash: make sure all options of type `Duration` use `%s`
According to https://pkg.go.dev/net/http#Transport, when TLSClientConfig
is provided, HTTP/2 will be disabled. Setting this option to true, when
using custom TLS config, will still attempt HTTP/2 upgrades.
Dimitri Herzog