Update golangci-lint: v1.18.0 --> v1.50.1
Linter was updated to fix internal panic which was probably caused by
the Golang update.
The tools directory contains helper programs for managing the health of
the codebase. The directory is based on the equivalent directory found
in the taro codebase at commit 64e9dc9ccadb9269a1985ce2ff03cfc3a83d0470
The target version of the tor.OnionStore interface methods do not accept
onion type as an argument. Further, tor V2 was depreciated in October
2021. We can therefore remove support for tor V2.
The lnd dependency is upgraded to the version currently used by taro.
The motivation behind upgrading lnd is to upgrade lnd/tor to v1.1.0
unit tests: mock signer: add missing interface methods
unit tests: mock signer: add missing interface method arguments
unit tests: mock walletkit: add missing interface method arguments
unit tests: mock lightning client: add missing interface method arguments
unit tests: mock chain notifier: add missing interface method arguments
In this commit, we start a timer if a mailbox stream is completely
un-occupied (neither read or write stream is occupied). The timer
stopped if either of the streams are occupied and is reset if both
streams are unoccupied.
In this commit, we add a mailbox-read-count metric. This will be
incremented each time a mailbox with an _odd_ stream ID is read from.
We do this because we assume that a full duplex connection is being used
meaning that there will be 2 streams that have a matching ID except for
the last byte. And so to avoid duplicating the data, we only record the
odd streams. We also assume that for every read, there will be a write
and so we only record the reads.
There's a race condition in the h2_bundle.go of go 1.16.10 and later. The
issue https://github.com/golang/go/issues/51799 mentions that this might
be fixed in go 1.19, so we'll need to wait for that.
We make sure we build our docker images with go 1.16.9 to not run into
the issue in the wild.
To avoid running into an issue in the race unit test with the Prometheus
histogram metrics that aren't concurrency safe, we don't register any of
them if Prometheus isn't enabled in the first place.
This shouldn't be an issue in production, since we don't start multiple
instances of Aperture _within the same process_ at the same time.
Set the ping and pong intervals of the websocket proxy to zero in order
to disable them. This is needed since a browser client is unable to
respond to these pings.
With this commit we fix a bug in the hashmail server that didn't return
a read stream properly if it was closed from the client side.
Co-authored-by: Elle Mouton <elle.mouton@gmail.com>
We want aperture to handle some of the incoming requests on its own,
without forwarding/proxying them to a remote backend. Those "local"
services can register themselves and will be given every request for
inspection. If a service decides to handle it locally, the request is
passed to that service and not forwarded.
