When an error occours (or the peer disconnect), we return from decorator
as an error is published to the chan.
There are though still 5 or 6 goroutines that want to write on that
channel and at least 3/4 of them will be hanging, leaving them stuck
publishing on the chan.
This though is probably not the cause of
https://github.com/status-im/infra-eth-cluster/issues/39
fills up the stack trace with hung go routines.
Why make this change?
Currently we only rate limit by number of messages.
This works well if we want to limit a large amount of small messages but
breaks down if sending a smaller amount of large messages.
This commit extends the current code to limit by size as well, setting
the default to 1MB per second, which should be enough.
What has changed?
- Rate limiter for waku only
* Added waku README.md, starting text and structure
Resolves GH-1944
* Added api, config and doc details
* Created Docs for waku.const, envelope and events
* Added notes on Filter
* Rewrote the filter section
* Restructured docs to match new versioning structure
* Moved package files out of the main ToC
* Added waku/common/README.md
* Added details to mailserver.go and notes to api.go
* Update to bloomfilter and const common
* Added common/ helpers.go, errors.go and message.go details
* Added common/metrics.go and protocol details
* Added rate_limiter and topic.go details
* fix rate limiter table
* Added Waku Versioning details
* Added Versioned sub-package file details
* Added better detail for Waku Filters
Addressing https://github.com/status-im/status-go/pull/1949#discussion_r419232235
* Added notes about light nodes not being actually new
* Added more detail for Waku node configs and defaults
* Added details to the version message.go section
* Added details to waku/common/rate_limiter
* Minor typo fix in docs
* Refactor tidy of waku package
* Added deprecation warning on whisper README.md
* Appeasing the lint gods and testing is good
* Place Whisper deprecation warning in the correct package README
:facepalm
* Implementing changes after team feedback
* More offerings to the lint gods
* Remove apparently redundant context params
* Correctly handle concurrent HandlePeer err
* Revert "Remove apparently redundant context params"
This reverts commit 557dbd0d64.
* Added note to waku/api.go about context
Why make the change?
As discussed previously, the way we will move across versions is to maintain completely separate
codebases and eventually remove those that are not supported anymore.
This has the drawback of some code duplication, but the advantage is that is more
explicit what each version requires, and changes in one version will not
impact the other, so we won't pile up backward compatible code.
This is the same strategy used by `whisper` in go ethereum and is influenced by
https://www.youtube.com/watch?v=oyLBGkS5ICk .
All the code that is used for the networking protocol is now under `v0/`.
Some of the common parts might still be refactored out.
The main namespace `waku` deals with `host`->`waku` interactions (through RPC),
while `v0` deals with `waku`->`remote-waku` interactions.
In order to support `v1`, the namespace `v0` will be copied over, and changed to
support `v1`. Once `v0` will be not used anymore, the whole namespace will be removed.
This PR does not actually implement `v1`, I'd rather get things looked over to
make sure the structure is what we would like before implementing the changes.
What has changed?
- Moved all code for the common parts under `waku/common/` namespace
- Moved code used for bloomfilters in `waku/common/bloomfilter.go`
- Removed all version specific code from `waku/common/const` (`ProtocolVersion`, status-codes etc)
- Added interfaces for `WakuHost` and `Peer` under `waku/common/protocol.go`
Things still to do
Some tests in `waku/` are still testing by stubbing components of a particular version (`v0`).
I started moving those tests to instead of stubbing using the actual component, which increases
the testing surface. Some other tests that can't be easily ported should be likely moved under
`v0` instead. Ideally no version specif code should be exported from a version namespace (for
example the various codes, as those might change across versions). But this will be a work-in-progress.
Some code that will be common in `v0`/`v1` could still be extract to avoid duplication, and duplicated only
when implementations diverge across versions.
Andrea Maria Piana
Samuel Hawksby-Robinson