* If upstreams are all using same host but with different ports
ie:
foobar:4001
foobar:4002
foobar:4003
...
Because fnv-1a has not a good enough avalanche effect
Then the hostByHashing result is not well balanced over
all upstreams
As last byte FNV input tend to affect few bits, the idea is to change
the concatenation order between the key and the upstream strings
So the upstream last byte have more impact on hash diffusion
* reverseproxy: Mask the WS close message when we're the client
* weakrand
* Bump golangci-lint version so path ignores work on Windows
* gofmt
* ugh, gofmt everything, I guess
* fileserver: Reject ADS and short name paths
* caddyhttp: Trim trailing space and dot on Windows
Windows ignores trailing dots and spaces in filenames.
* Fix test
* Adjust path filters
* Revert Windows test
* Actually revert the test
* Just check for colons
* httpcaddyfile: Skip some logic if auto_https off
* Try removing this check altogether...
* Refine test timeouts slightly, sigh
* caddyhttp: Assume udp for unrecognized network type
Seems like the reasonable thing to do if a plugin registers its own
network type.
* Add comment to document my lack of knowledge
* Clean up and prepare to merge
Add comments to try to explain what happened
PR #4066 added a dark color scheme to the file_server browse template.
PR #4356 later set the links for the `:visited` pseudo-class, but did
not set anything for the dark mode, resulting in poor contrast. I
selected some new colors by feel.
This commit also adds an `a:visited:hover` for both, to go along with
the normal blue hover colors.
It was not accurate. Placeholders could be used in outputs that are
defined in the same mapping as long as that placeholder does not do the
same.
A more general solution would be to detect it at run-time in the
replacer directly, but that's a bit tedious
and will require allocations I think.
A better implementation of this check could still be done, but I don't
know if it would always be accurate. Could be a "best-effort" thing?
But I've also never heard of an actual case where someone configured
infinite recursion...
* core: Refactor, improve listener logic
Deprecate:
- caddy.Listen
- caddy.ListenTimeout
- caddy.ListenPacket
Prefer caddy.NetworkAddress.Listen() instead.
Change:
- caddy.ListenQUIC (hopefully to remove later)
- caddy.ListenerFunc signature (add context and ListenConfig)
- Don't emit Alt-Svc header advertising h3 over HTTP/3
- Use quic.ListenEarly instead of quic.ListenEarlyAddr; this gives us
more flexibility (e.g. possibility of HTTP/3 over UDS) but also
introduces a new issue:
https://github.com/lucas-clemente/quic-go/issues/3560#issuecomment-1258959608
- Unlink unix socket before and after use
* Appease the linter
* Keep ListenAll
e338648fed3263200dfd6abc9f8100c6f1c0eb67 introduced multiple upstream
addresses. A comment notes that mixing schemes isn't supported and
therefore the first valid scheme is supposed to be used.
Fixes setting the first scheme.
fixes#5087
* caddyhttp: Honor grace period in background
This avoids blocking during config reloads.
* Don't quit process until servers shut down
* Make tests more likely to pass on fast CI (#5045)
* caddyhttp: Even faster shutdowns
Simultaneously shut down all HTTP servers, rather than one at a time.
In practice there usually won't be more than 1 that lingers. But this
code ensures that they all Shutdown() in their own goroutine
and then we wait for them at the end (if exiting).
We also wait for them to start up so we can be fairly confident the
shutdowns have begun; i.e. old servers no longer
accepting new connections.
* Fix comment typo
* Pull functions out of loop, for readability
