Skip to content

N
net
Switch branch/tag
  1. 28 Feb, 2019 1 commit
  3. 27 Feb, 2019 8 commits
  5. 26 Feb, 2019 3 commits
  7. 25 Feb, 2019 1 commit
  9. 13 Feb, 2019 1 commit
  11. 06 Feb, 2019 1 commit
    • Brad Fitzpatrick's avatar
      http2/h2demo: fix the HTTP/1-vs-HTTP/2 demo after HSTS breakage · 65e2d4e1
      Brad Fitzpatrick authored 
      We were previously demoing HTTP/1 vs HTTP/2 loading performance using
http:// scheme (which uses HTTP/1 implicitly) on the http2.golang.org
domain for HTTP/1, and https://http2.golang.org for HTTP/2.

But then golang.org got into the HSTS preload list, forcing all
*.golang.org to only be HTTPS.

So now, rather than find a new base domain name, to compare against
HTTP/1 we instead use https://http1.golang.org/ and then use the SNI
name to determine whether we advertise "h2".

Also, some cleanup:

* remove launch.go; it's no longer used since we moved to kubernetes
* use a multi-stage Dockerfile rather than the hacky workarounds
  that used to be necessary to simulate multiple stages
* modernize the kubernetes deployment stuff to match how we do
  it elsewhere ("gcloud docker" is long deprecated too)
* update from Go 1.9 to Go 1.11 for the prod base

Fixes golang/go#30033

Change-Id: I9f6b1f496d4005e5a08bf990843d440005a5b3e8
Reviewed-on: https://go-review.googlesource.com/c/160857Reviewed-by: 's avatarAndrew Bonventre <andybons@golang.org>
      65e2d4e1
  13. 25 Jan, 2019 2 commits
  15. 19 Jan, 2019 1 commit
  17. 10 Jan, 2019 2 commits
  19. 08 Jan, 2019 3 commits
  21. 07 Jan, 2019 3 commits
  23. 20 Dec, 2018 2 commits
    • Nigel Tao's avatar
      publicsuffix: hold icann-ness until wildcards fully match · 927f9776
      Nigel Tao authored 
      Consider the "uberspace.de" domain. The relevant rules in the Public
Suffix List are "*.uberspace.de" (in the PRVIATE DOMAIN section) and
"de" (in the ICANN DOMAIN section).

The PublicSuffix function returns a string and a bool. Both before and
after this commit, the string returned is "de", which is correct
according to a literal interpretation of the formal algorithm. But the
bool returned, icann-ness, is false before and true after. The correct
answer is true, since the matching rule, "de", is in the ICANN DOMAIN
section of the PSL.

Before this commit, the two-stage match for "*.uberspace" set the icann
bit when matching the back part, "uberspace", before checking that the
front part, the "*" wildcard, also matched.

A couple more examples, for the "bd" and "ck" domains. The relevant
rules are "*.bd" and "*.ck", with no non-wildcard "bd" or "ck" rule.
Before this commit, the PublicSuffix function would return (icann ==
true), when the correct result is (icann == false), the same as for
"nosuchtld".

Benchmarks get worse, but correctness trumps performance. Future commits
may be able to recover some of the loss. In any case, in absolute terms,
15µs is still pretty fast.

name             old time/op  new time/op  delta
PublicSuffix-56  11.0µs ± 0%  14.8µs ± 2%  +34.57%  (p=0.000 n=9+10)

Change-Id: I85ca6ab57a31308af5a29c46313197897eab5ab6
Reviewed-on: https://go-review.googlesource.com/c/154977Reviewed-by: 's avatarNigel Tao <nigeltao@golang.org>
      927f9776
    • Nigel Tao's avatar
      publicsuffix: add comments on the icann return value · 45fec1d2
      Nigel Tao authored 
      Change-Id: I51d236b07f187a6cbf4f6b8167f445e3280bbdf6
Reviewed-on: https://go-review.googlesource.com/c/155199Reviewed-by: 's avatarNigel Tao <nigeltao@golang.org>
      45fec1d2
  25. 17 Dec, 2018 1 commit
  27. 13 Dec, 2018 1 commit
    • Michael Fraenkel's avatar
      http2/hpack: track the beginning of a header block · 891ebc4b
      Michael Fraenkel authored 
      dynamic table size updates must occur at the beginning of the first
header block. The original fix, golang/go#25023, guaranteed it was at
the beginning of the very first block. The Close method implicitly
marked the end of the current header. We now document the Close behavior
and can track when we are at the beginning of the first block.

Updates golang/go#29187

Change-Id: I83ec39546527cb17d7de8a88ec417a46443d2baa
Reviewed-on: https://go-review.googlesource.com/c/153978Reviewed-by: 's avatarBrad Fitzpatrick <bradfitz@golang.org>
Run-TryBot: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>
      891ebc4b
  29. 07 Dec, 2018 1 commit
  31. 01 Dec, 2018 1 commit
    • Brad Fitzpatrick's avatar
      http2: revert Transport's strict interpretation of MAX_CONCURRENT_STREAMS · 351d144f
      Brad Fitzpatrick authored 
      And add the http2.Transport.StrictMaxConcurrentStreams bool knob to
behavior being reverted.

In CL 53250 for golang/go#13774 (for Go 1.10) we changed the HTTP/2
Transport's policy such that a server's advertisement of a
MAX_CONCURRENT_STREAMS value meant that it was a maximum for the
entire process, instead of just a single connection.

We thought that was a reasonable interpretation of the spec and
provided nice safety against slamming a server from a bunch of
goroutines doing concurrent requests, but it's been largely
unpopular (see golang/go#27044). It's also different behavior from
HTTP/1 and because you're usually not sure which protocol version
you're going to get, you need to limit your outbound HTTP requests
anyway in case you're hitting an HTTP/1 server.

And nowadays we have the Go 1.11 Transport.MaxConnsPerHost knob too
(CL 71272 for golang/go#13957). It doesn't yet work for HTTP/2, but it
will in either Go 1.12 or Go 1.13 (golang/go#27753)

After this is bundled into net/http's, the default HTTP client will
have this knob set false, restoring the old Go 1.9 behavior where new
TCP connections are created as necessary. Users wanting the strict
behavior and import golang.org/x/net/http2 themselves and make a
Transport with StrictMaxConcurrentStreams set to true. Or they can set
Transport.MaxConnsPerHost, once that works for HTTP/2.

Updates golang/go#27044 (fixes after bundle into std)

Change-Id: I4efdad7698feaf674ee8e01032d2dfa5c2f8a3a8
Reviewed-on: https://go-review.googlesource.com/c/151857Reviewed-by: 's avatarAndrew Bonventre <andybons@golang.org>
      351d144f
  33. 29 Nov, 2018 1 commit
  35. 14 Nov, 2018 2 commits
  37. 13 Nov, 2018 2 commits
  39. 08 Nov, 2018 3 commits