ca8b627...
by
Brad Fitzpatrick <email address hidden>
net/http: add Response.Uncompressed bool
The Transport's automatic gzip uncompression lost information in the
process (the compressed Content-Length, if known). Normally that's
okay, but it's not okay for reverse proxies which have to be able to
generate a valid HTTP response from the Transport's provided
*Response.
Reverse proxies should normally be disabling compression anyway and
just piping the compressed pipes though and not wasting CPU cycles
decompressing them. So also document that on the new Uncompressed
field.
Then, using the new field, fix Response.Write to not inject a bogus
"Connection: close" header when it doesn't see a transfer encoding or
content-length.
Updates #15366 (the http2 side remains, once this is submitted)
a9cf0b1...
by
Brad Fitzpatrick <email address hidden>
net/http: provide access to the listener address an HTTP request arrived on
This adds a context key named LocalAddrContextKey (for now, see #15229) to
let users access the net.Addr of the net.Listener that accepted the connection
that sent an HTTP request. This is similar to ServerContextKey which provides
access to the *Server. (A Server may have multiple Listeners)
abc1472...
by
Brad Fitzpatrick <email address hidden>
net/http: add Transport.IdleConnTimeout
Don't keep idle HTTP client connections open forever. Add a new knob,
Transport.IdleConnTimeout, and make the default be 90 seconds. I
figure 90 seconds is more than a minute, and less than infinite, and I
figure enough code has things waking up once a minute polling APIs.
This also removes the Transport's idleCount field which was unused and
redundant with the size of the idleLRU map (which was actually used).
The B/op number is effectively meaningless. There
is a surprisingly large one-time cost that gets
divided by the number of iterations that your
machine can get through in a second.
This CL discards the first run, which helps.
It is not a panacea. Running with -benchtime=10s
will allow the sync.Pool to be emptied,
which brings the problem back.
However, since there are more iterations to divide
the cost through, it’s not quite as bad,
and running with a high benchtime is rare.
This CL changes the meaning of the B/op number,
which is unfortunate, since it won’t have the
same order of magnitude as previous Go versions.
But it wasn’t really comparable before anyway,
since it didn’t have any reliable meaning at all.