Currently, we use same number of HTTP/2 sessions per worker with given
backend addresses. New option to specify the number of HTTP/2 session
per worker will follow.
It turns out that writing successfully to network is not enough.
After apparently successful network write, read fails and then we
first know network has been lost (at least my android mobile network).
In this change, we say connection check is successful only when
successful read. We already send PING in this case, so we just wait
PING ACK with short timeout. If timeout has expired, drop connection.
Since waiting for PING ACK could degrade performance for fast reliably
connected network, we decided to disable connection check by default.
Use --backend-http2-connection-check to enable it.
Previously we did not check HTTP semantics and it is left out for
application. Although checking is relatively easy, but they are
scattered and error prone. We have implemented these checks in our
applications and also feel they are tedious. To make application
development a bit easier, this commit adds basic HTTP semantics
validation to library code. We do following checks:
server:
* HEADERS is either request header or trailer header. Other type of
header is disallowed.
client:
* HEADERS is either zero or more non-final response header or final
response header or trailer header. Other type of header is
disallowed.
For both:
* Check mandatory pseudo header fields.
* Make sure that content-length matches the amount of DATA we
received.
If validation fails, RST_STREAM of type PROTOCOL_ERROR is issued.
Previously if HTTP/1 proxy is used for backend connection, we read all
incoming bytes from proxy including response body, which may be part
of HTTP/2 protocol. While investigating this issue, we found that
http_parser_execute() returns 1-less length when we call
http_parser_pause() inside on_headers_complete callback. To
workaround this, we increment the return value by 1. This commit also
fixes possible segmentation fault error, which could be caused by the
lack of stopping libev watcher in disconnect().
Previously when requests are issued to HTTP/2 downstream connection,
but it turns out that connection is down, handlers of those requests
are deleted. In some situations, we only know connection is down when
we write something to network, so we'd like to handle this kind of
situation in more robust manner. In this change, certain seconds
passed after last network activity, we first issue PING frame to
downstream connection before issuing new HTTP request. If writing
PING frame is failed, it means connection was lost. In this case,
instead of deleting handler, pending requests are migrated to new
HTTP2/ downstream connection, so that it can continue without
affecting upstream connection.
For HTTP/1 backend, -b option can be used several times to specify
multiple backend address. HTTP/2 backend does not support multiple
addresses and only uses first address even if multiple addresses are
specified.
With the combination of HTTP/1 upstream and HTTP/2 downstream,
downstream tells SHRPX_NO_BUFFER while connecting to the backend
server. Previously, we did not call upstream resume_read and upload
was blocked. This commit now calls upstream resume_read to unblock.
This commit also remove pending output buffer size of Http2Session
when calculating downstream connection's buffer is full. This is
desirable since we only operate resume_read by stream basis.
Even after on_stream_close_callback, Http2DownstreamConnection is
still alive and upstream keeps sending response to the client. The
consumed bytes are processed normally (data_source_read_callback) and
also we have a code to consume all allocated bytes for
Http2DownstreamConnection object when it is deleted. This means that
we don't need to and should not consume response data in downstream
on_stream_close_callback. If we do, we may get assertion error in
Http2DownstreamConnection::resume_read().
It is not used by library for a while. It could be used to pass
unsupported extension frames to application, but its interface
requires library to buffer entire frame, which we'd like to avoid.
For unsupported extension frames, we will add new callbacks which does
not require buffering if they are required.
h2-14 now allows extensions to define new error codes. To allow
application callback to access such error codes, we uses uint32_t as
error_code type for structs and function parameters. Previously we
treated unknown error code as INTERNAL_ERROR, but this change removes
this and unknown error code is passed to application callback as is.
To make it possible to add new callbacks without bumping so name, we
decided to hide details of nghttp2_session_callbacks. We provide
setter like functions to set individual callback function.
Previously we only update consumed flow control window when number of
bytes read in nghttp2 and spdylay callback is 0. Now we notify
nghttp2 library the consumed bytes even if number of bytes read > 0.
This change also uses newly added spdylay_session_consume() API, so we
require spdylay >= 1.3.0.
libnghttp2 will call on_stream_close callback when RST_STREAM is
received. So we can use on_stream_close callback to handle existing
stream, instead of on_frame_recv callback.
nghttpx supports hot deploy feature using signals. The host deploy in
nghttpx is multi step process. First send USR2 signal to nghttpx
process. It will do fork and execute new executable, using same
command-line arguments and environment variables. At this point, both
current and new processes can accept requests. To gracefully shutdown
current process, send QUIT signal to current nghttpx process. When
all existing frontend connections are done, the current process will
exit. At this point, only new nghttpx process exists and serves
incoming requests.
Reworked no automatic WINDOW_UPDATE feature. We added new API
nghttp2_session_consume() which tells the library how many bytes are
consumed by the application. Instead of submitting WINDOW_UPDATE by
the application, the library is now responsible to submit
WINDOW_UPDATE based on consumed bytes. This is more reliable method,
since it enables us to properly send WINDOW_UPDATE for stream and
connection individually. The previous implementation of nghttpx had
broken connection window management.
This change rewrites logging system of nghttpx. Previously access log
and error log are written to stderr or syslog and there was no option
to change stderr to something else. With this change, file path of
access log and error log can be configured separately and logging to
regular file is now added. To support rotating log, if SIGUSR1 signal
is received by nghttpx, it closes the current log files and reopen it
with the same name. The format of access log is changed and has same
look of apache's. But not all columns are not supported yet.
Previously, we use evbuffer_pullup(buf, -1) to linearize the memory
region and it may cause buffer copy. To avoid this, we use the return
value of evbuffer_get_contiguous_space() as 2nd parameter. According
to the libevent manual, by doing so evbuffer_pullup() will not copy or
modify any data in evbuffer.
Previously stream ID was assigned just before HEADERS or PUSH_PROMISE
was serialized and nghttp2_submit_{request, headers, push_promise} did
not return stream ID. The application has to check assigned stream ID
using before_frame_send_callback. Now it is apparent that priority is
meant to DATA transfer only. Also application can reorder the
requests if it wants. Therefore we can assign stream ID in
nghttp2_submit_* functions and return stream ID from them. With this
change, now application does not have to check stream ID using
before_frame_send_callback and its code will be simplified.
To make adding new option easier, we decided to make the details of
option struct private and hide it from public API. We provide
functions to set individual option value.
If SPDY or HTTP/2 ustream is used and HTTP/2 downstream is used, only
call {spdylay,nghttp2}_resume_data when complete DATA frame was read
in backend to avoid to transmit too small DATA frame to the upstream.
NGHTTP2_CLIENT_CONNECTION_PREFACE has the same content with
NGHTTP2_CLIENT_CONNECTION_HEADER, which is now obsoleted by
NGHTTP2_CLIENT_CONNECTION_PREFACE.
The profiler and benchmarking showed that calling evbuffer_add()
repeatedly is very costly. To avoid this, we buffer up small writes
into one large chunk and call evbuffer_add() less times.
This function behaves like nghttp2_session_send(), but it does not
use nghttp2_send_callback to send data. Instead, it returns the
serialized data to trasmit and its length to the caller.
Previously, there is inconsistency when on_frame_recv_callback
is called between HEADERS/PUSH_PROMISE and the other frames.
For former case, it is called before header block, in latter
case, it is called after whole frame is received. To make it
consistent, we call on_frame_recv_callback for HEADERS/PUSH_PROMISE
after its frame is fully received. Since on_frame_recv_callback
can signal the end of header block, we replaced on_end_headers_callback
with on_begin_headers_callback, which is called when the reception
of the header block is started.
nghttp2 library itself now accept octet header/value pairs,
completely not restricted by HTTP/1 header name/value rule.
The applications may impose restriction about them using
validators.
Now, in nghttp2_on_frame_recv_callback, nva and nvlen in
HEADERS and PUSH_PROMISE frames are always NULL and 0 respectively.
The header name/value pairs are emitted successive
nghttp2_on_header_callback functions. The end of header fields are
signaled with nghttp2_on_end_headers_callback function.
Since NGHTTP2_ERR_PAUSE for nghttp2_on_frame_recv_callback is
introduced to handle header block, it is now deprecated.
Instead, nghttp2_on_header_callback can be paused using
NGHTTP2_ERR_PAUSE.
This commit also changes SPDY's flow control size. Previously,
the size for SPDY is the same amount of bytes with HTTP/2.
For example, --frontend-http2-upstream-window-bits=N,
the window size is 2**N - 1. Now SPDY code uses 2**N.
Tatsuhiro Tsujikawa