d25b65bb53
testing: add tests for Netty bug where a channel would break after a RST_STREAM frame. Fixes #1682
2016-06-02 18:34:07 +02:00
c6c74799f4
netty: add toString() to GrpcHttp2Headers
2016-06-02 18:18:55 +02:00
90323ad5d4
core: add custom CallOptions
2016-06-02 09:14:48 -07:00
2ed39c92eb
core: cache decompressor registry encodings, and make it copy on write
2016-06-02 09:10:42 -07:00
90506405c9
Differentiate transport to LB services and to servers.
...
TransportManager has a new method, createOobTransportProvider(), which
accepts an EquivalentAddressGroup and the authority string. This
addresses two requirements:
1. Per GRPCLB protocol, connections to the remote load-balancer may use a
different authority than the channel's (#1137 ).
2. For idle state determination, Channel needs to exclude the transport to
the LB service when looking at live RPCs and (#1276 ).
2016-06-01 18:23:09 -07:00
676bf4854f
core: fix nags from linter
2016-06-01 13:58:07 -07:00
02eb24b3bd
core,netty,okhttp: move user agent removal closer to where it is set
2016-06-01 12:55:21 -07:00
028d0844dd
core: add test for deframer closed exception. fixes #1795
2016-06-01 20:44:29 +02:00
134451f62c
benchmarks: use nextUp for nextDelay calculation in OpenLoopClient
2016-05-31 14:05:48 -07:00
c08d74fec4
TransportSet shutdown() also shuts down the pending transport.
...
Previously TransportSet.shutdown() only shuts down the active transport,
which means a transport will not be shutdown if it's not ready yet. This
issue was introduced by #1494 that postponed the assignment of the
active transport till transport ready.
2016-05-29 15:45:03 -07:00
9f37951680
benchmarks: honor transport in AsyncClient
2016-05-27 11:55:18 -07:00
361936401f
core/internal: fix regex pattern
2016-05-26 14:02:15 -07:00
7052d8b5e9
core: Promote Deadline API to stable
...
Fixes #1706
2016-05-26 13:33:48 -07:00
fa22259e3a
netty: use custom Http2Headers class for encoding Metadata
...
Before:
Benchmark (headerCount) Mode Cnt Score Error Units
HeadersBenchmark.convertClientHeaders 10 sample 127008 631.214 ± 3.543 ns/op
HeadersBenchmark.convertClientHeaders 20 sample 142036 1125.874 ± 21.114 ns/op
HeadersBenchmark.convertClientHeaders 50 sample 117570 2678.635 ± 47.764 ns/op
HeadersBenchmark.convertClientHeaders 100 sample 115919 5427.720 ± 67.956 ns/op
HeadersBenchmark.convertServerHeaders 10 sample 133903 610.970 ± 3.094 ns/op
HeadersBenchmark.convertServerHeaders 20 sample 138155 1154.304 ± 4.595 ns/op
HeadersBenchmark.convertServerHeaders 50 sample 120078 2658.175 ± 38.679 ns/op
HeadersBenchmark.convertServerHeaders 100 sample 120509 5212.341 ± 49.062 ns/op
After:
Benchmark (headerCount) Mode Cnt Score Error Units
HeadersBenchmark.convertClientHeaders 10 sample 102473 407.383 ± 2.693 ns/op
HeadersBenchmark.convertClientHeaders 20 sample 103205 791.241 ± 38.054 ns/op
HeadersBenchmark.convertClientHeaders 50 sample 173817 1840.311 ± 5.718 ns/op
HeadersBenchmark.convertClientHeaders 100 sample 169984 3690.753 ± 44.308 ns/op
HeadersBenchmark.convertServerHeaders 10 sample 103615 401.661 ± 2.922 ns/op
HeadersBenchmark.convertServerHeaders 20 sample 99060 823.453 ± 5.553 ns/op
HeadersBenchmark.convertServerHeaders 50 sample 171824 1846.788 ± 29.840 ns/op
HeadersBenchmark.convertServerHeaders 100 sample 171622 3670.354 ± 7.127 ns/op
2016-05-25 17:23:51 -07:00
149ce204e2
netty: Bump netty dep to 4.1.0.Final
...
Is such a thing even possible? Yes it is.
Note that HTTP/2 in Netty still is not stable, so we remain
version-pinned.
2016-05-25 16:58:51 -07:00
ca8e166591
okhttp: generate user agent once for whole transport
2016-05-25 16:37:54 -07:00
f172170649
benchmarks: use more realistic header counts in benchmark
2016-05-25 16:30:29 -07:00
4204eb8cd9
core: allocate less garbage iterators in Metadata.serialize
...
Before:
HeadersBenchmark.convertClientHeaders 10 sample 99004 806.749 ± 4.692 ns/op
HeadersBenchmark.convertClientHeaders 20 sample 105673 1540.814 ± 86.361 ns/op
HeadersBenchmark.convertClientHeaders 50 sample 188978 3347.881 ± 71.778 ns/op
HeadersBenchmark.convertClientHeaders 100 sample 189913 6626.884 ± 94.348 ns/op
HeadersBenchmark.convertServerHeaders 10 sample 107884 766.328 ± 49.338 ns/op
HeadersBenchmark.convertServerHeaders 20 sample 122568 1303.501 ± 8.820 ns/op
HeadersBenchmark.convertServerHeaders 50 sample 109281 2920.586 ± 93.654 ns/op
HeadersBenchmark.convertServerHeaders 100 sample 196784 6429.566 ± 108.588 ns/op
After:
HeadersBenchmark.convertClientHeaders 10 sample 113274 715.382 ± 5.412 ns/op
HeadersBenchmark.convertClientHeaders 20 sample 128654 1294.677 ± 67.867 ns/op
HeadersBenchmark.convertClientHeaders 50 sample 112598 2814.925 ± 62.291 ns/op
HeadersBenchmark.convertClientHeaders 100 sample 116920 5383.146 ± 90.205 ns/op
HeadersBenchmark.convertServerHeaders 10 sample 130004 626.243 ± 3.528 ns/op
HeadersBenchmark.convertServerHeaders 20 sample 142054 1185.193 ± 80.272 ns/op
HeadersBenchmark.convertServerHeaders 50 sample 115153 2795.715 ± 92.956 ns/op
HeadersBenchmark.convertServerHeaders 100 sample 119520 5249.636 ± 56.089 ns/op
2016-05-25 15:12:40 -07:00
1cc76d8132
core,netty,okhttp: move user agent out of client call and into the transport
2016-05-25 15:11:42 -07:00
c102dd4e4f
Add trailers to Status for BlockingResponseStream
...
Fixes #1861
2016-05-25 12:04:55 -07:00
6e40edbc21
Add trailers to Status for BlockingResponseStream
2016-05-25 06:54:31 -07:00
641cb357c6
Tweak -Xlint warnings
...
This now catches a few more places we needed -Xlint:-options.
InProcessSocketAddress is technically already in our stable API, so I
maintained its current serialVersionUID.
2016-05-24 15:04:07 -07:00
e4ea237597
core: Refactor TransportSet in prep for fail fast
...
This is functionally equivalent. Some tests needed changes because they
were verifying internal behavior.
2016-05-24 14:40:07 -07:00
cca473df99
benchmarks: add client/server header benchmarks
2016-05-24 13:11:14 -07:00
5e30b2f7ba
netty: speed up header conversion by caching user agent string
...
Performance gain seems to be about 100-200ns based on a a couple trials.
Benchmark (headerCount) (validate) Mode Cnt Score Error Units
HeadersBenchmark.convertClientHeadersOld 10 false sample 187490 858.234 ± 3.992 ns/op
HeadersBenchmark.convertClientHeadersOld 20 false sample 113589 1407.557 ± 45.178 ns/op
HeadersBenchmark.convertClientHeadersOld 50 false sample 100725 3141.936 ± 55.175 ns/op
HeadersBenchmark.convertClientHeadersOld 100 false sample 109742 5707.748 ± 38.222 ns/op
HeadersBenchmark.convertHeaders 10 false sample 109137 748.486 ± 4.060 ns/op
HeadersBenchmark.convertHeaders 20 false sample 133639 1238.528 ± 51.914 ns/op
HeadersBenchmark.convertHeaders 50 false sample 107914 2915.602 ± 10.017 ns/op
HeadersBenchmark.convertHeaders 100 false sample 110305 5682.404 ± 44.032 ns/op
2016-05-23 17:49:30 -07:00
ac9a5a5679
Make Code.status() public, add Status#fromCode()
...
Closes #1722
2016-05-23 17:03:48 -07:00
8033cdd3a7
core: Add unit test for Contexts.interceptCall
2016-05-23 13:51:10 -07:00
39d1e31547
netty: Fix test to support netty 4.1.0.Final
2016-05-23 13:22:18 -07:00
6382015f9d
internal: Split-state AbstractStream; sending and receiving
...
This introduces an AbstractStream2 that is intended to replace the
current AbstractStream. Only server-side is implemented in this commit
which is why AbstractStream remains. This is mostly a reorganization of
AbstractStream and children, but minor internal behavioral changes were
required which makes it appear more like a reimplementation.
A strong focus was on splitting state that is maintained on the
application's thread (with Stream) and state that is maintained by the
transport (and used for StreamListener). By splitting the state it makes
it much easier to verify thread-safety and to reason about interactions.
I consider this a stepping stone for making even more changes to
simplify the Stream implementations and do not think some of the changes
are yet at their logical conclusion. Some of the changes may also
immediately be replaced with something better. The focus was to improve
readability and comprehesibility to more easily make more interesting
changes.
The only thing really removed is some state checking during sending
which is already occurring in ServerCallImpl.
2016-05-23 12:33:49 -07:00
8090effa2d
netty: don't revalidate when converting between Metadata and Http2Headers
2016-05-23 11:02:22 -07:00
b7eea7c13a
implements health check service
...
resolves #647
2016-05-20 14:03:29 -07:00
65dd5db5e3
protolite: Use 'unused' variable to avoid CheckReturnValue
...
Internally toByteArray is annotated with CheckReturnValue, which can
cause a failure during compilation if the value is ignored.
2016-05-20 09:46:46 -07:00
aa92d6c643
build: Update jetty for JDK 8u92
2016-05-19 11:43:48 -07:00
8c7440e141
benchmarks: add NETTY_EPOLL as an option for transport testing
2016-05-18 16:37:42 -07:00
e19848092d
core, interop-testing: always set message encoding, and use it in interop tests
...
This change updates the behavior of the core compression semantics. Previously,
if the codec was "identity", nothing was set on the wire. This is allowed by
the spec, but doesn't match what wrapped languages do.
Additionally, the interop tests will now attempt to honor the requested
compression.
2016-05-18 12:04:25 -07:00
6a802a2027
benchmarks: revert use of CMS GC
2016-05-17 10:38:59 -07:00
d08c167bc7
benchmarks: use Concurrent Mark and Sweep GC for benchmark worker
2016-05-16 16:44:58 -07:00
5a2e5e16a2
android-interop-test: Enable checkstyle; fix violations
2016-05-16 12:44:15 -07:00
2825361767
stub: use correct javadoc link for Guava
2016-05-16 10:25:03 -07:00
a70e1a4d8b
interop: Don't catch AssertionError as it makes test failure confusing
2016-05-16 09:52:48 -07:00
13fe13bf9f
okhttp: Enable transport test
...
OkHttpClientTransport has a fix for shutdown during start which
prevented transportTerminated from being called. It also no longer fails
pending streams during shutdown. Lifecycle management in general was
revamped to be hopefully simpler and more precise. In the process GOAWAY
handling (both sending and receiving) was improved.
With some of the changes, the log spam generated was immense and
unhelpful (since many exceptions are part of normal operation on
shutdown). This change reduces the amount of log spam to nothing.
2016-05-16 09:52:48 -07:00
188840b920
benchmarks: update to JMH 1.12
2016-05-13 21:25:31 -07:00
4e268da29b
core: Don't lose server call exception when message fails to close
2016-05-13 15:59:06 -07:00
951b5a4ca2
Adding metadata to Status exceptions.
...
Fixes #681
2016-05-13 15:52:06 -07:00
27dffdfe03
netty: call transportShutdown immediately on ID exhaustion
...
Fixes #1819
2016-05-13 13:39:47 -07:00
ce002bd449
core: allow ClientCall.cancel before start. Fixes #1536 ( #1822 )
2016-05-13 22:20:02 +02:00
f13fbaa782
interop-testing: update testing protos to include compression fields
2016-05-13 11:00:33 -07:00
46edcad0c1
Don't enforce a timeout in stress test client. Fixes #1812 ( #1823 )
2016-05-13 19:18:13 +02:00
4e8cf923ca
docs: Don't suggest downloading grpc-all
...
grpc-all is now just a husk of a JAR. Instead, users will need to
download each JAR they need separately.
Fixes #1805
2016-05-12 13:00:47 -07:00
f85ecdfbc4
interop-testing: Print JVM Flags by default for Stress tests
2016-05-11 13:20:23 -07:00
Jakob Buchgraber
elandau
Carl Mastrangelo
Kun Zhang
ZHANG Dapeng
Eric Anderson
nmittler
Łukasz Strzałkowski