a12250e98f
xds/cds: add env var for aggregated and DNS cluster ( #4440 )
2021-05-14 15:20:45 -07:00
50c071e9b5
example: correct the default value for server_host_override ( #4407 )
2021-05-14 14:09:26 -07:00
b759b408e8
xds: moved shared matchers to internal/xds ( #4441 )
...
* Moved shared matchers to internal/xds
2021-05-14 17:02:10 -04:00
71a1ca6c7f
interop/xds: support xds credentials in interop client ( #4436 )
2021-05-14 11:13:26 -07:00
dc77d7ffe3
xds: revert a workaround made in #4413 ( #4428 )
2021-05-12 18:03:52 -07:00
a16b156e99
internal: fix flaky test KeepaliveClientStaysHealthyWithResponsiveServer ( #4427 )
...
Server should allow `NoStream`, otherwise there's a small chance
(5/1000) the connection will be closed due to `too many pings`.
2021-05-12 17:43:29 -07:00
6fea90d7a8
benchmark: do not allow addition of values lower than the minimum allowed in histogram stats
2021-05-12 17:15:47 -07:00
a712a73889
xds/cds: add separate fields for cluster name and eds service name ( #4414 )
2021-05-12 16:25:07 -07:00
397adad6a0
update go.mod and go.sum to point to latest go-control-plane ( #4425 )
2021-05-12 15:52:15 -07:00
9cb99a5211
xds: pretty print xDS updates and service config ( #4405 )
2021-05-12 15:48:16 -07:00
45e60095da
xds: add support for aggregate clusters ( #4332 )
...
Add support for aggregate clusters in CDS Balancer
2021-05-12 17:28:49 -04:00
8bf65c69b9
xds: use same format while registering and watching resources ( #4422 )
2021-05-12 10:18:50 -07:00
aa59641d5d
interop: use credentials.NewTLS() when possible ( #4390 )
2021-05-12 10:17:13 -07:00
a95a5c3bac
transport: remove decodeState from client to reduce allocations ( #3313 )
2021-05-12 08:49:07 -07:00
62adda2ece
client: fix ForceCodec to set content-type header appropriately ( #4401 )
2021-05-11 17:05:16 -07:00
81b8cca6a9
Change version to 1.39.0-dev ( #4420 )
2021-05-11 15:28:46 -07:00
5f95ad6233
xds: workaround to deflake xds e2e tests ( #4413 )
2021-05-11 10:39:31 -07:00
b1940e15f6
xds: register resources at the mgmt server before requesting them ( #4406 )
2021-05-10 10:11:31 -07:00
98c895f7e0
cleanup: use testutils.MarshalAny in more places ( #4404 )
2021-05-10 09:35:55 -07:00
12a377b1e4
xds: nack route configuration with regexes that don't compile ( #4388 )
2021-05-07 15:42:59 -07:00
c15291b0f5
client: initialize safe config selector when creating ClientConn ( #4398 )
2021-05-07 15:24:10 -07:00
328b1d171a
transport: allow InTapHandle to return status errors ( #4365 )
2021-05-07 14:37:52 -07:00
aff517ba8a
xds: make e2e tests use a single management server instance ( #4399 )
2021-05-07 14:35:48 -07:00
0439465fe2
xds_resolver: fix flaky Test/XDSResolverDelayedOnCommitted ( #4393 )
...
Before this change, if two xds client updates came too close together, the second one could replace the first one. The fix is to wait for the effects of the first update before sending the second update. I injected a synthetic delay into handling the updates from the channel to reproduce this flake 100%, and confirmed this change fixes it.
As part of this change I also noticed that we're actually calling the context cancellation function twice via defers, and never the cancel function from the test setup, so I fixed that, too.
2021-05-07 11:57:56 -07:00
0ab423af82
test: fix flaky GoAwayThenClose ( #4394 )
...
In this test, we
1. make a streaming RPC on a connection
1. graceful stop it to send a GOAWAY
1. hard stop it, so the client will create a connection to another server
Before this fix, 2 and 3 can happen too soon, so the RPC in 1 would fail
and then transparent retry (because the stream is unprocessed by the
server in that case).
This retry attempt could pick the new connection, and then the RPC would
block until timeout.
After this streaming RPC fails, we make unary RPCs with the same
deadline (note: deadline not timeout) as the streaming RPC and expect
them to succeed. But they will also fail due to timeout.
The fix is to make a round-trip on the streaming RPC first, to make sure
it actually goes on the first connection.
2021-05-07 11:55:48 -07:00
b6f206b84f
grpc: improve docs on StreamDesc ( #4397 )
2021-05-07 11:17:26 -07:00
c7ea734087
dns: fix flaky TestRateLimitedResolve ( #4387 )
...
* Rewrote TestRateLimitedResolve in dns resolver test to get rid of flakiness.
2021-05-07 08:28:34 -04:00
cb396472c2
Revert "grpc: call balancer.Close() before returning from ccBalancerWrapper.close()" ( #4391 )
...
This reverts commit 28078834f3
2021-05-06 13:28:27 -07:00
d2d6bdae07
server: add ForceServerCodec() to set a custom encoding.Codec on the server ( #4205 )
2021-05-06 09:40:54 -07:00
d426aa5f2e
test: extend the xDS interop tests timeout to 360 mins ( #4380 )
2021-05-05 13:37:13 -07:00
40b25c5b2c
xds: set correct order of certificate providers in handshake info ( #4350 )
2021-05-05 12:34:15 -07:00
0fc0397d77
xds: actually close stuff in cds/eds `Close()` ( #4381 )
2021-05-04 16:54:57 -07:00
4f3aa7cfa1
xds: optimize fault injection filter with empty config ( #4367 )
2021-05-04 15:29:58 -07:00
79e55d6444
xds: use SendContext() to fail in time when the channel is full ( #4386 )
2021-05-04 15:06:43 -07:00
11bd77660d
xds: work around xdsclient race in fault injection test ( #4377 )
2021-05-04 14:51:32 -07:00
75497df97f
meshca: remove meshca certificate provider implementation ( #4385 )
2021-05-04 14:38:47 -07:00
ebd6aba675
Revert "xds/cds: add separate fields for cluster name and eds service name" ( #4382 )
...
This reverts PRs #4352 (and two follow up fixes #4372 #4378 ).
Because the xds interop tests were flaky.
Revert before the branch cut.
2021-05-03 15:16:49 -07:00
b418de839e
xds/eds: restart EDS watch after previous was canceled ( #4378 )
2021-04-30 11:53:31 -07:00
28078834f3
grpc: call balancer.Close() before returning from ccBalancerWrapper.close() ( #4364 )
2021-04-29 21:44:26 -07:00
aa3ef8fb8f
internal: regenerate proto ( #4373 )
2021-04-29 12:17:56 -07:00
c3b66015bd
xds/circuit_breaking: use cluster name as key, not EDS service name ( #4372 )
2021-04-29 11:56:50 -07:00
91d8f0c916
serviceconfig: support marshalling BalancerConfig to JSON ( #4368 )
2021-04-28 18:11:45 -07:00
b602d17e45
metadata: reduce memory footprint in FromOutgoingContext ( #4360 )
...
When Looking at memory profiles for cockroachdb/cockroach, we observed
that the intermediate metadata.MD array constructed to iterate over
appended metadata escaped to the heap. Fortunately, this is easily
rectifiable.
go build -gcflags '-m' google.golang.org/grpc/metadata
...
google.golang.org/grpc/metadata/metadata.go:198:13: make([]MD, 0, len(raw.added) + 1) escapes to heap
2021-04-28 10:05:50 -07:00
24d03d9f76
xds/priority: add ignore reresolution boolean to config ( #4275 )
2021-04-27 15:22:25 -07:00
7c5e73795d
xds/cds: add separate fields for cluster name and eds service name ( #4352 )
2021-04-27 13:37:48 -07:00
145f12a95b
reflection: accept interface instead of grpc.Server struct in Register() ( #4340 )
2021-04-27 13:15:08 -07:00
52a707c0da
xds: serving mode changes outlined in gRFC A36 ( #4328 )
2021-04-26 14:29:06 -07:00
9572fd6fae
client: include details about GOAWAYs in status messages ( #4316 )
2021-04-23 17:26:26 -07:00
e158e3e82c
xds/lrs: server name is not required to be non-empty ( #4356 )
2021-04-23 13:15:21 -07:00
74fe6eaa41
github: testing action workflow improvements and update to test Go1.16 ( #4358 )
2021-04-22 14:59:51 -07:00
Menghan Li
Zeke Lu
Zach Reyes
Easwar Swaminathan
Mayank Singhal
James Protzman
Doug Fawley
Mikhail Mazurskiy
Lidi Zheng
irfan sharif
Joshua Humphries
apolcyn