Steve Jenson

gRPC unary throughput and latency issues

1

We are getting reports of linkerd adding 10ms latency overhead at p50 with high concurrency. I'm not able to reproduce the 10ms overhead at p50 but I have been able...

Chunked encoding with lots of small chunks appear to cause throughput issues on small core hosts

2

In trying to nail down why tests with echoheaders is problematic, @siggy and I decided to replicate the configuration of echoheaders: an nginx process with a built-in lua script. What...

gRPC interop: compute_engine_creds test fails

1

``` Jan 05, 2017 8:48:33 PM io.grpc.internal.ManagedChannelImpl INFO: [io.grpc.internal.ManagedChannelImpl-1] Created with target directaddress:///proxy-test-4b/10.240.0.10: 4141 Running test compute_engine_creds Exception in thread "main" java.lang.AssertionError: expected: but was: at org.junit.Assert.fail(Assert.java:88) at org.junit.Assert.failNotEquals(Assert.java:743) at...

``` /Users/stevej% ./go-grpc-interop-client -use_tls=false -server_port=60011 -test_case=unimplemented_service 2017/01/25 12:59:27 transport: http2Client.notifyError got notified that the client transport was broken EOF. 2017/01/25 12:59:27 &{0xc420180500}.UnimplementedCall() = _, Internal, want _, Unimplemented ``` Server...

`/shutdown` on the admin port no longer results in the process shutting down

1

``` $ curl -X POST -vvvv http://127.0.0.1:9989/shutdown * Trying 127.0.0.1... * TCP_NODELAY set * Connected to 127.0.0.1 (127.0.0.1) port 9989 (#0) > POST /shutdown HTTP/1.1 > Host: 127.0.0.1:9989 > User-Agent:...

Periodic waves of EOF every 60 seconds.

3

testing linkerd-tcp with the stock example.yml and slow_cooker at 20k qps is resulting in large waves of EOFs. ``` Get http://proxy-test-4d:7474: EOF [96/25466] Get http://proxy-test-4d:7474: EOF Get http://proxy-test-4d:7474: EOF Get...

Add more client stats

3

Debugging the mystery of the failing iperf tests would have been easier had `/metrics` shown me that there were other clients connected and where they were connected from.

linkerd-tcp connection reuse issue

1

Running a load test without connection reuse (this is part of my testing out the tacho OrderMap changes) and linkerd-tcp starts rejecting traffic after a thousand requests. ``` stevej@netty-test-8:~$ ./slow_cooker_linux_amd64...

Adds failfast mode for unroutable requests.

3

* Fixes Issue #26. * Adds a failfast mode for currently unroutable requests. * Adds an integration test for failfast mode. * Adds a mock namerd. * Adds a mock...

*Problem* 'Too many open files' leaves linkerd-tcp in a bad state rather than gracefully exiting. Symptoms: In a shell with a `ulimit -n 1024`, if I open 500 connections (which...