Permalink
Fetching contributors…
Cannot retrieve contributors at this time
381 lines (274 sloc) 18.1 KB

Build Status

GoRouter

This repository contains the source code for the Cloud Foundry L7 HTTP router. GoRouter is deployed by default with Cloud Foundry (cf-release) which includes routing-release as submodule.

Note: This repository should be imported as code.cloudfoundry.org/gorouter.

Development

The following instructions may help you get started with gorouter.

Prerequisites

Setup

GoRouter is part of routing-release.

git clone https://github.com/cloudfoundry-incubator/routing-release
cd routing-release
./scripts/update
cd src/code.cloudfoundry.org/gorouter

Note: direnv will automatically set your GOPATH when you cd into the routing-release directory. You will need to run direnv allow the first time.

Running Tests

We are using Ginkgo, to run tests.

Running scripts/test will:

  • Checks for Go
  • Checks that GOPATH is set
  • Installs gnatsd and ginkgo (or use the one already downloaded into the GOPATH)
  • Runs all the tests with ginkgo (in random order, without benchmarks)

Any flags passed into scripts/test will be passed into ginkgo.

# run all the tests
scripts/test

# run only tests whose names match Registry
scripts/test -focus=Registry

# run only the tests in the registry package
scripts/test registry

Building

Building creates an executable in the gorouter/ dir:

go build

Installing

Installing creates an executable in the $GOPATH/bin dir:

go install

Start

# Start NATS server in daemon mode
go get github.com/nats-io/gnatsd
gnatsd &

# Start gorouter
gorouter

Performance

See Routing Release 0.144.0 Release Notes

Dynamic Routing Table

Gorouters routing table is updated dynamically via the NATS message bus. NATS can be deployed via BOSH with (cf-release) or standalone using nats-release.

To add or remove a record from the routing table, a NATS client must send register or unregister messages. Records in the routing table have a maximum TTL of 120 seconds, so clients must heartbeat registration messages periodically; we recommend every 20s. Route Registrar is a BOSH job that comes with Routing Release that automates this process.

When deployed with Cloud Foundry, registration of routes for apps pushed to CF occurs automatically without user involvement. For details, see Routes and Domains.

Registering Routes via NATS

When the gorouter starts, it sends a router.start message to NATS. This message contains an interval that other components should then send router.register on, minimumRegisterIntervalInSeconds. It is recommended that clients should send router.register messages on this interval. This minimumRegisterIntervalInSeconds value is configured through the start_response_delay_interval configuration property. GoRouter will prune routes that it considers to be stale based upon a seperate "staleness" value, droplet_stale_threshold, which defaults to 120 seconds. GoRouter will check if routes have become stale on an interval defined by prune_stale_droplets_interval, which defaults to 30 seconds. All of these values are represented in seconds and will always be integers.

The format of the router.start message is as follows:

{
  "id": "some-router-id",
  "hosts": ["1.2.3.4"],
  "minimumRegisterIntervalInSeconds": 20,
  "prunteThresholdInSeconds": 120,
}

After a router.start message is received by a client, the client should send router.register messages. This ensures that the new router can update its routing table and synchronize with existing routers.

If a component comes online after the router, it must make a NATS request called router.greet in order to determine the interval. The response to this message will be the same format as router.start.

The format of the router.register message is as follows:

{
  "host": "127.0.0.1",
  "port": 4567,
  "uris": [
    "my_first_url.vcap.me",
    "my_second_url.vcap.me"
  ],
  "tags": {
    "another_key": "another_value",
    "some_key": "some_value"
  },
  "app": "some_app_guid",
  "stale_threshold_in_seconds": 120,
  "private_instance_id": "some_app_instance_id"
}

stale_threshold_in_seconds is the custom staleness threshold for the route being registered. If this value is not sent, it will default to the router's default staleness threshold.

app is a unique identifier for an application that the endpoint is registered for. This value will be included in router access logs with the label app_id, as well as being sent with requests to the endpoint in an HTTP header X-CF-ApplicationId.

private_instance_id is a unique identifier for an instance associated with the app identified by the app field. Gorouter includes an HTTP header X-CF-InstanceId set to this value with requests to the registered endpoint.

Such a message can be sent to both the router.register subject to register URIs, and to the router.unregister subject to unregister URIs, respectively.

Example

Create a simple app

$ nohup ruby -rsinatra -e 'get("/") { "Hello!" }' &

Send a register message

$ nats-pub 'router.register' '{"host":"127.0.0.1","port":4567,"uris":["my_first_url.vcap.me","my_second_url.vcap.me"],"tags":{"another_key":"another_value","some_key":"some_value"}}'

Published [router.register] : '{"host":"127.0.0.1","port":4567,"uris":["my_first_url.vcap.me","my_second_url.vcap.me"],"tags":{"another_key":"another_value","some_key":"some_value"}}'

See that it works!

$ curl my_first_url.vcap.me:8081
Hello!

Healthchecking from a Load Balancer

To scale GoRouter horizontally for high-availability or throughput capacity, you must deploy it behind a highly-available load balancer (F5, AWS ELB, etc).

GoRouter has a health endpoint /health on port 8080 that returns a 200 OK which indicates the GoRouter instance is healthy; any other response indicates unhealthy. This port can be configured via the router.status.port property in the BOSH deployment manifest or via the status.port property under /var/vcap/jobs/gorouter/config/gorouter.yml

$ curl -v http://10.0.32.15:8080/health
*   Trying 10.0.32.15..
* Connected to 10.0.32.15 (10.0.32.15) port 8080 (#0)
> GET /health HTTP/1.1
> Host: 10.0.32.15:8080
> User-Agent: curl/7.43.0
> Accept: */*
>
< HTTP/1.1 200 OK
< Cache-Control: private, max-age=0
< Expires: 0
< Date: Thu, 22 Sep 2016 00:13:54 GMT
< Content-Length: 3
< Content-Type: text/plain; charset=utf-8
<
ok
* Connection #0 to host 10.0.32.15 left intact

DEPRECATED: Your load balancer can be configured to send an HTTP healthcheck on port 80 with the User-Agent HTTP header set to HTTP-Monitor/1.1. A 200 response indicates the GoRouter instance is healthy; any other response indicates unhealthy. GoRouter can be configured to accept alternate values for the User Agent header using the healthcheck_user_agent configuration property; as an example, AWS ELBS send User-Agent: ELB-HealthChecker/1.0.

$ curl -v -A "HTTP-Monitor/1.1" "http://10.0.32.15"
* Rebuilt URL to: http://10.0.32.15/
* Hostname was NOT found in DNS cache
*   Trying 10.0.32.15...
* Connected to 10.0.32.15 (10.0.32.15) port 80 (#0)
> GET / HTTP/1.1
> User-Agent: HTTP-Monitor/1.1
> Host: 10.0.32.15
> Accept: */*
>
< HTTP/1.1 200 OK
< Cache-Control: private, max-age=0
< Expires: 0
< X-Vcap-Request-Id: 04ad84c6-43dd-4d20-7818-7c47595d9442
< Date: Thu, 07 Jan 2016 22:30:02 GMT
< Content-Length: 3
< Content-Type: text/plain; charset=utf-8
<
ok
* Connection #0 to host 10.0.32.15 left intact

DEPRECATED: The /healthz endpoint provides a similar response, but it always returns a 200 response regardless of whether or not the GoRouter instance is healthy.

Instrumentation

The Routing Table

The /routes endpoint returns the entire routing table as JSON. This endpoint requires basic authentication and is served on port 8080. Each route has an associated array of host:port entries.

$ curl "http://someuser:somepass@10.0.32.15:8080/routes"
{"api.catwoman.cf-app.com":[{"address":"10.244.0.138:9022","ttl":0,"tags":{"component":"CloudController"}}],"dora-dora.catwoman.cf-app.com":[{"address":"10.244.16.4:60035","ttl":0,"tags":{"component":"route-emitter"}},{"address":"10.244.16.4:60060","ttl":0,"tags":{"component":"route-emitter"}}]}

Because of the nature of the data present in /varz and /routes, they require http basic authentication credentials. These credentials can be found the BOSH manifest for cf-release under the router job:

properties:
  router:
    status:
      password: zed292_bevesselled
      port:
      user: paronymy61-polaric

Or on the Gorouter VM under /var/vcap/jobs/gorouter/config/gorouter.yml:

status:
  port: 8080
  user: some_user
  pass: some_password

Metrics

The /varz endpoint provides status and metrics. This endpoint requires basic authentication.

$ curl "http://someuser:somepass@10.0.32.15:8080/varz"
{"bad_gateways":0,"bad_requests":20,"cpu":0,"credentials":["user","pass"],"droplets":26,"host":"10.0.32.15:8080","index":0,"latency":{"50":0.001418144,"75":0.00180639025,"90":0.0070607187,"95":0.009561058849999996,"99":0.01523927838000001,"samples":1,"value":5e-07},"log_counts":{"info":9,"warn":40},"mem":19672,"ms_since_last_registry_update":1547,"num_cores":2,"rate":[1.1361328993362565,1.1344545494448148,1.1365784133171992],"requests":13832,"requests_per_sec":1.1361328993362565,"responses_2xx":13814,"responses_3xx":0,"responses_4xx":9,"responses_5xx":0,"responses_xxx":0,"start":"2016-01-07 19:04:40 +0000","tags":{"component":{"CloudController":{"latency":{"50":0.009015199,"75":0.0107408015,"90":0.015104917100000005,"95":0.01916497394999999,"99":0.034486261410000024,"samples":1,"value":5e-07},"rate":[0.13613289933245148,0.13433569936308343,0.13565885617276216],"requests":1686,"responses_2xx":1684,"responses_3xx":0,"responses_4xx":2,"responses_5xx":0,"responses_xxx":0},"HM9K":{"latency":{"50":0.0033354,"75":0.00751815875,"90":0.011916812100000005,"95":0.013760064,"99":0.013760064,"samples":1,"value":5e-07},"rate":[1.6850238803894876e-12,5.816129919395257e-05,0.00045864309255845694],"requests":12,"responses_2xx":6,"responses_3xx":0,"responses_4xx":6,"responses_5xx":0,"responses_xxx":0},"dea-0":{"latency":{"50":0.001354994,"75":0.001642107,"90":0.0020699939000000003,"95":0.0025553900499999996,"99":0.003677146940000006,"samples":1,"value":5e-07},"rate":[1.0000000000000013,1.0000000002571303,0.9999994853579043],"requests":12103,"responses_2xx":12103,"responses_3xx":0,"responses_4xx":0,"responses_5xx":0,"responses_xxx":0},"uaa":{"latency":{"50":0.038288465,"75":0.245610809,"90":0.2877324668,"95":0.311816554,"99":0.311816554,"samples":1,"value":5e-07},"rate":[8.425119401947438e-13,2.9080649596976205e-05,0.00022931374141467497],"requests":17,"responses_2xx":17,"responses_3xx":0,"responses_4xx":0,"responses_5xx":0,"responses_xxx":0}}},"top10_app_requests":[{"application_id":"063f95f9-492c-456f-b569-737f69c04899","rpm":60,"rps":1}],"type":"Router","uptime":"0d:3h:22m:31s","urls":21,"uuid":"0-c7fd7d76-f8d8-46b7-7a1c-7a59bcf7e286"}

Profiling the Server

The GoRouter runs the debugserver, which is a wrapper around the go pprof tool. In order to generate this profile, do the following:

# Establish a SSH tunnel to your server (not necessary if you can connect directly)
ssh -L localhost:8080:[INTERNAL_SERVER_IP]:17001 vcap@[BOSH_DIRECTOR]
# Run the profile tool.
go tool pprof http://localhost:8080/debug/pprof/profile

Load Balancing

The GoRouter is, in simple terms, a reverse proxy that load balances between many backend instances. The default load balancing algorithm that GoRouter will use is a simple round-robin strategy. GoRouter will retry a request if the chosen backend does not accept the TCP connection.

Round-Robin

Default load balancing algorithm that gorouter will use or may be explicity set in gorouter.yml

default_balancing_algorithm: round-robin

Least-Connection

The GoRouter also supports least connection based routing and this can be enabled in gorouter.yml

default_balancing_algorithm: least-connection

Least connection based load balancing will select the endpoint with the least number of connections. If multiple endpoints match with the same number of least connections, it will select a random one within those least connections.

NOTE: GoRouter currently only supports changing the load balancing strategy at the gorouter level and does not yet support a finer-grained level such as route-level. Therefore changing the load balancing algorithm from the default (round-robin) should be proceeded with caution.

When terminating TLS in front of Gorouter with a component that does not support sending HTTP headers

Enabling apps and CF to detect that request was encrypted using X-Forwarded-Proto

If you terminate TLS in front of Gorouter, your component should send the X-Forwarded-Proto HTTP header in order for applications and Cloud Foundry system components to correctly detect when the original request was encrypted. As an example, UAA will reject requests that do not include X-Forwarded-Proto: https.

If your TLS-terminating component does not support sending HTTP headers, we recommend also terminating TLS at Gorouter. In this scenario you should only disable TLS at Gorouter if your TLS-terminating component rejects unencrypted requests and your private network is completely trusted. In this case, use the following property to inform applications and CF system components that requests are secure.

properties:
  router:
    force_forwarded_proto_https: true

Enabling apps to detect the requestor's IP address uing PROXY Protocol

If you terminate TLS in front of Gorouter, your component should also send the X-Forwarded-Proto HTTP header in order for X-Forwarded-For header to applications can detect the requestor's IP address.

If your TLS-terminating component does not support sending HTTP headers, you can use the PROXY protocol to send Gorouter the requestor's IP address.

If your TLS-terminating component supports the PROXY protocol, enable the PROXY protocol on Gorouter using the following cf-release manifest property:

properties:
  router:
    enable_proxy: true

You can test this feature manually:

echo -e "PROXY TCP4 1.2.3.4 [GOROUTER IP] 12345 [GOROUTER PORT]\r\nGET / HTTP/1.1\r\nHost: [APP URL]\r\n" | nc [GOROUTER IP] [GOROUTER PORT]

You should see in the access logs on the GoRouter that the X-Forwarded-For header is 1.2.3.4. You can read more about the PROXY Protocol here.

HTTP/2 Support

The GoRouter does not currently support proxying HTTP/2 connections, even over TLS. Connections made using HTTP/1.1, either by TLS or cleartext, will be proxied to backends over cleartext.

Logs

The router's logging is specified in its YAML configuration file. It supports the following log levels:

  • fatal - A fatal error has occurred that makes gorouter unable to handle any requests. Examples: the router can't bind to its TCP port, a CF component has published invalid data to the router.
  • error - An unexpected error has occurred. Examples: the router failed to fetch token from UAA service.
  • info, debug - An expected event has occurred. Examples: a new CF component was registered with the router, the router has begun to prune routes for stale droplets.

Sample log message in gorouter.

[2017-02-01 22:54:08+0000] {"log_level":0,"timestamp":1485989648.0895808,"message":"endpoint-registered","source":"vcap.gorouter.registry","data":{"uri":"0-*.login.bosh-lite.com","backend":"10.123.0.134:8080","modification_tag":{"guid":"","index":0}}}

  • log_level: This represents logging level of the message
  • timestamp: Epoch time of the log
  • message: Content of the log line
  • source: The function within Gorouter that initiated the log message
  • data: Additional information that varies based on the message

Access logs provide information for the following fields when recieving a request:

<Request Host> - [<Start Date>] "<Request Method> <Request URL> <Request Protocol>" <Status Code> <Bytes Received> <Bytes Sent> "<Referer>" "<User-Agent>" <Remote Address> x_forwarded_for:"<X-Forwarded-For>" x_forwarded_proto:"<X-Forwarded-Proto>" vcap_request_id:<X-Vcap-Request-ID> response_time:<Response Time> app_id:<Application ID> <Extra Headers>

  • Status Code, Response Time, Application ID, and Extra Headers are all optional fields
  • The absence of Status Code, Response Time or Application ID will result in a "-" in the corresponding field

Access logs are also redirected to syslog.

Headers

If an user wants to send requests to a specific app instance, the header X-CF-APP-INSTANCE can be added to indicate the specific instance to be targeted. The format of the header value should be X-Cf-App-Instance: APP_GUID:APP_INDEX. If the instance cannot be found or the format is wrong, a 404 status code is returned. Usage of this header is only available for users on the Diego architecture.

Docs

There is a separate docs folder which contains more advanced topics.

Troubleshooting

Refer doc to learn more troubleshooting slow requests.

Contributing

Please read the contributors' guide