Introduction

A Linkerd config example

admin:
  port: 9990

routers:
- protocol: http
  label: int-http
  dtab: |
    /svc       => /#/io.l5d.fs;
  servers:
  - port: 4140
    ip: 0.0.0.0

- protocol: thrift
  servers:
  - port: 8081
    ip: 0.0.0.0
    thriftFramed: true
  client:
    thriftFramed: true
  thriftMethodInDst: false
  dtab: |
    /svc => /#/io.l5d.fs/thrift;

namers:
- kind: io.l5d.fs
  rootDir: disco

telemetry:
- kind: io.l5d.statsd
  experimental: true
  prefix: linkerd
  hostname: 127.0.0.1
  port: 8125
  gaugeIntervalMs: 10000
  sampleRate: 0.01
- kind: io.l5d.tracelog
  sampleRate: 0.2
  level: TRACE

Welcome to the Configuration Reference for linkerd!

linkerd’s configuration is controlled via config file, which must be provided as a command-line argument. It may be a local file path or - to indicate that the configuration should be read from the standard input. For convenience, the release package includes a default linkerd.yaml file in the config/ directory.

File Format

The configuration may be specified as a JSON or YAML object. There are no requirements on field ordering, though it’s generally good style to start a router with the protocol. Four top level keys are supported:

Key	Required	Description
admin	no	Configures Linkerd’s administrative interface.
routers	yes	Configures Linkerd’s RPC support for various protocols.
namers	no	Configures Linkerd’s integration with various service discovery backends.
telemetry	no	Configures Linkerd’s metrics instrumentation.
usage	no	Configures Linkerd usage reporting.

Administrative interface

admin:
  ip: 127.0.0.1
  port: 9990

Linkerd supports an administrative interface, both as a web ui and a collection of json endpoints. The exposed admin port and ip to listen on are configurable via a top-level admin section.

Key	Default Value	Description
ip	`0.0.0.0`	IP for the admin interface.
port	`9990`	Port for the admin interface.

Administrative endpoints

Example admin requests

curl :9990/admin
curl :9990/admin/tracing?enable=true
curl -H "Accept: application/json" :9990/admin/lint
curl -H "Accept: application/json" :9990/admin/threads
curl ":9990/admin/pprof/profile?seconds=10&hz=100"
curl ":9990/delegator.json?path=/http/1.1/GET/foo&dtab=/http/*=>/$/inet/127.1/9990"

Default admin endpoints available in both Linkerd and namerd:

Endpoint	Description
`/admin`	retrieve a list of all available admin endpoints
`/admin/announcer`	set of announcement chains that have run through the announcers
`/admin/contention`	call stacks of blocked and waiting threads
`/admin/lint`	results for all registered linters, set `Accept: application/json` to force json
`/admin/lint.json`	identical to `admin/lint`
`/admin/ping`	simple health check endpoint, returns `pong`
`/admin/pprof/contention`	CPU contention profile which identifies blocked threads (Thread.State.BLOCKED), in pprof format. The process will be profiled for 10 seconds at a frequency of 100 hz. These values can be controlled via HTTP request parameters `seconds` and `hz` respectively.
`/admin/pprof/heap`	heap profile computed by the heapster agent, output is in pprof format
`/admin/pprof/profile`	CPU usage profile in pprof format. The process will be profiled for 10 seconds at a frequency of 100 hz. These values can be controlled via HTTP request parameters `seconds` and `hz` respectively.
`/admin/registry.json`	displays how Linkerd is currently configured across a variety of dimensions including the client stack, server stack, flags, service loader values, system properties, environment variables, build properties and more
`/admin/server_info`	build information about this linkerd
`/admin/shutdown`	initiate a graceful shutdown
`/admin/threads`	capture the current stacktraces, set `Accept: application/json` to force json
`/admin/threads.json`	identical to `admin/threads`
`/admin/tracing`	enable (`/admin/tracing?enable=true`) or disable tracing (`/admin/tracing?disable=true`)
`/config.json`	current Linkerd configuration, this should match your config file

Endpoints only available in linkerd:

Endpoint	Description
`/bound-names.json`	list of all known bound names from configured namers. namers must support the `EnumeratingNamer` trait to make this available, currently only supported by the io.l5d.k8s and io.l5d.fs namers.
`/delegator.json`	given `path`, `dtab`, and an optional `namespace` param, return the delegation tree
`/logging.json`	currently configured loggers and log levels

Note that in addition to a default set of admin endpoints, Linkerd plugins may dynamically add their own endpoints.

This administrative interface was originally based on TwitterServer, more information may be found at TwitterServer HTTP Admin interface.

Routers Intro

A minimal Linkerd configuration example, which forwards all requests on localhost:8080 to localhost:8888

routers:
- protocol: http
  dtab: /svc/* => /$/inet/127.1/8888
  servers:
  - port: 8080

All configurations must define a routers key, the value of which must be an array of router configurations. Each router implements RPC for a supported protocol. Linkerd doesn’t need to understand the payload in an RPC call, but it does need to know enough about the protocol to determine the logical name of the destination.

See routers.

Namers and Service Discovery

namers:
- kind: io.l5d.fs
  rootDir: disco

Linkerd supports a variety of common service discovery backends, including ZooKeeper and Consul. Linkerd provides abstractions on top of service discovery lookups that allow the use of arbitrary numbers of service discovery backends, and for precedence and failover rules to be expressed between them. This logic is governed by the routing configuration.

Naming and service discovery are configured via the namers section of the configuration file. A namer acts on paths that start with /# followed by the namer’s prefix. See namers.

Telemetry Intro

telemetry:
- kind: io.l5d.statsd
  experimental: true
  prefix: linkerd
  hostname: 127.0.0.1
  port: 8125
  gaugeIntervalMs: 10000
  sampleRate: 0.01
- kind: io.l5d.tracelog
  sampleRate: 0.2
  level: TRACE

A telemeter may receive stats and trace annotations, i.e. to send to a collector or export. Telemetry data can be collected and exported from a Linkerd process by configuring telemeters via a top-level telemetry section. See telemetry.

Usage

Example usage config

usage:
 orgId: my-org

In order to make improvements and prioritize features, we’d like to gain a broader picture of how users run Linkerd. Linkerd gathers anonymized usage data, and sends it to Buoyant once an hour. This behavior can be configured or disabled in the usage config.

The kind of data captured is as follows:

How linkerds are configured (The kinds of namers, initializers, identifiers, transformers, protocols, & interpreters used)
What environments are linkerds running in (Operating System, Container orchestration solution),
How do linkerds perform in those contexts (JVM performance, Number of requests served)

We do not collect the labels of namers/routers, designated service addresses or directories, dtabs, request/response data, or any other possibly identifying information.

To see the exact payload that is sent query localhost:9990/admin/metrics/usage

Key	Default Value	Description
orgId	empty by default	Optional string of your choosing that identifies your organization
enabled	true	If set to true, data is sent to Buoyant once per hour

Routers

All configurations must define a routers key, the value of which must be an array of router configurations. Routers also include servers, which define their entry points, client, which configures how clients are built, and service, which configures service level policy.

Router Parameters

routers:
- protocol: http
  servers: ...
  service: ...
  client: ...
  announcers: ...
  bindingCache: ...
  label: myPackIce
  dstPrefix: /walruses/http
  dtab: |
    /host                => /#/io.l5d.fs;
    /walruses/http => /host;
  originator: true
  bindingTimeoutMs: 5000

Key	Default Value	Description
protocol	required	Either `http`, `h2`, `thrift`, or `mux`.
servers	required	A list of server objects.
announcers	an empty list	A list of service discovery announcers that servers can announce to.
dtab	an empty dtab	Sets the base delegation table. See dtabs for more.
bindingTimeoutMs	10 seconds	The maximum amount of time in milliseconds to spend binding a path.
bindingCache	see binding cache	Binding cache size configuration.
client	an empty object	A client configuration object.
dstPrefix	protocol dependent	A path prefix to be used on request destinations.
originator	`false`	If `true`, indicates that this router is the first hop for linker-to-linker requests, and reflects that in the router’s stats. Useful for deduping linker-to-linker stats.
interpreter	default interpreter	An interpreter object determining what module will be used to process destinations.
label	the value of protocol	The name of the router (in stats and the admin ui)

Binding Cache

- protocol: http
  servers:
  - port: 9000
  bindingCache:
    paths: 100
    trees: 100
    bounds: 100
    clients: 10

Key	Default Value	Description
paths	`1000`	Max number of paths in the path cache.
trees	`1000`	Max number of trees in the tree cache.
bounds	`1000`	Max number of bounds in the bounds cache.
clients	`1000`	Max number of clients in the clients cache.

Server Parameters

servers:
- port: 8080
  ip: 0.0.0.0
  tls:
    certPath: /foo/cert.pem
    keyPath: /foo/key.pem
  maxConcurrentRequests: 1000
  announce:
    - /#/io.l5d.serversets/discovery/prod/web

Key	Default Value	Description
port	protocol dependent	The TCP port number. Protocols may provide default values. If no default is provided, the port parameter is required.
ip	loopback address	The local IP address. A value like 0.0.0.0 configures the server to listen on all local IPv4 interfaces.
tls	no tls	The server will serve over TLS if this parameter is provided. see TLS.
maxConcurrentRequests	unlimited	The maximum number of concurrent requests the server will accept.
announce	an empty list	A list of concrete names to announce using the router’s announcers.
clearContext	`false`	If `true`, all headers that set Linkerd contexts are removed from inbound requests. Useful for servers exposed on untrusted networks.

Service Configuration

This section defines the policy that Linkerd will use when talking to services. The structure of this section depends on its kind.

Key	Default Value	Description
kind	`io.l5d.global`	Either io.l5d.global or io.l5d.static

Global Service Config

- protocol: http
  service:
    kind: io.l5d.global
    totalTimeoutMs: 500ms
    retries:
      budget:
        minRetriesPerSec: 5
        percentCanRetry: 0.5
        ttlSecs: 15
      backoff:
        kind: jittered
        minMs: 10
        maxMs: 10000

This service configuration allows you to specify service parameters which will be applied to all services.

Static Service Config

- protocol: http
  service:
    kind: io.l5d.static
    configs:
    - prefix: /svc
      retries:
        budget:
          minRetriesPerSec: 5
          percentCanRetry: 0.5
          ttlSecs: 15
        backoff:
          kind: jittered
          minMs: 10
          maxMs: 10000
    - prefix: /svc/foo
      totalTimeoutMs: 500ms
    - prefix: /svc/bar
      totalTimeoutMs: 200ms

This service configuration allows you to specify service parameters which will be applied to all services that match a specified prefix. The service configuration must contain a property called configs which contains a list of config objects. Each config object must specify a prefix and the service parameters to apply to services that match that prefix. If a service matches more than one prefix, all parameters from the matching configs will be applied, with parameters defined later in the configuration file taking precedence over those defined earlier.

Service Parameters

Key	Default Value	Description
retries	see retries	A retry policy for applicaiton-level retries.
totalTimeoutMs	no timeout	The timeout for an entire request, including all retries, in milliseconds.
responseClassifier	`io.l5d.http.nonRetryable5XX`	A (sometimes protocol-specific) response classifier that determines which responses should be considered failures and, of those, which should be considered retryable.

Client Configuration

This section defines how the clients that Linkerd creates will be configured. The structure of this section depends on its kind.

Key	Default Value	Description
kind	`io.l5d.global`	Either io.l5d.global or io.l5d.static

Global Client Config

- protocol: http
  client:
    kind: io.l5d.global
    loadBalancer:
      kind: ewma
    failureAccrual:
      kind: io.l5d.consecutiveFailures
      failures: 10

This client configuration allows you to specify client parameters which will be applied to all clients.

Static Client Config

- protocol: http
  client:
    kind: io.l5d.static
    configs:
    - prefix: /#/io.l5d.fs
      loadBalancer:
        kind: ewma
    - prefix: /#/io.l5d.fs/{service}
      tls:
        commonName: "{service}.linkerd.io"
    - prefix: /$/inet/*/80
      failureAccrual:
        kind: io.l5d.consecutiveFailures
        failures: 10

This client configuration allows you to specify client parameters which will be applied to all clients that match a specified prefix. The client configuration must contain a property called configs which contains a list of config objects. Each config object must specify a prefix and the client parameters to apply to clients that match that prefix. A prefix may contain wildcards (*) and capture variables ({foo}) which can be referenced in some client parameters. If a client matches more than one config’s prefix, all parameters from the matching configs will be applied, with parameters defined later in the configuration file taking precedence over those defined earlier.

Client Parameters

client:
  tls:
    kind: io.l5d.noValidation
    commonName: foo
    caCertPath: /foo/caCert.pem
  requestAttemptTimeoutMs: 100
  loadBalancer:
    kind: ewma
    enableProbation: false
  requeueBudget:
    percentCanRetry: 0.25
  failureAccrual:
    kind: io.l5d.consecutiveFailures
    failures: 10

Key	Default Value	Description
hostConnectionPool	An empty object	see hostConnectionPool.
tls	no tls	The router will make requests using TLS if this parameter is provided. It must be a client TLS object.
loadBalancer	p2c	A load balancer object.
failFast	`false`	If `true`, connection failures are punished more aggressively. Should not be used with small destination pools.
requeueBudget	see retry budget	A requeue budget for connection-level retries.
failureAccrual	5 consecutive failures	a failure accrual policy for all clients created by this router.
requestAttemptTimeoutMs	no timeout	The timeout, in milliseconds, for each attempt (original or retry) of the request made by this client.

Host Connection Pool

client:
  hostConnectionPool:
    minSize: 0
    maxSize: 1000
    idleTimeMs: 10000
    maxWaiters: 5000

Key	Default Value	Description
minSize	`0`	The minimum number of connections to maintain to each host.
maxSize	Int.MaxValue	The maximum number of connections to maintain to each host.
idleTimeMs	forever	The amount of idle time for which a connection is cached in milliseconds.
maxWaiters	Int.MaxValue	The maximum number of connection requests that are queued when the connection concurrency exceeds maxSize.

HTTP/1.1 protocol

Below: http-specific configuration options

routers:
- protocol: http
  httpAccessLog: access.log
  identifier:
    kind: io.l5d.methodAndHost
  maxChunkKB: 8
  maxHeadersKB: 8
  maxInitialLineKB: 4
  maxRequestKB: 5120
  maxResponseKB: 5120
  servers:
  - port: 5000
    addForwardedHeader:
      by: {kind: "ip:port"}
      for: {kind: ip}

protocol: http

The HTTP/1.1 protocol is used when the protocol option of the routers configuration block is set to http. This protocol has additional configuration options on the routers block.

Key	Default Value	Description
dstPrefix	`/svc`	A path prefix used by Http-specific identifiers.
httpAccessLog	none	Sets the access log path. If not specified, no access log is written.
identifier	The `io.l5d.header.token` identifier	An identifier or list of identifiers. See Http-specific identifiers.
maxChunkKB	8	The maximum size of an HTTP chunk.
maxHeadersKB	8	The maximum size of all headers in an HTTP message.
maxInitialLineKB	4	The maximum size of an initial HTTP message line.
maxRequestKB	5120	The maximum size of a non-chunked HTTP request payload.
maxResponseKB	5120	The maximum size of a non-chunked HTTP response payload.
compressionLevel	`-1`, automatically compresses textual content types with compression level 6	The compression level to use (on 0-9).

HTTP Servers

HTTP servers accept additional configuration parameters.

Example: default

addForwardedHeader: {}

Key	Default Value	Description
addForwardedHeader	null	If set, a `Forwarded` header is added to all requests. See below.

Adding the `Forwarded` header

RFC 7239 describes how a Forwarded header may be added to requests by proxies. This RFC requests that this header not be added unless explicitly configured and that proxies obfuscate IP addresses unless explicitly configured to transmit them.

Key	Default Value	Description
by	`{kind: requestRandom}`	The labeler to use with the router’s server address
for	`{kind: requestRandom}`	The labeler to use with the upstream client’s address

Endpoint labelers

The Forwarded header includes labels describing the endpoints of the upstream connection. Because this is sensitive information, it is typically randomized.

addForwardedHeader:
  for: {kind: ip}
  by:
    kind: static
    label: linkerd

Kind	Description
ip	A textual IP address like `192.168.1.1` or `"[2001:db8:cafe::17]"`.
ip:port	A textual IP:PORT address like `"192.168.1.1:80"` or `"[2001:db8:cafe::17]:80"`.
connectionRandom	An obfuscated random label like `_6Oq8jJ` generated for all requests on a connection.
requestRandom	An obfuscated random label like `_6Oq8jJ` generated for each request.
router	Uses the router’s `label` as an obfuscated static label.
static	Accepts a `label` parameter. Produces obfuscated static labels like `_linkerd`.

HTTP/1.1 Identifiers

Identifiers are responsible for creating logical names from an incoming request; these names are then matched against the dtab. (See the linkerd routing overview for more details on this.) All HTTP/1.1 identifiers have a kind. If a list of identifiers is provided, each identifier is tried in turn until one successfully assigns a logical name to the request.

If no identifier is specified the io.l5d.header.token identifier is used.

Key	Default Value	Description
kind	required	Either `io.l5d.methodAndHost`, `io.l5d.path`, `io.l5d.header`, `io.l5d.header.token`, or `io.l5d.static`.

Method and Host Identifier

kind: io.l5d.methodAndHost.

With this identifier, HTTP requests are turned into logical names using a combination of Host header, method, and (optionally) URI. Host header value is lower-cased as per RFC 2616.

Identifier Configuration:

Configuration example

identifier:
  kind: io.l5d.methodAndHost
  httpUriInDst: true

Key	Default Value	Description
httpUriInDst	`false`	If `true` http paths are appended to destinations. This allows a form of path-prefix routing. This option is not recommended as performance implications may be severe; Use the path identifier instead.

Identifier Path Parameters:

Dtab Path Format for HTTP/1.1

  / dstPrefix / "1.1" / method / host [/ uri* ]

Dtab Path Format for HTTP/1.0

  / dstPrefix / "1.0" / method [/ uri* ]

Key	Default Value	Description
dstPrefix	`/svc`	The `dstPrefix` as set in the routers block.
method	N/A	The HTTP method of the current request, ie `OPTIONS`, `GET`, `HEAD`, `POST`, `PUT`, `DELETE`, `TRACE`, or `CONNECT`.
host	N/A	The value of the current request’s Host header. Case sensitive!. Not used in HTTP/1.0.
uri	Not used	Only considered a part of the logical name if the config option `httpUriInDst` is `true`.

Path Identifier

kind: io.l5d.path

With this identifier, HTTP requests are turned into names based only on the path component of the URL, using a configurable number of “/” separated segments from the start of their HTTP path.

Identifier Configuration:

With this configuration, a request to :5000/true/love/waits.php will be mapped to /http/true/love and will be routed based on this name by the corresponding dtab. Additionally, because consume is true, after routing, requests will be proxied to the destination service with /waits.php as the path component of the URL.

routers:
- protocol: http
  identifier:
    kind: io.l5d.path
    segments: 2
    consume: true
  servers:
  - port: 5000

Key	Default Value	Description
segments	`1`	Number of segments from the path that are appended to destinations.
consume	`false`	Whether to additionally strip the consumed segments from the HTTP request proxied to the final destination service. This only affects the request sent to the destination service; it does not affect identification or routing.

Identifier Path Parameters:

Dtab Path Format

  / dstPrefix [/ *urlPath ]

Key	Default Value	Description
dstPrefix	`/svc`	The `dstPrefix` as set in the routers block.
urlPath	N/A	A path from the URL whose number of segments is set in the identifier block.

Header Identifier

kind: io.l5d.header

With this identifier, HTTP requests are turned into names based only on the value of an HTTP header. The value of the HTTP header is interpreted as a path and therefore must start with a /.

Identifier Configuration:

With this configuration, the value of the my-header HTTP header will be used as the logical name.

routers:
- protocol: http
  identifier:
    kind: io.l5d.header
    header: my-header
  servers:
  - port: 5000

Key	Default Value	Description
header	`l5d-name`	The name of the HTTP header to use

Identifier Path Parameters:

Dtab Path Format

  / dstPrefix [*headerValue ]

Key	Default Value	Description
dstPrefix	`/svc`	The `dstPrefix` as set in the routers block.
headerValue	N/A	The value of the HTTP header as a path.

Header Token Identifier

kind: io.l5d.header.token

With this identifier, HTTP requests are turned into names based only on the value of an HTTP header. The name is a path with one segment and the value of that segment is taken from the HTTP header.

Identifier Configuration:

With this configuration, the value of the my-header HTTP header will be used as the logical name.

routers:
- protocol: http
  identifier:
    kind: io.l5d.header.token
    header: my-header
  servers:
  - port: 5000

Key	Default Value	Description
header	`Host`	The name of the HTTP header to use

Identifier Path Parameters:

Dtab Path Format

  / dstPrefix / [headerValue]

Key	Default Value	Description
dstPrefix	`/svc`	The `dstPrefix` as set in the routers block.
headerValue	N/A	The value of the HTTP header as a path segment.

Ingress Identifier

kind: io.l5d.ingress

Using this identifier enables Linkerd to function as a Kubernetes ingress controller. The ingress identifier compares HTTP requests to ingress resource rules, and assigns a name based on those rules.

Identifier Configuration:

This example watches all ingress resources in the default namespace:

routers:
- protocol: http
  identifier:
    kind: io.l5d.ingress
    namespace: default
  servers:
  - port: 4140
  dtab: /svc => /#/io.l5d.k8s

namers:
- kind: io.l5d.k8s

An example ingress resource watched by the Linkerd ingress controller:

apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  name: my-first-ingress
  namespace: default
annotations:
  kubernetes.io/ingress.class: "linkerd"
spec:
  rules:
  - http:
      paths:
      - path: /testpath
        backend:
          serviceName: test
          servicePort: 80

So an HTTP request like http://localhost:4140/testpath would have an identified name of /svc/default/80/test

Key	Default Value	Description
namespace	(all)	The Kubernetes namespace where the ingress resources are deployed. If not specified, Linkerd will watch all namespaces.
host	`localhost`	The Kubernetes master host.
port	`8001`	The Kubernetes master port.

Identifier Path Parameters

Dtab Path Format

  / dstPrefix / namespace / port / service

Key	Default Value	Description
dstPrefix	`/svc`	The `dstPrefix` as set in the routers block.
namespace	N/A	The Kubernetes namespace.
port	N/A	The port name.
svc	N/A	The name of the service.

Static Identifier

kind: io.l5d.static

This identifier always assigns the same static name to all requests.

Identifier Configuration:

routers:
- protocol: http
  identifier:
    kind: io.l5d.static
    path: /foo/bar

Key	Default Value	Description
path	required	The name to assign to all requests

Identifier Path Parameters

Dtab Path Format

  / dstPrefix / *path

Key	Default Value	Description
dstPrefix	`/svc`	The `dstPrefix` as set in the routers block.
path	N/A	The path given in the configuration.

HTTP Engines

This configures an HTTP router that uses the new netty4 implementation on both the client and server:

- protocol: http
  servers:
  - port: 4141
    ip: 0.0.0.0
    engine:
      kind: netty4
  client:
    engine:
      kind: netty4

An engine may be configured on HTTP clients and servers, causing an alternate HTTP implementation to be used.

Key	Default Value	Description
kind	`netty3`	Either `netty3` or `netty4` (`netty4` will become default in an upcoming release).

HTTP Headers

Linkerd reads and sets several headers prefixed by l5d-.

Context Headers

Context headers (l5d-ctx-*) are generated and read by linkerd instances. Applications should forward all context headers in order for all Linkerd features to work.

Header	Description
`dtab-local`	Deprecated. Use `l5d-ctx-dtab` and `l5d-dtab`.
`l5d-ctx-deadline`	Describes time bounds within which a request is expected to be satisfied. Currently deadlines are only advisory and do not factor into request cancellation.
`l5d-ctx-trace`	Encodes Zipkin-style trace IDs and flags so that trace annotations emitted by Linkerd may be correlated.

User Headers

Append a dtab override to the dtab for this request

curl -H 'l5d-dtab: /host/web => /host/web-v2' "localhost:5000"

User headers enable user-overrides.

Header	Description
`l5d-dtab`	A client-specified delegation override.
`l5d-sample`	A client-specified trace sample rate override.

Informational Request Headers

The informational headers Linkerd emits on outgoing requests.

Header	Description
`l5d-dst-service`	The logical service name of the request as identified by linkerd.
`l5d-dst-client`	The concrete client name after delegation.
`l5d-dst-residual`	An optional residual path remaining after delegation.
`l5d-reqid`	A token that may be used to correlate requests in a callgraph across services and Linkerd instances.

Applications are not required to forward these headers on downstream requests.

Informational Response Headers

The informational headers Linkerd emits on outgoing responses.

Header	Description
`l5d-err`	Indicates a linkerd-generated error. Error responses that do not have this header are application errors.
`l5d-retryable`	Indicates that the request for this response is known to be safe to retry (for example, because it was not delivered to its destination).

Applications should not forward these headers on upstream responses.

HTTP/2 protocol

Below: Authority (Host) based routing for HTTP/2 over TLS

routers:
- protocol: h2
  experimental: true
  servers:
  - port: 4143
    tls:
      certPath: .../public/linkerd.pem
      keyPath: .../private/linkerd.pem
      caCertPath: .../ca.pem
  identifier:
    kind: io.l5d.header.token
    header: ":authority"
  dtab: |
    /svc => /#/io.l5d.fs ;
  client:
    tls:
      kind: io.l5d.boundPath
      caCertPath: .../ca.pem
      names:
      - prefix: "/#/io.l5d.fs/{service}"
        commonNamePattern: "{service}"

Below: plaintext gRPC

routers:
- protocol: h2
  experimental: true
  label: grpc
  servers:
  - port: 4142
  identifier:
    kind: io.l5d.header.path
    segments: 2
  dtab: |
    /svc => /#/io.l5d.fs ;

because gRPC encodes URLs as /serviceName/methodName, we can simply register service names into a discovery system and route accordingly. Note that gRPC may be configured over TLS as well.

protocol: h2

Linkerd now has experimental support for HTTP/2. There are a number of open issues that are being addressed. Please report any additional issues with this protocol!

Key	Default Value	Description
dstPrefix	`/svc`	A path prefix used by H2-specific identifiers.
experimental	`false`	Set this to `true` to opt-in to experimental h2 support.
identifier	The `io.l5d.header.token` identifier	An identifier or list of identifiers. See HTTP/2 Identifiers.

When TLS is configured, h2 routers negotiate to communicate over HTTP/2 via ALPN.

When TLS is not configured, h2 servers accept both prior knowledge and HTTP Upgrade requests. Plaintext clients are currently only capable of issuing prior-knowledge requests.

HTTP/2 Server Parameters

Key	Default Value	Description
windowUpdateRatio:	`0.99`	A number between 0 and 1, exclusive, indicating the ratio at which window updates should be sent. With a value of 0.75, updates will be sent when the available window size is 75% of its capacity.
headerTableBytes	none	Configures `SETTINGS_HEADER_TABLE_SIZE` on new streams.
initialStreamWindowBytes	64KB	Configures `SETTINGS_INITIAL_WINDOW_SIZE` on streams.
maxConcurrentStreamsPerConnection	unlimited	Configures `SETTINGS_MAX_CONCURRENT_STREAMS` on new streams.
maxFrameBytes	16KB	Configures `SETTINGS_MAX_FRAME_SIZE` on new streams.
maxHeaderListByts	none	Configures `SETTINGS_MAX_HEADER_LIST_SIZE` on new streams.

HTTP/2 Client Parameters

Key	Default Value	Description
windowUpdateRatio:	`0.99`	A number between 0 and 1, exclusive, indicating the ratio at which window updates should be sent. With a value of 0.75, updates will be sent when the available window size is 75% of its capacity.
headerTableBytes	none	Configures `SETTINGS_HEADER_TABLE_SIZE` on new streams.
initialStreamWindowBytes	64KB	Configures `SETTINGS_INITIAL_WINDOW_SIZE` on streams.
maxFrameBytes	16KB	Configures `SETTINGS_MAX_FRAME_SIZE` on new streams.
maxHeaderListByts	none	Configures `SETTINGS_MAX_HEADER_LIST_SIZE` on new streams.

HTTP/2 Identifiers

Identifiers are responsible for creating logical names from an incoming request; these names are then matched against the dtab. (See the linkerd routing overview for more details on this.) All h2 identifiers have a kind. If a list of identifiers is provided, each identifier is tried in turn until one successfully assigns a logical name to the request.

Key	Default Value	Description
kind	required	Either `io.l5d.header.token`, `io.l5d.header.path`, or `io.l5d.ingress`.

HTTP/2 Header Token identifier

kind: io.l5d.header.token.

With this identifier, requests are turned into logical names using the value of the named header. By default, the :authority pseudo-header is used to provide host-based routing.

Namer Configuration:

With this configuration, the value of the my-header header will be used as the logical name.

routers:
- protocol: h2
  experimental: true
  identifier:
    kind: io.l5d.header.token
    header: my-header
  servers:
  - port: 5000

Key	Default Value	Description
header	`:authority`	The name of the header to extract a token from. If there are multiple headers with this name, the last one is used.

Namer Path Parameters:

Dtab Path Format

  / dstPrefix / headerValue

Key	Default Value	Description
dstPrefix	`/svc`	The `dstPrefix` as set in the routers block.
headerValue	N/A	The value of the header.

HTTP/2 Header Path Identifier

kind: io.l5d.header.path

With this identifier, requests are identified using a path read from a header. This is useful for routing gRPC requests. By default, the :path psuedo-header is used.

Namer Configuration:

With this configuration, a request to :5000/true/love/waits.php?thing=1 will be mapped to /h2/true/love and will be routed based on this name by the corresponding Dtab.

routers:
- protocol: h2
  experimental: true
  identifier:
    kind: io.l5d.header.path
    segments: 2
  servers:
  - port: 5000

Key	Default Value	Description
header	`:path`	The name of the header to extract a Path from. If there are multiple headers with this name, the last one is used.
segments	None	If specified, the number of path segments that are required extracted from each request.

Namer Path Parameters:

Dtab Path Format

  / dstPrefix / *urlPath

Key	Default Value	Description
dstPrefix	`/svc`	The `dstPrefix` as set in the routers block.
urlPath	N/A	The first `segments` elements of the path from the URL

HTTP/2 Ingress Identifier

kind: io.l5d.ingress

Using this identifier enables Linkerd to function as a Kubernetes ingress controller. The ingress identifier compares HTTP/2 requests to ingress resource rules, and assigns a name based on those rules.

Identifier Configuration:

This example watches all ingress resources in the default namespace:

routers:
- protocol: h2
  experimental: true
  identifier:
    kind: io.l5d.ingress
    namespace: default
  servers:
  - port: 4140
  dtab: /svc => /#/io.l5d.k8s

namers:
- kind: io.l5d.k8s

An example ingress resource watched by the Linkerd ingress controller:

apiVersion: extensions/v1beta1
kind: Ingress
metadata:
  name: my-first-ingress
  namespace: default
annotations:
  kubernetes.io/ingress.class: "linkerd"
spec:
  rules:
  - http:
      paths:
      - path: /testpath
        backend:
          serviceName: test
          servicePort: 80

So an HTTP/2 request like https://localhost:4140/testpath would have an identified name of /svc/default/80/test

Key	Default Value	Description
namespace	(all)	The Kubernetes namespace where the ingress resources are deployed. If not specified, Linkerd will watch all namespaces.
host	`localhost`	The Kubernetes master host.
port	`8001`	The Kubernetes master port.

Identifier Path Parameters

Dtab Path Format

  / dstPrefix / namespace / port / service

Key	Default Value	Description
dstPrefix	`/svc`	The `dstPrefix` as set in the routers block.
namespace	N/A	The Kubernetes namespace.
port	N/A	The port name.
svc	N/A	The name of the service.

HTTP/2 Headers

Linkerd reads and sets several headers prefixed by l5d-, as is done by the http protocol.

HTTP/2 Context Headers

Context headers (l5d-ctx-*) are generated and read by linkerd instances. Applications should forward all context headers in order for all Linkerd features to work.

Header	Description
`dtab-local`	Deprecated. Use `l5d-ctx-dtab` and `l5d-dtab`.
`l5d-ctx-deadline`	Describes time bounds within which a request is expected to be satisfied. Currently deadlines are only advisory and do not factor into request cancellation.
`l5d-ctx-trace`	Encodes Zipkin-style trace IDs and flags so that trace annotations emitted by Linkerd may be correlated.

HTTP/2 User Headers

Append a dtab override to the dtab for this request

curl -H 'l5d-dtab: /host/web => /host/web-v2' "localhost:5000"

User headers enable user-overrides.

Header	Description
`l5d-dtab`	A client-specified delegation override.
`l5d-sample`	A client-specified trace sample rate override.

HTTP/2 Informational Request Headers

The informational headers Linkerd emits on outgoing requests.

Header	Description
`l5d-dst-service`	The logical service name of the request as identified by linkerd.
`l5d-dst-client`	The concrete client name after delegation.
`l5d-dst-residual`	An optional residual path remaining after delegation.
`l5d-reqid`	A token that may be used to correlate requests in a callgraph across services and Linkerd instances.

Applications are not required to forward these headers on downstream requests.

HTTP/2 Informational Response Headers

The informational headers Linkerd emits on outgoing responses.

Header	Description
`l5d-err`	Indicates a linkerd-generated error. Error responses that do not have this header are application errors.

Applications are not required to forward these headers on upstream responses.

Thrift Protocol

This config routes thrift (via buffered transport using the TCompactProtocol) from port 4004 to port 5005

routers:
- protocol: thrift
  label: port-shifter
  dtab: |
    /svc => /$/inet/127.1/5005;
  servers:
  - port: 4004
    ip: 0.0.0.0
    thriftFramed: false
    thriftProtocol: compact
  client:
    thriftFramed: false
    thriftProtocol: compact

protocol: thrift

If the TTwitter thrift protocol is used, the value from the dest request header will be used for routing:

Dtab Path Format For Thrift / dstPrefix [/ dest] [/ thriftMethod ]

Otherwise, the Thrift protocol does not encode a destination name in the message itself and the dest part of the path will be absent.

Thrift Router Parameters

Key	Default Value	Description
dstPrefix	`/svc`	A path prefix used in `dtab`.
thriftMethodInDst	`false`	If `true`, thrift method names are appended to destinations for outgoing requests.

Thrift Server Parameters

Key	Default Value	Description
port	`4114`	The TCP port number.
thriftFramed	`true`	If `true`, a framed thrift transport is used for incoming requests; otherwise, a buffered transport is used. Typically this setting matches the router’s `thriftFramed` param.
thriftProtocol	`binary`	Either `binary` (TBinaryProtocol) or `compact` (TCompantProtocol). Typically this setting matches the router’s client `thriftProtocol` param.

Thrift Client Parameters

Key	Default Value	Description
thriftFramed	`true`	If `true`, a framed thrift transport is used for outgoing requests; otherwise, a buffered transport is used. Typically this setting matches the router’s servers’ `thriftFramed` param.
thriftProtocol	`binary`	Either `binary` (TBinaryProtocol) or `compact` (TCompantProtocol). Typically this setting matches the router’s servers’ `thriftProtocol` param.
attemptTTwitterUpgrade	`false`	Controls whether thrift protocol upgrade should be attempted.

ThriftMux Protocol (experimental)

This config routes thriftmux from port 4400 to port 5005.

routers:
- protocol: thriftmux
  experimental: true
  label: port-shifter
  dtab: |
    /svc => /$/inet/127.1/5005;
  servers:
  - port: 4400
    ip: 0.0.0.0

protocol: thriftmux

Linkerd experimentally supports the thriftmux

Thriftmux protocol is capable of routing traffic to pure thrift service and will use Thrift protocol on the client.

Protocol configuration uses the same parameters as Thrift protocol.

ThriftMux Router Parameters

See Thrift Router Parameters

ThriftMux Server Parameters

See Thrift Server Parameters

ThriftMux Client Parameters

See Thrift Client Parameters

Mux Protocol (experimental)

A mux router configuration that routes requests to port 9001


routers:
- protocol: mux
  label: power-level-router
  dstPrefix: /overNineThousand
  dtab: |
    /overNineThousand => /$/inet/127.0.1/9001;

protocol: mux

Linkerd experimentally supports the mux protocol.

Mux Router Parameters

Key	Default Value	Description
dstPrefix	`/svc`	A path prefix used in `dtab`.

Mux Server Parameters

Key	Default Value	Description
port	`4141`	The TCP port number.

Interpreter

Example Interpreter Configuration

routers:
- ...
  interpreter:
    kind: io.l5d.namerd
    dst: /$/inet/1.2.3.4/4180

An interpreter determines how names are resolved.

Key	Default Value	Description
kind	`default`	Either `default`, `io.l5d.namerd`, `io.l5d.namerd.http`, `io.l5d.mesh`, or `io.l5d.fs`.
transformers	No transformers	A list of transformers to apply to the resolved addresses.

Default

kind: default

The default interpreter resolves names via the configured namers, with a fallback to the default Finagle Namer.Global that handles paths of the form /$/.

namerd thrift

kind: io.l5d.namerd

The namerd thrift interpreter offloads the responsibilities of name resolution to the namerd service. Any namers configured in this Linkerd are not used. The interpreter users namerd’s long-poll thrift interface (io.l5d.thriftNameInterpreter). Note that the protocol that the interpreter uses to talk to namerd is unrelated to the protocols of Linkerd’s routers.

Key	Default Value	Description
dst	required	A Finagle path locating the namerd service.
namespace	`default`	The name of the namerd dtab to use.
retry	see namerd retry	An object configuring retry backoffs for requests to namerd.
tls	no tls	Requests to namerd will be made using TLS if this parameter is provided. It must be a namerd client TLS object.

namerd retry

Key	Default Value	Description
baseSeconds	5 seconds	The base number of seconds to wait before retrying.
maxSeconds	10 minutes	The maximum number of seconds to wait before retrying.

namerd client tls

Key	Default Value	Description
commonName	required	The common name to use for namerd requests.
caCert	N/A	The path to the CA cert used for common name validation.

namerd http

kind: io.l5d.namerd.http

The namerd http interpreter offloads the responsibilities of name resolution to the namerd service. Any namers configured in this Linkerd are not used. The interpreter uses namerd’s HTTP streaming interface (io.l5d.httpController). Note that the protocol that the interpreter uses to talk to namerd is unrelated to the protocols of Linkerd’s routers.

Key	Default Value	Description
experimental	required	Because the http interpreter is still considered experimental, you must set this to `true` to use it.
dst	required	A Finagle path locating the namerd service.
namespace	`default`	The name of the namerd dtab to use.
retry	see namerd retry	An object configuring retry backoffs for requests to namerd.
tls	no tls	Requests to namerd will be made using TLS if this parameter is provided. It must be a namerd client TLS object.

namerd mesh

kind: io.l5d.mesh

The namerd mesh interpreter offloads the responsibilities of name resolution to the namerd service. Any namers configured in this Linkerd are not used. The interpreter uses namerd’s gRPC mesh interface (io.l5d.mesh). Note that the protocol that the interpreter uses to talk to namerd is unrelated to the protocols of Linkerd’s routers.

Key	Default Value	Description
experimental	required	Because the mesh interpreter is still considered experimental, you must set this to `true` to use it.
dst	required	A Finagle path locating the namerd service.
root	`/default`	A single-element Finagle path representing the namerd namespace.
retry	see namerd retry	An object configuring retry backoffs for requests to namerd.
tls	no tls	The namerd mesh interface does not support server TLS configuration at this time.

File-System

kind: io.l5d.fs

The file-system interpreter resolves names via the configured namers, just like the default interpreter, but also uses a dtab read from a file on the local file-system. The specified file is watched for changes so that the dtab may be edited live.

Key	Default Value	Description
dtabFile	required	The file-system path to a file containing a dtab.

Transformer

Example Transformer Configuration

routers:
- ...
  interpreter:
    ...
    transformers:
    - kind: io.l5d.localhost

Transformers perform a transformation on the addresses resolved by the interpreter. Transformations are applied sequentially in the order they appear.

Key	Default Value	Description
kind	required	Either `io.l5d.localhost`, `io.l5d.port`, `io.l5d.k8s.daemonset`, `io.l5d.k8s.localnode`, `io.l5d.replace`, or `io.l5d.const`.

Localhost

kind: io.l5d.localhost

The localhost transformer filters the list of addresses down to only addresses that have the same IP address as localhost. The IP of localhost is determined by doing a one-time dns lookup of the local hostname. This transformer can be used by an incoming router to only route traffic to local destinations.

Port

kind: io.l5d.port

The port transformer replaces the port number in every addresses with a configured value. This can be used if there is an incoming Linkerd router (or other reverse-proxy) running on a fixed port on each host and you with to send traffic to that port instead of directly to the destination address.

Key	Default Value	Description
port	required	The port number to use.

DaemonSet (Kubernetes)

kind: io.l5d.k8s.daemonset

The DaemonSetTransformer maps each address in the destination NameTree to a member of a given daemonset that is on the same /24 subnet. Since each k8s node is its own /24 subnet, the result is that each destination address is mapped to the member of the daemonset that is running on the same node. This can be used to redirect traffic to a reverse-proxy that runs as a daemonset.

This transformer assumes that there is a Kubernetes service for the daemonset which can be used to find all pods in the daemonset.

Key	Default Value	Description
k8sHost	`localhost`	The Kubernetes master host.
k8sPort	`8001`	The Kubernetes master port.
namespace	required	The Kubernetes namespace of the daemonset.
service	required	The Kubernetes service name for the daemonset.
port	required	The name of the daemonset port to use.
hostNetwork	`false`	If true, use nodeName instead of /24 subnet to determine which daemonset pod is on the destination node. Set this to true if the daemonset is running with `hostNetwork: true`.

Localnode (Kubernetes)

kind: io.l5d.k8s.localnode

The localnode transformer filters the list of addresses down to only addresses that are on the same /24 subnet as localhost. Since each k8s node is its own /24 subnet, the result is that only addresses on the local node are used.

This transformer does not have any configuration properties but it does require the POD_IP environment variable be set with the localhost IP address. This is most easily done with the Kubernetes downward API.

In your container spec:

env:
- name: POD_IP
    valueFrom:
      fieldRef:
        fieldPath: status.podIP

Key	Default Value	Description
hostNetwork	`false`	If true, use nodeName to determine which pods are on the local node. This adds the requirement that the `NODE_NAME` environment variable be set with the node name. Set this to true if the pod is running with `hostNetwork: true`.

Replace

kind: io.l5d.replace

The replace transformer replaces all bound names with a configurable path. This differs from the Const transformer in that if the original NameTree is Neg then the result will be Neg as well. This is useful if you want to use a namer to check the validity of a name but then actually route to a different name.

Key	Default Value	Description
path	required	Bound names will be replaced with this path.

Const

kind: io.l5d.const

The const transformer ignores the input and always returns a constant configurable path. This differs from the Replace transformer in that it always returns the configured path, even when the original NameTree is Neg.

Key	Default Value	Description
path	required	Ignore the input and return this path.

HTTP Response Classifiers

Example response classifier config

routers:
- protocol: http
  service:
    responseClassifier:
      kind: io.l5d.http.retryableRead5XX

Response classifiers determine which HTTP responses are considered to be failures (for the purposes of success rate calculation) and which of these responses may be retried.

Key	Default Value	Description
kind	`io.l5d.http.nonRetryable5XX`	Either `io.l5d.http.nonRetryable5XX`, `io.l5d.h2.nonRetryable5XX`, `io.l5d.http.retryableRead5XX`, `io.l5d.h2.retryableRead5XX`, `io.l5d.http.retryableIdempotent5XX`, or `io.l5d.h2.retryableIdempotent5XX`.

Non-Retryable 5XX

kind: io.l5d.http.nonRetryable5XX kind: io.l5d.h2.nonRetryable5XX

All 5XX responses are considered to be failures and none of these requests are considered to be retryable.

Retryable Read 5XX

kind: io.l5d.http.retryableRead5XX kind: io.l5d.h2.retryableRead5XX

All 5XX responses are considered to be failures. However, GET, HEAD, OPTIONS, and TRACE requests may be retried automatically.

Retryable Idempotent 5XX

kind: io.l5d.http.retryableIdempotent5XX kind: io.l5d.h2.retryableIdempotent5XX

Like io.l5d.http.retryableRead5XX/io.l5d.h2.retryableRead5XX, but PUT and DELETE requests may also be retried.

Failure Accrual

routers:
- ...
  client:
    failureAccrual:
      kind: io.l5d.successRate
      successRate: 0.9
      requests: 1000
      backoff:
        kind: jittered
        minMs: 5000
        maxMs: 300000

Linkerd uses failure accrual to track the number of requests that have failed to a given node, and it will backoff sending requets to any nodes whose failures have exceeded a given threshold. Both the failure threshold and the backoff behavior are configurable. By default, if Linkerd observes 5 consecutive failures from a node, it will mark the node as dead and only attempt to resend it traffic in increasing intervals between 5 seconds and 5 minutes.

Key	Default Value	Description
kind	required	Either `io.l5d.consecutiveFailures`, `io.l5d.successRate`, `io.l5d.successRateWindowed`, or `none`.
backoff	jittered backoff from 5s to 300s	A backoff policy that determines how long to wait before resending traffic.

Consecutive Failures

kind: io.l5d.consecutiveFailures

Observes the number of consecutive failures to each node, and backs off sending requests to nodes that have exceeded the specified number of failures.

Key	Default Value	Description
failures	required	Number of consecutive failures.

Success Rate

kind: io.l5d.successRate

Computes an exponentially-weighted moving average success rate for each node, and backs off sending requests to nodes that have fallen below the specified success rate. The window size for computing success rate is constrained to a fixed number of requests.

Key	Default Value	Description
successRate	required	Target success rate.
requests	required	Number of requests over which success rate is computed.

Success Rate (windowed)

kind: io.l5d.successRateWindowed

Key	Default Value	Description
successRate	required	Target success rate.
window	required	Number of seconds over which success rate is computed.

None

kind: none

Disables failure accrual altogether. This policy does not accept any additional parameters.

TLS

Server TLS

routers:
- protocol: http
  servers:
  - port: 4140
    # accept incoming TLS traffic from remote linkerd
    tls:
      certPath: /certificates/certificate.pem
      keyPath: /certificates/key.pem
  dtab: |
    /http => /$/inet/127.1/8080;

In order to accept incoming tls traffic, the tls parameter must be defined on the server.

Key	Default Value	Description
certPath	required	File path to the TLS certificate file.
keyPath	required	File path to the TLS key file.

See Transparent TLS with linkerd for more on how to generate certificate and key files.

Client TLS

Client TLS is defined in the client section of routers:

routers:
- protocol: http
  client:
    tls:
      commonName: linkerd.io
      trustCerts:
      - /certificates/cacert.pem

In order to send outgoing tls traffic, the tls parameter must be defined as a client parameter.

A client TLS object describes how Linkerd should use TLS when sending requests to destination services.

Key	Default Value	Description
disableValidation	false	Enable this to skip hostname validation (unsafe).
commonName	required unless disableValidation is set	The common name to use for all TLS requests.
trustCerts	empty list	A list of file paths of CA certs to use for common name validation.

Any variables captured from the client prefix may be used in the common name.

routers:
- protocol: http
  client:
    kind: io.l5d.static
    configs:
    - prefix: /#/io.l5d.fs/{service}
      tls:
        commonName: "{service}.linkerd.io"
        trustCerts:
        - /certificates/cacert.pem

Load Balancer

Example load balancer configuration

routers:
- ...
  client:
    loadBalancer:
      kind: ewma
      maxEffort: 10
      decayTimeMs: 15000

Key	Default Value	Description
kind	`p2c`	Either `p2c`, `ewma`, `aperture`, `heap`, or `roundRobin`.
enableProbation	`false`	If `true`, endpoints are eagerly evicted from service discovery. See Finagle’s LoadBalancerFactory.EnableProbation.

Power of Two Choices: Least Loaded

kind: p2c

Key	Default Value	Description
maxEffort	`5`	The number of times a load balancer can retry if the previously picked node was marked unavailable.

Power of Two Choices: Peak EWMA

kind: ewma

Key	Default Value	Description
maxEffort	`5`	The number of times a load balancer can retry if the previously picked node was marked unavailable.
decayTimeMs	10 seconds	The window of latency observations.

Aperture: Least Loaded

kind: aperture

Key	Default Value	Description
maxEffort	`5`	The number of times a load balancer can retry if the previously picked node was marked unavailable.
smoothWin	5 seconds	The window of concurrent load observation.
lowLoad	`0.5`	The lower bound of the load band used to adjust an aperture.
highLoad	`2`	The upper bound of the load band used to adjust an aperture.
minAperture	`1`	The minimum size of the aperture.

Heap: Least Loaded

kind: heap

Round Robin

kind: roundRobin

Key	Default Value	Description
maxEffort	`5`	The number of times a load balancer can retry if the previously picked node was marked unavailable.

Retries

routers:
- ...
  service:
    retries:
      budget:
        minRetriesPerSec: 5
        percentCanRetry: 0.5
        ttlSecs: 15
      backoff:
        kind: jittered
        minMs: 10
        maxMs: 10000

Linkerd can automatically retry requests on certain failures and can be configured via the retries block. Retries fall into two categories: retries and requeues.

Key	Default Value	Description
budget	See retry budget	Object that determins how many failed requests are eligible to be retried.
backoff	See retry backoff	Object that determines which backoff algorithm should be used.

Retries

Retries are for application-level failures (such as 5XX responses in the case of HTTP) as determined by the response classifier. If the response classifier determines that a request is a retryable failure, and the retry budget is not empty, then the request will be retried. Retries are configured by the retries parameter on the service object. On the retries object you may specify the retry budget and retry backoff schedule. Each service has its own retry budget that is not shared with other services or clients.

Requeues

Requeues are for connection-level failures that are guaranteed to be idempotent. If a connection-level failure is encountered and there is requeue budget available, then the request will be retried. Requeue budgets are configured by the requeueBudget parameter on the client object. Requeues happen immediately with no backoff. Each client has its own requeue budget that is not shared with other clients or services.

Retry Budget Parameters

For every 10 non-retry calls, allow 1 retry

service:
  retries:
    budget:
      percentCanRetry: 0.1

For every non-retry call, allow 2 retries

service:
  retries:
    budget:
      percentCanRetry: 2.0

Key	Default Value	Description
minRetriesPerSec	`10`	The minimum rate of retries allowed in order to accommodate clients that have just started issuing requests, as well as clients that do not issue many requests per window. Must be non-negative. If `0`, no reserve is given.
percentCanRetry	`0.2`	The percentage of calls that can be retried. This is in addition to any retries allowed via `minRetriesPerSec`. Must be >= `0` and <= `1000`.
ttlSecs	`10`	The amount of time in seconds that successful calls are considered when calculating retry budgets.

Retry Backoff Parameters

Key	Default Value	Description
kind	required	Either `constant` or `jittered`.

Constant Backoff

kind: constant

Key	Default Value	Description
ms	`0`	The number of milliseconds to wait before each retry.

Jittered Backoff

kind: jittered

Uses a decorrelated jitter backoff algorithm.

Key	Default Value	Description
minMs	required	The minimum number of milliseconds to wait before each retry.
maxMs	required	The maximum number of milliseconds to wait before each retry.

Namers

namers:
- kind: io.l5d.fs
  prefix: /disco
  rootDir: disco

A namer binds a concrete name to a physical address.

Key	Default Value	Description
kind	required	Either `io.l5d.fs`, `io.l5d.serversets`, `io.l5d.consul`, `io.l5d.k8s`, `io.l5d.marathon`, `io.l5d.zkLeader`, `io.l5d.curator`, or `io.l5d.rewrite`.
prefix	namer dependent	Resolves names with `/#/<prefix>`.
experimental	`false`	Set this to `true` to enable the namer if it is experimental.
transformers	No transformers	A list of transformers to apply to the resolved addresses.

File-based service discovery

kind: io.l5d.fs

File-based Configuration

Example fs configuration:

namers:
- kind: io.l5d.fs
  rootDir: disco

Then reference the namer in the dtab to use it:

dtab: |
  /svc => /#/io.l5d.fs

With the filesystem directory:

$ ls disco/
apps    users   web

The contents of the files look similar to this:

$ cat config/web
192.0.2.220 8080
192.0.2.105 8080
192.0.2.210 8080 * 2.0

Linkerd ships with a simple file-based service discovery mechanism, called the file-based namer. This system is intended to act as a structured form of basic host lists.

While simple, the file-based namer is a full-fledged service discovery system, and can be useful in production systems where host configurations are largely static. It can act as an upgrade path for the introduction of an external service discovery system, since application code will be isolated from these changes. Finally, when chained with precedence rules, the file-based namer can be a convenient way to add local service discovery overrides for debugging or experimentation.

This service discovery mechanism is tied to the directory set by the namers/rootDir key in config.yaml. This directory must be on the local filesystem and relative to Linkerd’s start path. Every file in this directory corresponds to a service, where the name of the file is the service’s concrete name, and the contents of the file must be a newline-delimited set of addresses.

Linkerd watches all files in this directory, so files can be added, removed, or updated, and Linkerd will pick up the changes automatically.

Key	Default Value	Description
prefix	`io.l5d.fs`	Resolves names with `/#/<prefix>`.
rootDir	required	the directory containing name files as described above.

File-based Path Parameters

Dtab Path Format:

/#/<prefix>/<fileName>

Key	Required	Description
prefix	yes	Tells Linkerd to resolve the request path using the fs namer.
fileName	yes	The file in `rootDir` to use when resolving this request.

ZooKeeper ServerSets service discovery

kind: io.l5d.serversets

ServerSets Configuration

Example ServerSets configuration:

namers:
- kind: io.l5d.serversets
  zkAddrs:
  - host: 127.0.0.1
    port: 2181

Then reference the namer in the dtab to use it:

dtab: |
  /svc => /#/io.l5d.serversets/discovery/prod;

Linkerd provides support for ZooKeeper ServerSets.

Key	Default Value	Description
prefix	`io.l5d.serversets`	Resolves names with `/#/<prefix>`.
zkAddrs	required	A list of ZooKeeper addresses, each of which have `host` and `port` parameters.

ServerSets Path Parameters

Dtab Path Format:

/#/<prefix>/<zkHosts>/<zkPath>[:<endpoint>]

Key	Required	Description
prefix	yes	Tells Linkerd to resolve the request path using the serversets namer.
zkHosts	yes	The ZooKeeper host to use for this request.
zkPath	yes	The ZooKeeper path to use for this request.
endpoint	no	The ZooKeeper endpoint to use for this request.

Consul service discovery

kind: io.l5d.consul

Consul Configuration

Configure a consul namer:

namers:
- kind: io.l5d.consul
  host: 127.0.0.1
  port: 2181
  includeTag: true
  useHealthCheck: true
  setHost: true
  consistencyMode: stale

Then reference the namer in the dtab to use it:

dtab: |
  /svc => /#/io.l5d.consul/dc1/prod;

linker provides support for service discovery via Consul.

Key	Default Value	Description
prefix	`io.l5d.consul`	Resolves names with `/#/<prefix>`.
host	`localhost`	The Consul host.
port	`8500`	The Consul port.
includeTag	`false`	If `true`, read a Consul tag from the path.
useHealthCheck	`false`	If `true`, exclude app instances that are failing Consul health checks. Even if `false`, linkerd’s built-in resiliency algorithms will still apply.
token	no authentication	The auth token to use when making API calls.
setHost	`false`	If `true`, HTTP requests resolved by Consul will have their Host header overwritten to `${serviceName}.service.${datacenter}.${domain}`. `$domain` is fetched from Consul.
consistencyMode	`default`	Select between Consul API consistency modes such as `default`, `stale` and `consistent`.
failFast	`false`	If `false`, disable fail fast and failure accrual for Consul client. Keep it `false` when using a local agent but change it to `true` when talking directly to an HA Consul API.
preferServiceAddress	`true`	If `true` use the service address if defined and default to the node address. If `false` always use the node address.

Consul Path Parameters

Dtab Path Format when includeTag is false

/#/<prefix>/<datacenter>/<serviceName>

Dtab Path Format when includeTag is true

/#/<prefix>/<datacenter>/<tag>/<serviceName>

Key	Required	Description
prefix	yes	Tells Linkerd to resolve the request path using the consul namer.
datacenter	yes	The Consul datacenter to use for this request. It can have a value `.local` (otherwise invalid datacenter name from Consul’s perspective) in order to reference a datacenter of the agent namer is connected to.
tag	yes if includeTag is `true`	The Consul tag to use for this request.
serviceName	yes	The Consul service name to use for this request.

Kubernetes service discovery

kind : io.l5d.k8s

K8s Configuration

Configure a K8s namer

namers:
- kind: io.l5d.k8s
  host: localhost
  port: 8001
  labelSelector: version

Then reference the namer in the dtab to use it:

dtab: |
  /svc => /#/io.l5d.k8s/prod/http;

Linkerd provides support for service discovery via Kubernetes.

Key	Default Value	Description
prefix	`io.l5d.k8s`	Resolves names with `/#/<prefix>`.
host	`localhost`	The Kubernetes master host.
port	`8001`	The Kubernetes master port.
labelSelector	none	The key of the label to filter services.

K8s Path Parameters

Dtab Path Format

/#/<prefix>/<namespace>/<port-name>/<svc-name>[/<label-value>]

Key	Required	Description
prefix	yes	Tells Linkerd to resolve the request path using the k8s namer.
namespace	yes	The Kubernetes namespace.
port-name	yes	The port name or port number.
svc-name	yes	The name of the service.
label-value	yes if `labelSelector` is defined	The value used to filter services.

K8s External Configuration

Configure a K8s External namer

namers:
- kind: io.l5d.k8s.external
  experimental: true
  host: localhost
  port: 8001
  labelSelector: version

Then reference the namer in the dtab to use it:

dtab: |
  /svc => /#/io.l5d.k8s.external/prod/http;

The Kubernetes External namer looks up the IP of the external load balancer for the given service on the given port. This can be used by Linkerd instances running outside of k8s to route to services running in k8s.

Key	Default Value	Description
prefix	`io.l5d.k8s.external`	Resolves names with `/#/<prefix>`.
experimental	required	Because this namer is still considered experimental, you must set this to `true` to use it.
host	`localhost`	The Kubernetes master host.
port	`8001`	The Kubernetes master port.
labelSelector	none	The key of the label to filter services.

K8s External Path Parameters

Dtab Path Format

/#/<prefix>/<namespace>/<port-name>/<svc-name>[/<label-value>]

Key	Required	Description
prefix	yes	Tells Linkerd to resolve the request path using the k8s external namer.
namespace	yes	The Kubernetes namespace.
port-name	yes	The port name.
svc-name	yes	The name of the service.
label-value	yes if `labelSelector` is defined	The label value used to filter services.

Marathon service discovery

kind: io.l5d.marathon

Marathon Configuration

Configure a marathon namer

namers:
- kind:           io.l5d.marathon
  prefix:         /io.l5d.marathon
  host:           marathon.mesos
  port:           80
  uriPrefix:      /marathon
  ttlMs:          5000
  useHealthCheck: false

Then reference the namer in the dtab to use it:

dtab: |
  /marathonId => /#/io.l5d.marathon;
  /host       => /$/io.buoyant.http.domainToPathPfx/marathonId;
  /svc => /host;

Linkerd provides support for service discovery via Marathon.

Key	Default Value	Description
prefix	`io.l5d.marathon`	Resolves names with `/#/<prefix>`.
host	`marathon.mesos`	The Marathon master host.
port	`80`	The Marathon master port.
uriPrefix	none	The Marathon API prefix. This prefix depends on your Marathon configuration. For example, running Marathon locally, the API is available at `localhost:8080/v2/`, while the default setup on AWS/DCOS is `$(dcos config show core.dcos_url)/marathon/v2/apps`.
ttlMs	`5000`	The polling interval in milliseconds against the Marathon API.
useHealthCheck	`false`	If `true`, exclude app instances that are failing Marathon health checks. Even if `false`, linkerd’s built-in resiliency algorithms will still apply.

Marathon Path Parameters

Dtab Path Format

/#/<prefix>/<appId>

Key	Required	Description
prefix	yes	Tells Linkerd to resolve the request path using the marathon namer.
appId	yes	The app id of a marathon application. This id can be multiple path segments long. For example, the app with id “/users” can be reached with `/#/io.l5d.marathon/users`. Likewise, the app with id “/appgroup/usergroup/users” can be reached with `/#/io.l5d.marathon/appgroup/usergroup/users`.

Marathon Authentication

Example DCOS environment variable

{
  "login_endpoint": "https://leader.mesos/acs/api/v1/auth/login",
  "private_key": "<private-key-value>",
  "scheme": "RS256",
  "uid": "service-acct"
}

Example basic HTTP authentication variable

dXNlcm5hbWU6cGFzc3dvcmQ=

The Marathon namer supports loading authentication data from an environment variable for DCOS private key in the DCOS_SERVICE_ACCOUNT_CREDENTIAL variable and standalone Marathon basic HTTP authentication in the MARATHON_HTTP_AUTH_CREDENTIAL environment variable. If both are provided the DCOS_SERVICE_ACCOUNT_CREDENTIAL takes precedence.

Basic authentication token is base64 encoded and should not include the Basic prefix, only in the format username:password.

ZooKeeper Leader

kind: io.l5d.zkLeader

ZK Leader Configuration

A namer backed by ZooKeeper leader election.

Key	Default Value	Description
prefix	`io.l5d.zkLeader`	Resolves names with `/#/<prefix>`.
zkAddrs	required	A list of ZooKeeper addresses, each of which have `host` and `port` parameters.

ZK Leader Path Parameters

Dtab Path Format

/#/<prefix>/<zkPath>

Key	Required	Description
prefix	yes	Tells Linkerd to resolve the request path using the marathon namer.
zkPath	yes	The ZooKeeper path of a leader group. This path can be multiple path segments long. The namer resolves to the address stored in the data of the leader.

Curator

kind: io.l5d.curator

Curator Configuration

A namer that uses the Curator discovery library to resolve names.

Note: If you have registered Curator services with a custom payload object, that class file must be on the classpath. Otherwise you will get a java.lang.IllegalArgumentException: Invalid type id '<some-payload-class' error.

Key	Default Value	Description
prefix	`io.l5d.curator`	Resolves names with `/#/<prefix>`.
experimental	required	Because this namer is still considered experimental, you must set this to `true` to use it.
zkAddrs	required	A list of ZooKeeper addresses, each of which have `host` and `port` parameters.
basePath	`/`	The ZooKeeper path for Curator discovery.

Curator Path Parameters

Dtab Path Format

/#/<prefix>/<serviceName>

Key	Required	Description
prefix	yes	Tells Linkerd to resolve the request path using the curator namer.
serviceName	yes	The name of the Curator service to lookup in ZooKeeper.

Rewrite

kind: io.l5d.rewrite

Rewrite Configuration

Example rewrite configuration:

namers:
- kind: io.l5d.rewrite
  pattern: "/{service}/api"
  name: "/srv/{service}"

Then reference the namer in the dtab to use it:

dtab: |
  /svc => /#/io.l5d.rewrite

A namer that completely rewrites a path. This is useful to do arbitrary reordering of the path segments that is not possible using standard prefix replacement. While this is a general purpose tool for reordering path segments, it cannot be used to modify or split individual segments (for modification or splitting of individual segments, see the rewriting namers section below).

If the name matches the pattern in the config, it will be replaced by the name in the config. Additionally, any variables in the pattern will capture the value of the matching path segment and may be used in the final name.

Key	Default Value	Description
prefix	`io.l5d.rewrite`	Resolves names with `/#/<prefix>`.
pattern	required	If the name matches this prefix, replace it with the name configured in the `name` parameter. Wildcards and variable capture are allowed (see: `io.buoyant.namer.util.PathMatcher`).
name	required	The replacement name. Variables captured in the pattern may be used in this string.

Rewrite Path Parameters

Dtab Path Format

/#/<prefix> [/ *name ]

Key	Required	Description
prefix	yes	Tells Linkerd to resolve the request path using the rewrite namer.
name	yes	Attempt to match this name against the pattern and replace it with the configured name.

Rewriting Namers

In addition to service discovery namers, Linkerd supplies a number of utility namers. These namers assist in path rewriting when the transformation is more complicated than just prefix substitution. They are prefixed with /$/ instead of /#/, and can be used without explicitly adding them to the namers section of the config.

domainToPathPfx

/marathonId => /#/io.l5d.marathon;
/host       => /$/io.buoyant.http.domainToPathPfx/marathonId;
/svc => /host;

Dtab Path Format

/$/io.buoyant.http.domainToPathPfx/<prefix>/<host>

Rewrites the path’s prefix with <prefix> first, followed by each subdomain of <host> separated and in reverse order.

For example, /$/io.buoyant.http.domainToPathPfx/pfx/foo.buoyant.io/resource/name would be rewritten to /pfx/io/buoyant/foo/resource/name.

subdomainOfPfx

/consulSvc  => /#/io.l5d.consul/.local
/host       => /$/io.buoyant.http.subdomainOfPfx/service.consul/consulSvc;
/svc => /host;

Dtab Path Format

/$/io.buoyant.http.subdomainOfPfx/<domain>/<prefix>/<host>

Rewrites the path’s prefix with <prefix> first, followed by <host> with the <domain> dropped.

For example, /$/io.buoyant.http.subdomainOfPfx/buoyant.io/pfx/foo.buoyant.io/resource/name would be rewritten to /pfx/foo/resource/name

hostportPfx

/ip-hostport => /$/inet;
/svc         => /$/io.buoyant.hostportPfx/ip-hostport;

Dtab Path Format

/$/io.buoyant.hostportPfx/<prefix>/<host>:<port>/etc

Rewrites a name of the form “host:ip” as a path with host followed by ip. Does not not support IPv6 host IPs (because ipv6 notation doesn’t work in Paths as- is, due to bracket characters).

For example, /$/io.buoyant.hostportPfx/pfx/host:port/etc would be rewritten to /pfx/host/port/etc.

porthostPfx

/k8s-porthost => /#/io.l5d.k8s/default;
/svc          => /$/io.buoyant.porthostPfx/k8s-porthost;

Dtab Path Format

/$/io.buoyant.porthostPfx/<prefix>/<host>:<port>/etc

Rewrites a name of the form “host:ip” as a path with ip followed by host. Does not not support IPv6 host IPs (because ipv6 notation doesn’t work in Paths as- is, due to bracket characters).

For example, /$/io.buoyant.porthostPfx/pfx/host:port/etc would be rewritten to /pfx/port/host/etc.

Telemetry

Key	Default Value	Description
kind	required	Either `io.l5d.prometheus`, `io.l5d.statsd`, `io.l5d.tracelog`, `io.l5d.recentRequests`, or `io.l5d.zipkin`.
experimental	`false`	Set this to `true` to enable the telemeter if it is experimental.

Prometheus

Example Prometheus config

telemetry:
- kind: io.l5d.prometheus

kind: io.l5d.prometheus

Exposes admin endpoints:

/admin/metrics/prometheus: retrieve all metrics in Prometheus format

This telemeter has no additional parameters.

InfluxDB

Example InfluxDB config

telemetry:
- kind: io.l5d.influxdb

kind: io.l5d.influxdb

This telemeter is intended for collection by Telegraf. Each measurement will have a host tag, set from the Host header on the collector’s incoming request. Recommended Telegraf configuration is using inputs.exec plugin with curl -s http://[LINKERD_IP]/admin/metrics/influxdb.

Exposes admin endpoints:

/admin/metrics/influxdb: retrieve all metrics in InfluxDB LINE protocol format

This telemeter has no additional parameters.

StatsD (experimental)

Example StatD config

telemetry:
- kind: io.l5d.statsd
  experimental: true
  prefix: linkerd
  hostname: 127.0.0.1
  port: 8125
  gaugeIntervalMs: 10000
  sampleRate: 1.0

kind: io.l5d.statsd

StatsD metrics exporting. This telemeter connects to a given StatsD server via UDP. Counters and timers/histograms are exported immediately, based on sample rate. Gauge export interval is configurable.

Key	Default Value	Description
experimental	required	Because this telemeter is still considered experimental, you must set this to `true` to use it.
prefix	`linkerd`	String to prefix all exported metric names with.
hostname	`127.0.0.1`	Hostname of the StatsD server.
port	`8125`	Port of the StatsD server.
gaugeIntervalMs	`10000`	Interval to export Gauges, in milliseconds.
sampleRate	`0.01`	Sample rate to export counter and timing/histogram events. Higher values will result in higher Linkerd latency.

TraceLog

Example TraceLog config

telemetry:
- kind: io.l5d.tracelog
  sampleRate: 0.2
  level: TRACE

kind: io.l5d.tracelog

Log all tracing data, given a log-level and sample rate.

Key	Default Value	Description
host	`localhost`	Host to send trace data to.
sampleRate	`1.0`	What percentage of traces to log.
level	`INFO`	Log-level, one of: `ALL`, `CRITICAL`, `DEBUG`, `ERROR`, `FATAL`, `INFO`, `OFF`, `TRACE`, `WARNING`. For full details, see com.twitter.logging.Level.

Recent Requests

Example Recent Requests config

telemetry:
- kind: io.l5d.recentRequests
  sampleRate: 1.0
  capacity: 10

kind: io.l5d.recentRequests

The recent requests telemeter keeps an in-memory record of recent requests and uses it to populate the recent requests table on the admin dashboard. This table can be viewed at /requests on the admin port. Recording requests can have an impact on Linkerd performance so make sure to set a sample rate that is appropriate for your level of traffic.

Key	Default Value	Description
sampleRate	required	What percentage of traces to record.
capacity	10	The maximum number of recent traces to store

Zipkin telemeter

Example zipkin config

telemetry:
- kind: io.l5d.zipkin
  host: localhost
  port: 9410
  sampleRate: 0.02

kind: io.l5d.zipkin

Finagle’s zipkin-tracer. Use this telemeter to send trace data to a Zipkin Scribe collector.

Finagle’s default Zipkin tracer uses Scribe transport, not HTTP. Therefore the data generated by this tracer must be sent to Zipkin’s Scribe collector, which runs on a separate port (default 9410) from Zipkin’s web UI (default 9411). The Scribe collector is not enabled by default; to enable it, start Zipkin with the SCRIBE_ENABLED=true environment variable set. Alternatively, to use Zipkin’s HTTP collector, use the io.zipkin.http telemeter plugin provided by the linkerd-zipkin repo.

Key	Default Value	Description
host	`localhost`	Host to send trace data to.
port	`9410`	Port to send trace data to (must be the Scribe collector port).
sampleRate	`0.001`	What percentage of requests to trace.

Announcers

An announcer registers servers in service discovery. Each server may specify a list of concrete names to announce as in the announce server key. Each announcer has a prefix and will only announce names that begin with that prefix.

Key	Default Value	Description
kind	required	Only `io.l5d.serversets` is available at this time.
prefix	kind-specific	Announces names beginning with `/#/<prefix>`.

Serversets

kind: io.l5d.serversets

Announce to ZooKeeper using the serverset format.

Key	Default Value	Description
prefix	`io.l5d.serversets`	Announces names beginning with `/#/<prefix>`.
zkAddrs	required	A list of ZooKeeper addresses, each of which have `host` and `port` parameters.
pathPrefix	`/discovery`	The ZooKeeper path under which services should be registered.

Introduction

File Format

Administrative interface

Administrative endpoints

Routers Intro

Namers and Service Discovery

Telemetry Intro

Usage

Routers

Router Parameters

Binding Cache

Server Parameters

Service Configuration

Global Service Config

Static Service Config

Service Parameters

Client Configuration

Global Client Config

Static Client Config

Client Parameters

Host Connection Pool

HTTP/1.1 protocol

HTTP Servers

Adding the Forwarded header

Endpoint labelers

HTTP/1.1 Identifiers

Method and Host Identifier

Identifier Configuration:

Identifier Path Parameters:

Path Identifier

Identifier Configuration:

Identifier Path Parameters:

Header Identifier

Identifier Configuration:

Identifier Path Parameters:

Header Token Identifier

Identifier Configuration:

Identifier Path Parameters:

Ingress Identifier

Identifier Configuration:

Identifier Path Parameters

Static Identifier

Identifier Configuration:

Identifier Path Parameters

HTTP Engines

HTTP Headers

Context Headers

User Headers

Informational Request Headers

Informational Response Headers

HTTP/2 protocol

HTTP/2 Server Parameters

HTTP/2 Client Parameters

HTTP/2 Identifiers

HTTP/2 Header Token identifier

Namer Configuration:

Namer Path Parameters:

HTTP/2 Header Path Identifier

Namer Configuration:

Namer Path Parameters:

HTTP/2 Ingress Identifier

Identifier Configuration:

Identifier Path Parameters

HTTP/2 Headers

HTTP/2 Context Headers

HTTP/2 User Headers

HTTP/2 Informational Request Headers

HTTP/2 Informational Response Headers

Thrift Protocol

Thrift Router Parameters

Thrift Server Parameters

Thrift Client Parameters

ThriftMux Protocol (experimental)

ThriftMux Router Parameters

ThriftMux Server Parameters

ThriftMux Client Parameters

Mux Protocol (experimental)

Mux Router Parameters

Mux Server Parameters

Interpreter

Adding the `Forwarded` header