Skip to content


This is the work in progress operations guide for showing information, which are relevant for production use.

Skipper is proven to scale with number of routes beyond 300.000 routes per instance. Skipper is running with peaks to 65.000 http requests per second using multiple instances.

Connection Options

Skipper’s connection options are allowing you to set Go’s http.Server Options on the client side and http.Transport on the backend side.

“It is recommended to read this blog post about net http timeouts in order to better understand the impact of these settings.


Backend is the side skipper opens a client connection to.

Closing idle connections is required for DNS failover, because Go’s http.Transport caches DNS lookups and needs to create new connections for doing so. Skipper will start a goroutine and use the specified time.Duration to call CloseIdleConnections() on that http.Transport.

-close-idle-conns-period string
    period of closing all idle connections in seconds or as a
    duration string. Not closing when less than 0 (default "20")

This will set MaxIdleConnsPerHost on the http.Transport to limit the number of idle connections per backend such that we do not run out of sockets.

-idle-conns-num int
    maximum idle connections per backend host (default 64)

This will set MaxIdleConns on the http.Transport to limit the number for all backends such that we do not run out of sockets.

-disable-http-keepalives bool
    forces backend to always create a new connection

This will set DisableKeepAlives on the http.Transport to disable HTTP keep-alives and to only use the connection for single request.

-max-idle-connection-backend int
    sets the maximum idle connections for all backend connections

This will set TLSHandshakeTimeout on the http.Transport to have timeouts based on TLS connections.

-tls-timeout-backend duration
    sets the TLS handshake timeout for backend connections (default 1m0s)

This will set Timeout on net.Dialer that is the implementation of DialContext, which is the TCP connection pool used in the http.Transport.

-timeout-backend duration
    sets the TCP client connection timeout for backend connections (default 1m0s)

This will set KeepAlive on net.Dialer that is the implementation of DialContext, which is the TCP connection pool used in the http.Transport.

-keepalive-backend duration
    sets the keepalive for backend connections (default 30s)

This will set DualStack (IPv4 and IPv6) on net.Dialer that is the implementation of DialContext, which is the TCP connection pool used in the http.Transport.

    enables DualStack for backend connections (default true)


Client is the side skipper gets incoming calls from. Here we can set timeouts in different parts of the http connection.

This will set ReadTimeout in http.Server handling incoming calls from your clients.

-read-timeout-server duration
    set ReadTimeout for http server connections (default 5m0s)

This will set ReadHeaderTimeout in http.Server handling incoming calls from your clients.

-read-header-timeout-server duration
    set ReadHeaderTimeout for http server connections (default 1m0s)

This will set WriteTimeout in http.Server handling incoming calls from your clients.

-write-timeout-server duration
    set WriteTimeout for http server connections (default 1m0s)

This will set IdleTimeout in http.Server handling incoming calls from your clients. If you have another loadbalancer layer in front of your Skipper http routers, for example AWS Application Load Balancers, you should make sure, that Skipper’s idle-timeout-server setting is bigger than the idle timeout from the loadbalancer in front. Wrong combinations of idle timeouts can lead to a few unexpected HTTP 502.

-idle-timeout-server duration
    maximum idle connections per backend host (default 1m0s)

This will set MaxHeaderBytes in http.Server to limit the size of the http header from your clients.

-max-header-bytes int
    set MaxHeaderBytes for http server connections (default 1048576)


Skipper implements now controlling the maximum incoming TCP client connections.

The purpose of the mechanism is to prevent Skipper requesting more memory than available in case of too many concurrent connections, especially in an autoscaling deployment setup, in those case when the scaling is not fast enough to follow sudden connection spikes.

This solution relies on a listener implementation combined with a LIFO queue. It allows only a limited number of connections being handled concurrently, defined by the max concurrency configuration. When the max concurrency limit is reached, the new incoming client connections are stored in a queue. When an active (accepted) connection is closed, the most recent pending connection from the queue will be accepted. When the queue is full, the oldest pending connection is closed and dropped, and the new one is inserted into the queue.

The feature can be enabled with the -enable-tcp-queue flag. The maximum concurrency can bet set with the -max-tcp-listener-concurrency flag, or, if this flag is not set, then Skipper tries to infer the maximum accepted concurrency from the system by reading the /sys/fs/cgroup/memory/memory.limit_in_bytes file. In this case, it uses the average expected per request memory requirement, which can be set with the -expected-bytes-per-request flag.

Note that the automatically inferred limit may not work as expected in an environment other than cgroups v1.

OAuth2 Tokeninfo

OAuth2 filters integrate with external services and have their own connection handling. Outgoing calls to these services have a default timeout of 2s, which can be changed by the flag -oauth2-tokeninfo-timeout=<OAuthTokeninfoTimeout>.

OAuth2 Tokenintrospection RFC7662

OAuth2 filters integrate with external services and have their own connection handling. Outgoing calls to these services have a default timeout of 2s, which can be changed by the flag -oauth2-tokenintrospect-timeout=<OAuthTokenintrospectionTimeout>.


Monitoring is one of the most important things you need to run in production and skipper has a godoc page for the metrics package, describing options and most keys you will find in the metrics handler endpoint. The default is listening on :9911/metrics. You can modify the listen port with the -support-listener flag. Metrics can exposed using formats Codahale (json) or Prometheus and be configured by -metrics-flavour=, which defaults to codahale. To expose both formats you can use a comma separated list: -metrics-flavour=codahale,prometheus.


In case you want to get metrics in Prometheus format exposed, use this option to enable it:


It will return Prometheus metrics on the common metrics endpoint :9911/metrics.

To monitor skipper we recommend the following queries:

  • P99 backend latency: histogram_quantile(0.99, sum(rate(skipper_serve_host_duration_seconds_bucket{}[1m])) by (le))
  • HTTP 2xx rate: histogram_quantile(0.99, sum(rate(skipper_serve_host_duration_seconds_bucket{code =~ "2.*"}[1m])) by (le) )
  • HTTP 4xx rate: histogram_quantile(0.99, sum(rate(skipper_serve_host_duration_seconds_bucket{code =~ "4.*"}[1m])) by (le) )
  • HTTP 5xx rate: histogram_quantile(0.99, sum(rate(skipper_serve_host_duration_seconds_bucket{code =~ "52.*"}[1m])) by (le) )
  • Max goroutines (depends on label selector): max(go_goroutines{application="skipper-ingress"})
  • Max threads (depends on label selector): max(go_threads{application="skipper-ingress"})
  • max heap memory in use in MB (depends on label selector): max(go_memstats_heap_inuse_bytes{application="skipper-ingress"}) / 1024 / 1000
  • Max number of heap objects (depends on label selector): max(go_memstats_heap_objects{application="skipper-ingress"})
  • Max of P75 Go GC runtime in ms (depends on label selector): max(go_gc_duration_seconds{application="skipper-ingress",quantile="0.75"}) * 1000 * 1000
  • P99 request filter duration (depends on label selector): histogram_quantile(0.99, sum(rate(skipper_filter_request_duration_seconds_bucket{application="skipper-ingress"}[1m])) by (le) )
  • P99 response filter duration (depends on label selector): histogram_quantile(0.99, sum(rate(skipper_filter_response_duration_seconds_bucket{application="skipper-ingress"}[1m])) by (le) )
  • If you use Kubernetes limits or Linux cgroup CFS quotas (depends on label selector): sum(rate(container_cpu_cfs_throttled_periods_total{container_name="skipper-ingress"}[1m]))

Connection metrics

This option will enable known loadbalancer connections metrics, like counters for active and new connections. This feature sets a metrics callback on http.Server and uses a counter to collect http.ConnState.

    enables connection metrics for http server connections

It will expose them in /metrics, for example json structure looks like this example:

  "counters": {
    "": {
      "count": 6
    "": {
      "count": 6
    "": {
      "count": 6
    "": {
      "count": 6
  /* stripped a lot of metrics here */

LIFO metrics

When enabled in the routes, LIFO queues can control the maximum concurrency level proxied to the backends and mitigate the impact of traffic spikes. The current level of concurrency and the size of the queue can be monitored with gauges per each route using one of the lifo filters. To enable monitoring for the lifo filters, use the command line option:


When queried, it will return metrics like:

  "gauges": {
    "": {
      "value": 245
    "skipper.lifo.routeXYZ.queued": {
      "value": 27

Application metrics

Application metrics for your proxied applications you can enable with the option:

    enables reporting total serve time metrics for each host
    enables reporting total serve time metrics for each route

This will make sure you will get stats for each “Host” header or the route name as “timers”. The following is an example for -serve-host-metrics:

"timers": {
  "skipper.servehost.app1_example_com.GET.200": {
    "15m.rate": 0.06830666203045982,
    "1m.rate": 2.162612637718806e-06,
    "5m.rate": 0.008312609284452856,
    "75%": 236603815,
    "95%": 236603815,
    "99%": 236603815,
    "99.9%": 236603815,
    "count": 3,
    "max": 236603815,
    "mean": 116515451.66666667,
    "mean.rate": 0.0030589345776699827,
    "median": 91273391,
    "min": 21669149,
    "stddev": 89543653.71950394
  "skipper.servehost.app1_example_com.GET.304": {
    "15m.rate": 0.3503336738177459,
    "1m.rate": 0.07923086447313292,
    "5m.rate": 0.27019839341602214,
    "75%": 99351895.25,
    "95%": 105381847,
    "99%": 105381847,
    "99.9%": 105381847,
    "count": 4,
    "max": 105381847,
    "mean": 47621612,
    "mean.rate": 0.03087161486272533,
    "median": 41676170.5,
    "min": 1752260,
    "stddev": 46489302.203724876
  "skipper.servehost.app1_example_com.GET.401": {
    "15m.rate": 0.16838468990057648,
    "1m.rate": 0.01572861413072501,
    "5m.rate": 0.1194724817779537,
    "75%": 91094832,
    "95%": 91094832,
    "99%": 91094832,
    "99.9%": 91094832,
    "count": 2,
    "max": 91094832,
    "mean": 58090623,
    "mean.rate": 0.012304914018033056,
    "median": 58090623,
    "min": 25086414,
    "stddev": 33004209

Note you can reduce the dimension of the metrics by removing the HTTP status code and method from it. Use the -serve-method-metric=false and/or -serve-status-code-metric=false. Both flags are enabled by default. For prometheus metrics flavour, a counter with both the HTTP method and status code can be enabled with -serve-host-counter or -serve-route-counter, even if these flags are disabled.

To change the sampling type of how metrics are handled from uniform to exponential decay, you can use the following option, which is better for not so huge utilized applications (less than 100 requests per second):

    use exponentially decaying sample in metrics

Go metrics

Metrics from the go runtime memstats are exposed from skipper to the metrics endpoint, default listener :9911, on path /metrics :

"gauges": {
  "skipper.runtime.MemStats.Alloc": {
    "value": 3083680
  "skipper.runtime.MemStats.BuckHashSys": {
    "value": 1452675
  "skipper.runtime.MemStats.DebugGC": {
    "value": 0
  "skipper.runtime.MemStats.EnableGC": {
    "value": 1
  "skipper.runtime.MemStats.Frees": {
    "value": 121
  "skipper.runtime.MemStats.HeapAlloc": {
    "value": 3083680
  "skipper.runtime.MemStats.HeapIdle": {
    "value": 778240
  "skipper.runtime.MemStats.HeapInuse": {
    "value": 4988928
  "skipper.runtime.MemStats.HeapObjects": {
    "value": 24005
  "skipper.runtime.MemStats.HeapReleased": {
    "value": 0
  "skipper.runtime.MemStats.HeapSys": {
    "value": 5767168
  "skipper.runtime.MemStats.LastGC": {
    "value": 1516098381155094500
  "skipper.runtime.MemStats.Lookups": {
    "value": 2
  "skipper.runtime.MemStats.MCacheInuse": {
    "value": 6944
  "skipper.runtime.MemStats.MCacheSys": {
    "value": 16384
  "skipper.runtime.MemStats.MSpanInuse": {
    "value": 77368
  "skipper.runtime.MemStats.MSpanSys": {
    "value": 81920
  "skipper.runtime.MemStats.Mallocs": {
    "value": 1459
  "skipper.runtime.MemStats.NextGC": {
    "value": 4194304
  "skipper.runtime.MemStats.NumGC": {
    "value": 0
  "skipper.runtime.MemStats.PauseTotalNs": {
    "value": 683352
  "skipper.runtime.MemStats.StackInuse": {
    "value": 524288
  "skipper.runtime.MemStats.StackSys": {
    "value": 524288
  "skipper.runtime.MemStats.Sys": {
    "value": 9246968
  "skipper.runtime.MemStats.TotalAlloc": {
    "value": 35127624
  "skipper.runtime.NumCgoCall": {
    "value": 0
  "skipper.runtime.NumGoroutine": {
    "value": 11
  "skipper.runtime.NumThread": {
    "value": 9
"histograms": {
  "skipper.runtime.MemStats.PauseNs": {
    "75%": 82509.25,
    "95%": 132609,
    "99%": 132609,
    "99.9%": 132609,
    "count": 12,
    "max": 132609,
    "mean": 56946,
    "median": 39302.5,
    "min": 28749,
    "stddev": 31567.015005117817


Redis - Rate limiting metrics

Timer metrics for the latencies and errors of the communication with the auxiliary Redis instances are enabled by the default, and exposed among the timers via the following keys:

  • skipper.swarm.redis.query.allow.success: successful allow requests to the rate limiter, ungrouped
  • skipper.swarm.redis.query.allow.failure: failed allow requests to the rate limiter, ungrouped, where the redis communication failed
  • skipper.swarm.redis.query.retryafter.success.: successful allow requests to the rate limiter, grouped by the rate limiter group name when used
  • skipper.swarm.redis.query.retryafter.failure.: failed allow requests to the rate limiter, ungrouped, where the redis communication faileds, grouped by the rate limiter group name when used

See more details about rate limiting at Rate limiting.


Skipper has support for different OpenTracing API vendors, including jaeger, lightstep and instana.

You can configure tracing implementations with a flag and pass information and tags to the tracer:

-opentracing=<vendor> component-name=skipper-ingress ... tag=cluster=mycluster ...

The best tested tracer is the lightstep tracer, because we use it in our setup. In case you miss something for your chosen tracer, please open an issue or pull request in our repository.

Skipper creates up to 5 different spans: Spans

Some Tag details are added to all spans. Span details

Ingress span

The Ingress span is active from getting the request in Skipper’s main http handler, until we served the response to the client of the request.


  • component: skipper
  • hostname: ip-10-149-64-142
  • http.method: GET
  • http.path: /
  • http.remote_addr:
  • http.url: /
  • span.kind: server

Ingress span with tags

Proxy span

The Proxy span starts just before executing the backend call.


  • component: skipper
  • hostname: ip-10-149-65-70
  • http.method: GET
  • http.path: /
  • http.remote_addr:
  • http.status_code: 200
  • http.url:
  • skipper.route_id: kube_default__example_ingress_hostname_example_org____example_backend
  • span.kind: client

Proxy span with tags

Proxy span has logs to measure connect (dial_context), http roundtrip (http_roundtrip), stream headers from backend to client (stream_Headers), stream body from backend to client (streamBody.byte) and events by the Go runtime.

Proxy span with logs

In addition to the manual instrumented proxy client logs, we use net/http/httptrace.ClientTrace to show events by the Go runtime. Full logs of the Proxy span:

Proxy span logs

Request filters span

The request filters span logs show start and end events for each filter applied.

request filter span with logs

Response filters span

The response filters span logs show start and end events for each filter applied.

response filter span with logs

Request and response filters event logging can be disabled by setting the -opentracing-log-filter-lifecycle-events=false flag and span creation can be disabled altogether by the -opentracing-disable-filter-spans flag.

Auth filters span

Auth filters are special, because they might call an authorization endpoint, which should be also visible in the trace. This span can have the name “tokeninfo”, “tokenintrospection” or “webhook” depending on the filter used by the matched route.

Tags: - http.url:

The auth filters have trace log values start and end for DNS, TCP connect, TLS handshake and connection pool:

tokeninfo auth filter span with logs

Redis rate limiting spans

Operation: redis_allow_check_card

Operation executed when the cluster rate limiting relies on the auxiliary Redis instances, and the Allow method checks if the rate exceeds the configured limit.

Operation: redis_allow_add_card

Operation setting the counter of the measured request rate for cluster rate limiting with auxiliary Redis instances.

Operation: redis_oldest_score

Operation querying the oldest request event for the rate limiting Retry-After header with cluster rate limiting when used with auxiliary Redis instances.


Dataclients poll some kind of data source for routes. To change the timeout for calls that polls a dataclient, which could be the Kubernetes API, use the following option:

-source-poll-timeout int
    polling timeout of the routing data sources, in milliseconds (default 3000)

Routing table information

Skipper allows you to get some runtime insights. You can get the current routing table from skipper with in the eskip file format:

curl localhost:9911/routes
-> "http://localhost:12345/"

You also can get the number of routes X-Count and the UNIX timestamp of the last route table update X-Timestamp, using a HEAD request:

curl -I localhost:9911/routes
HTTP/1.1 200 OK
Content-Type: text/plain
X-Count: 1
X-Timestamp: 1517777628
Date: Sun, 04 Feb 2018 20:54:31 GMT

The number of routes given is limited (1024 routes by default). In order to control this limits, there are two parameters: limit and offset. The limit defines the number of routes to get and offset where to start the list. Thanks to this, it’s possible to get the results paginated or getting all of them at the same time.

curl localhost:9911/routes?offset=200&limit=100

Memory consumption

While Skipper is generally not memory bound, some features may require some attention and planning regarding the memory consumption.

Potentially high memory consumers:

  • Metrics
  • Filters
  • Slow Backends and chatty clients

Make sure you monitor backend latency, request and error rates. Additionally use Go metrics for the number of goroutines and threads, GC pause times should be less than 1ms in general, route lookup time, request and response filter times and heap memory.


Memory consumption of metrics are dependent on enabled command line flags. Make sure to monitor Go metrics.

If you use -metrics-flavour=codahale,prometheus you enable both storage backends.

If you use the Prometheus histogram buckets -histogram-metric-buckets.

If you enable route based -route-backend-metrics -route-response-metrics -serve-route-metrics, error codes -route-response-metrics and host -serve-host-metrics based metrics it can count up. Please check the support listener endpoint (default 9911) to understand the usage:

% curl localhost:9911/metrics

By default, the route and host metrics include the labels for the request HTTP response status code and the HTTP method. You can customize it by setting -serve-method-metric=false and/or -serve-status-code-metric=false. These two flags will enable or disable the method and status code labels from your metrics reducing the number of metrics generated and memory consumption.


Ratelimit filter clusterClientRatelimit implementation using the swim based protocol, consumes roughly 15MB per filter for 100.000 individual clients and 10 maximum hits. Make sure you monitor Go metrics. Ratelimit filter clusterClientRatelimit implementation using the Redis ring based solution, adds 2 additional roundtrips to redis per hit. Make sure you monitor redis closely, because skipper will fallback to allow traffic if redis can not be reached.

Slow Backends

Skipper has to keep track of all active connections and http Requests. Slow Backends can pile up in number of connections, that will consume each a little memory per request. If you have high traffic per instance and a backend times out it can start to increase your memory consumption. Make sure you monitor backend latency, request and error rates.

Default Filters

Default filters will be applied to all routes created or updated.

Global Default Filters

Global default filters can be specified via two different command line flags -default-filters-prepend and -default-filters-append. Filters passed to these command line flags will be applied to all routes. The difference prepend and append is where in the filter chain these default filters are applied.

For example a user specified the route: r: * -> setPath("/foo") If you run skipper with -default-filters-prepend=enableAccessLog(4,5) -> lifo(100,100,"10s"), the actual route will look like this: r: * -> enableAccessLog(4,5) -> lifo(100,100,"10s") -> setPath("/foo"). If you run skipper with -default-filters-append=enableAccessLog(4,5) -> lifo(100,100,"10s"), the actual route will look like this: r: * -> setPath("/foo") -> enableAccessLog(4,5) -> lifo(100,100,"10s").

Kubernetes Default Filters

Kubernetes dataclient supports default filters. You can enable this feature by specifying default-filters-dir. The defined directory must contain per-service filter configurations, with file name following the pattern ${service}.${namespace}. The content of the files is the actual filter configurations. These filters are then prepended to the filters already defined in Ingresses.

The default filters are supposed to be used only if the filters of the same kind are not configured on the Ingress resource. Otherwise, it can and will lead to potentially contradicting filter configurations and race conditions, i.e. you should specify a specific filter either on the Ingress resource or as a default filter.


HTTP request schedulers change the queuing behavior of in-flight requests. A queue has two generic properties: a limit of requests and a concurrency level. The limit of request can be unlimited (unbounded queue), or limited (bounded queue). The concurrency level is either limited or unlimited.

The default scheduler is an unbounded first in first out (FIFO) queue, that is provided by Go’s standard library.

Skipper provides 2 last in first out (LIFO) filters to change the scheduling behavior.

On failure conditions, Skipper will return HTTP status code:

  • 503 if the queue is full, which is expected on the route with a failing backend
  • 502 if queue access times out, because the queue access was not fast enough
  • 500 on unknown errors, please create an issue

The problem

Why should you use boundaries to limit concurrency level and limit the queue?

The short answer is resiliency. If you have one route, that is timing out, the request queue of skipper will pile up and consume much more memory, than before. This can lead to out of memory kill, which will affect all other routes. In this comment you can see the memory usage increased in Go’s standard library bufio package.

Why LIFO queue instead of FIFO queue?

In normal cases the queue should not contain many requests. Skipper is able to process many requests concurrently without letting the queue piling up. In overrun situations you might want to process at least some fraction of requests instead of timing out all requests. LIFO would not time out all requests within the queue, if the backend is capable of responding some requests fast enough.

A solution

Skipper has two filters lifo() and lifoGroup(), that can limit the number of requests for a route. A documented load test shows the behavior with an enabled lifo(100,100,"10s") filter for all routes, that was added by default. You can do this, if you pass the following flag to skipper: -default-filters-prepend=lifo(100,100,"10s").

Both LIFO filters will, use a last in first out queue to handle most requests fast. If skipper is in an overrun mode, it will serve some requests fast and some will timeout. The idea is based on Dropbox bandaid proxy, which is not opensource. Dropbox shared their idea in a public blogpost.

Skipper’s scheduler implementation makes sure, that one route will not interfere with other routes, if these routes are not in the same scheduler group. LifoGroup has a user chosen scheduler group and lifo() will get a per route unique scheduler group.

URI standards interpretation

Considering the following request path: /foo%2Fbar, Skipper can handle it in two different ways. The current default way is that when the request is parsed purely relying on the Go stdlib url package, this path becomes /foo/bar. According to RFC 2616 and RFC 3986, this may be considered wrong, and this path should be parsed as /foo%2Fbar. This is possible to achieve centrally, when Skipper is started with the -rfc-patch-path flag. It is also possible to allow the default behavior and only force the alternative interpretation on a per-route basis with the rfcPath() filter. See rfcPath().

If the second interpretation gets considered the right way, and the other one a bug, then the default value for this flag may become to be on.

Debugging Requests

Skipper provides filters, that can change HTTP requests. You might want to inspect how the request was changed, during the route processing and check the request that would be made to the backend. Luckily with -debug-listener=:9922, Skipper can provide you this information.

For example you have the following route:

kube_default__foo__foo_teapot_example_org_____foo: Host(/^foo[.]teapot[.]example[.]org$/) && PathSubtree("/")
  -> setRequestHeader("X-Foo", "hello-world")
  -> <roundRobin, "", "">;

If you sent now a request to the debug listener, that will be matched by the route, Skipper will respond with information that show you the matched route, the incoming request, the transformed request and all predicates and filters involved in the route processing:

% curl -s -H"Host:" | jq .
  "route_id": "kube_default__foo__foo_teapot_example_org_____foo",
  "route": "Host(/^foo[.]teapot[.]example[.]org$/) && PathSubtree(\"/\") -> setRequestHeader(\"X-Foo\", \"hello-world\") -> <roundRobin, \"\", \"\">",
  "incoming": {
    "method": "GET",
    "uri": "/",
    "proto": "HTTP/1.1",
    "header": {
      "Accept": [
      "User-Agent": [
    "host": "",
    "remote_address": ""
  "outgoing": {
    "method": "GET",
    "uri": "",
    "proto": "HTTP/1.1",
    "header": {
      "Accept": [
      "User-Agent": [
      "X-Foo": [
    "host": ""
  "response_mod": {
    "header": {
      "Server": [
  "filters": [
      "name": "setRequestHeader",
      "args": [
  "predicates": [
      "name": "PathSubtree",
      "args": [

Profiling skipper

Go profiling is explained in Go’s diagnostics documentation.

To enable profiling in skipper you have to use -enable-profile. This will start a profiling route at /debug/pprof/profile on the support listener, which defaults to :9911.

Profiling example

Start skipper with enabled profiling:

skipper -inline-routes='r1: * -> inlineContent("hello") -> <shunt>' -enable-profile

Use Go tool pprof to download profiling sample to analyze (sample is not from the example):

go tool pprof
Fetching profile over HTTP from
Saved profile in /$HOME/pprof/pprof.skipper.samples.cpu.004.pb.gz
File: skipper
Build ID: 272c31a7bd60c9fabb637bdada37a3331a919b01
Type: cpu
Time: Oct 7, 2020 at 6:17pm (CEST)
Duration: 30s, Total samples = 0
No samples were found with the default sample value type.
Try "sample_index" command to analyze different sample values.
Entering interactive mode (type "help" for commands, "o" for options)
(pprof) top
Showing nodes accounting for 2140ms, 50.00% of 4280ms total
Dropped 330 nodes (cum <= 21.40ms)
Showing top 10 nodes out of 303
      flat  flat%   sum%        cum   cum%
     560ms 13.08% 13.08%      640ms 14.95%  syscall.Syscall
     420ms  9.81% 22.90%      430ms 10.05%  runtime.nanotime
     410ms  9.58% 32.48%      410ms  9.58%  runtime.futex
     170ms  3.97% 36.45%      450ms 10.51%  runtime.mallocgc
     170ms  3.97% 40.42%      180ms  4.21%  runtime.walltime
     160ms  3.74% 44.16%      220ms  5.14%  runtime.scanobject
      80ms  1.87% 46.03%       80ms  1.87%  runtime.heapBitsSetType
      70ms  1.64% 47.66%       70ms  1.64%  runtime.epollwait
      50ms  1.17% 48.83%      120ms  2.80%  compress/flate.(*compressor).deflate
      50ms  1.17% 50.00%       50ms  1.17%  encoding/json.stateInString
(pprof) web
--> opens browser with SVG

pprof svg in web browser

Response serving

When serving a response from a backend, Skipper serves first the HTTP response headers. After that Skipper streams the response payload and uses one 8kB buffer to stream the data through this 8kB buffer. It uses Flush() to make sure the 8kB chunk is written to the client. Details can be observed by opentracing in the logs of the Proxy Span.

Forwarded headers

Skipper can be configured to add X-Forwarded-* headers:

  -forwarded-headers value
        comma separated list of headers to add to the incoming request before routing
        X-Forwarded-For sets or appends with comma the remote IP of the request to the X-Forwarded-For header value
        X-Forwarded-Host sets X-Forwarded-Host value to the request host
        X-Forwarded-Port=<port> sets X-Forwarded-Port value
        X-Forwarded-Proto=<http|https> sets X-Forwarded-Proto value
  -forwarded-headers-exclude-cidrs value
        disables addition of forwarded headers for the remote host IPs from the comma separated list of CIDRs

Converting Routes

For migrations you need often to convert X to Y. This is also true in case you want to switch one predicate to another one or one filter to another one. In skipper we have -edit-route and -clone-route that either modifies matching routes or copy matching routes and change the copy.


A route with edit-route

% skipper -inline-routes='Path("/foo") -> setResponseHeader("X-Foo","bar") -> inlineContent("hi") -> <shunt>' \
-edit-route='/inlineContent[(]["](.*)["][)]/inlineContent("modified \"$1\" response")/'
[APP]INFO[0000] Expose metrics in codahale format
[APP]INFO[0000] support listener on :9911
[APP]INFO[0000] route settings, reset, route: : Path("/foo") -> setResponseHeader("X-Foo", "bar") -> inlineContent("hi") -> <shunt>
[APP]INFO[0000] proxy listener on :9090
[APP]INFO[0000] TLS settings not found, defaulting to HTTP
[APP]INFO[0000] route settings received
[APP]INFO[0000] route settings applied

Modified route:

curl localhost:9911/routes
  -> setResponseHeader("X-Foo", "bar")
  -> inlineContent("modified \"hi\" response")
  -> <shunt>

Modified response body:

% curl -v http://localhost:9090/foo
*   Trying ::1...
* Connected to localhost (::1) port 9090 (#0)
> GET /foo HTTP/1.1
> Host: localhost:9090
> User-Agent: curl/7.49.0
> Accept: */*
< HTTP/1.1 200 OK
< Content-Length: 22
< Content-Type: text/plain; charset=utf-8
< Server: Skipper
< X-Foo: bar
< Date: Thu, 14 Oct 2021 08:41:53 GMT
* Connection #0 to host localhost left intact
modified "hi" response

With edit-route and -clone-route you can modify Predicates and Filters to convert from SourceFromLast() to ClientIP, for example if you want to migrate AWS cloud load balancer from Application Load Balancer to Network Load Balancer, you can use -clone-route='/SourceFromLast[(](.*)[)]/ClientIP($1)/' to create additional routes for

r: SourceFromLast("","2001:67c:20a0::/48") -> ...`
to change to
r: SourceFromLast("","2001:67c:20a0::/48") -> ...`
clone_r: ClientIP("","2001:67c:20a0::/48") -> ...`
for migration time.

/ symbol is not the only option for the separator for -edit-route and -clone-route, any first symbol you will specify in those options could be used as separator. This could be useful for IP mask changes, for example, you can use -edit-route='#/26#/24#. In this case

r: SourceFromLast("","2001:67c:20a0::/48") -> ...`
will be changed to
r: SourceFromLast("","2001:67c:20a0::/48") -> ...`