Skip to content

Latest commit

 

History

History
572 lines (451 loc) · 23.1 KB

README.md

File metadata and controls

572 lines (451 loc) · 23.1 KB
Raw

Crawlera Headless Proxy

Build Status Go Report Card

Crawlera Headless proxy is a proxy which main intent is to help users with headless browsers to use Crawlera. This includes different implementations of headless browsers such as Splash, headless Chrome, and Firefox. Also, this proxy should help users of such frameworks as Selenium and Puppeteer to use Crawlera without a need to build Squid chains or install Polipo.

The biggest problem with headless browsers is their configuration:

  1. Crawlera uses proxy authentication protocol described in RFC 7235 but it is rather hard to configure such authentication in headless browsers. The most popular way of bypassing this problem is to use Polipo which is, unfortunately, unsupported for a long time.
  2. Crawlera uses X-Headers as configuration. To use this API with headless browsers, users have to install plugins or extensions in their browsers and configure them to propagate such headers to Crawlera.
  3. Also, it is rather hard and complex to maintain best practices of using these headers. For example, support of Browser Profiles requires to have a minimal possible set of headers. For example, it is recommended to remove Accept header by default. It is rather hard to do that using headless browsers API.
  4. Crawlera works best with browsers only with some preconditions which users have to repeat everytime: session usage, some recommended headers like Referer etc.

Crawlera Headless Proxy intended to help users to avoid such problems. You should generally think about it as a proxy which should be accessible by your headless browser of Selenium grid. This proxy propagates your requests to Crawlera maintaining API key and injecting headers into the requests. Basically, you have to do a bare minimum:

  1. Get Crawlera API key
  2. Run this proxy on your local machine or any machine accessible by headless browser, configuring it with a configuration file, command line parameters or environment variables.
  3. Propagate TLS certificate of this proxy to your browsers or operating system vault.
  4. Access this proxy as a local proxy, plain, without any authentication.

Installation

Check out the Using Headless Browsers with Zyte Smart Proxy Manager manual.

Install binaries

There are some prebuilt binaries available on Release pages. Please download required one for your operating system and CPU architecture.

Install from sources

Install prerequisites

You need to have a distributions of Go programming language, git, bash and make installed. We use Go >= 1.11 so please be sure that you have fresh enough version.

To install them on Ubuntu/Debian, please execute the following command:

$ sudo apt install -y bash make git golang-go

If you have Ubuntu older than 18.10, please install go with snap package:

$ sudo snap install --classic go

To install them on OS X with Homebrew, please execute the following command:

$ brew install go make git

Install from HomeBrew

If you use HomeBrew, you can use it to install headless proxy:

$ curl -L https://raw.githubusercontent.com/zytedata/zyte-smartproxy-headless-proxy/master/crawlera-headless-proxy.rb > crawlera-headless-proxy.rb && brew install --HEAD crawlera-headless-proxy.rb

Build binary

$ git clone https://github.com/zytedata/zyte-smartproxy-headless-proxy.git
$ cd zyte-smartproxy-headless-proxy

The next step is to execute make

$ make

This will build binary crawlera-headless-proxy. If you are interested in compiling for other OS/CPU architecture, please cross-compile:

$ make crosscompile

You'll find a set of compiled biaries in ./ccbuilds directory after the process is finished.

Docker container

To download prebuilt container, please do the following:

$ docker pull zytedata/zyte-smartproxy-headless-proxy

If you want to build this image locally, please do it with make (also, be sure that docker is installed).

$ make docker

This will build an image with tag crawlera-headless-proxy. It can be configured by environment variables or command flags. Default configuration file path within a container is /config.toml.

If you want to have a smaller image (but build time will grow a lot), you can build it with docker-slim make target.

$ make docker-slim

Usage

Help output

$ crawlera-headless-proxy --help
usage: crawlera-headless-proxy [<flags>]

Local proxy for Crawlera to be used with headless browsers.

Flags:
      --help                 Show context-sensitive help (also try --help-long and --help-man).
  -d, --debug                Run in debug mode.
  -b, --bind-ip=BIND-IP      IP to bind to. Default is 127.0.0.1.
  -m, --proxy-api-ip=PROXY-API-IP
                             IP to bind proxy API to. Default is the bind-ip value.
  -p, --bind-port=BIND-PORT  Port to bind to. Default is 3128.
  -w, --proxy-api-port=PROXY-API-PORT
                             Port to bind proxy api to. Default is 3130.
  -c, --config=CONFIG        Path to configuration file.
  -l, --tls-ca-certificate=TLS-CA-CERTIFICATE
                             Path to TLS CA certificate file.
  -r, --tls-private-key=TLS-PRIVATE-KEY
                             Path to TLS private key.
  -t, --no-auto-sessions     Disable automatic session management.
  -n, --concurrent-connections=CONCURRENT-CONNECTIONS
                             Number of concurrent connections.
  -a, --api-key=API-KEY      API key to Crawlera.
  -u, --crawlera-host=CRAWLERA-HOST
                             Hostname of Crawlera. Default is proxy.crawlera.com.
  -o, --crawlera-port=CRAWLERA-PORT
                             Port of Crawlera. Default is 8010.
  -v, --dont-verify-crawlera-cert
                             Do not verify Crawlera own certificate.
  -x, --xheader=XHEADER ...  Crawlera X-Headers.
  -k, --adblock-list=ADBLOCK-LIST ...
                             A list to requests to filter out (ADBlock compatible).
  -z, --direct-access-hostpath-regexps=DIRECT-ACCESS-HOSTPATH-REGEXPS ...
                             A list of regexps for hostpath for direct access, bypassing Crawlera.
  -e, --direct-access-except-hostpath-regexps=DIRECT-ACCESS-EXCEPT-HOSTPATH-REGEXPS ...
                             A list of regexps for hostpath for direct access. No effect without `-z`. Not required.
      --version              Show application version.

Docker example:

$ docker run --name crawlera-headless-proxy -p 3128:3128 zytedata/zyte-smartproxy-headless-proxy --help

Configuration

Defaults are sensible. If you run this tool without any configuration, it will start HTTP/HTTPS proxy on localhost:3128. The only thing you usually need to do is to propagate API key.

$ crawlera-headless-proxy -a myapikey

This will start local HTTP/HTTPS proxy on localhost:3128 and will proxy all requests to proxy.crawlera.com:8010 with API key myapikey.

Also, it is possible to configure this tool using environment variables. Here is the complete table of configuration options and corresponding environment variables.

Description Environment variable Comandline parameter Parameter in configuration file Default value
Run in debug/verbose mode. CRAWLERA_HEADLESS_DEBUG -d, --debug debug false
Which IP this tool should listen on (0.0.0.0 for all interfaces). CRAWLERA_HEADLESS_BINDIP -b, --bind-ip bind_ip 127.0.0.1
Which port this tool should listen. CRAWLERA_HEADLESS_BINDPORT -p, --bind-port bind_port 3128
Path to the configuration file. CRAWLERA_HEADLESS_CONFIG -c, --config -
API key of Crawlera. CRAWLERA_HEADLESS_APIKEY -a, --api-key api_key
Hostname of Crawlera. CRAWLERA_HEADLESS_CHOST -u, --crawlera-host crawlera_host proxy.crawlera.com
Port of Crawlera. CRAWLERA_HEADLESS_CPORT -o, --crawlera-port crawlera_port 8010
Do not verify Crawlera own TLS certificate. CRAWLERA_HEADLESS_DONTVERIFY -v, --dont-verify-crawlera-cert dont_verify_crawlera_cert false
Path to own TLS CA certificate. CRAWLERA_HEADLESS_TLSCACERTPATH -l, --tls-ca-certificate tls_ca_certificate
Path to own TLS private key. CRAWLERA_HEADLESS_TLSPRIVATEKEYPATH -r, --tls-private-key tls_private_key
Disable automatic session management CRAWLERA_HEADLESS_NOAUTOSESSIONS -t, --no-auto-sessions no_auto_sessions false
Maximal ammount of concurrent connections to process CRAWLERA_HEADLESS_CONCURRENCY -n, --concurrent-connections concurrent_connections 0
Additional Crawlera X-Headers. CRAWLERA_HEADLESS_XHEADERS -x, --xheaders Section xheaders
Adblock-compatible filter lists. CRAWLERA_HEADLESS_ADBLOCKLISTS -k, --adblock-list adblock_lists
Regular expressions for hostpath URL part for direct access, bypassing Crawlera. CRAWLERA_HEADLESS_DIRECTACCESS -z, --direct-access-hostpath-regexps direct_access_hostpath_regexps
Exceptions to DirectAccess. Always proxied irrespective of direct acces regex. CRAWLERA_HEADLESS_DIRECTACCESS_EXCEPT -e, --direct-access-except-hostpath-regexps direct_access_except_hostpath_regexps
Which IP should proxy API listen on (default is bind-ip value). CRAWLERA_HEADLESS_PROXYAPIIP -m, --proxy-api-ip proxy_api_ip <same as bind_ip>
Which port proxy API should listen on. CRAWLERA_HEADLESS_PROXYAPIPORT -w, --proxy-api-port proxy_api_port 3130

0 concurrent connections means unlimited. Embedded TLS key/certificate means that headless proxy will use ones from the repository.

Configuration is implemented in TOML language. If you haven't heard about TOML, please consider it as a hardened INI configuration file. Every configuration goes to top-level section (unnamed). X-Headers go to its own section. Let's express following command line in the configuration file:

$ crawlera-headless-proxy -b 0.0.0.0 -p 3129 -u proxy.crawlera.com -o 8010 -x profile=desktop -x cookies=disable

Configuration file will look like:

bind_ip = "0.0.0.0"
bind_port = 3129
crawlera_host = "proxy.crawlera.com"
crawlera_port = 8010

[xheaders]
profile = "desktop"
cookies = "disable"

You can use both command line flags, environment variables, and configuration files. This tool will resolve these options according to this order (1 has max priority, 4 - minimal):

  1. Environment variables
  2. Commandline flags
  3. Configuration file
  4. Defaults

Docker example:

$ docker run --name crawlera-headless-proxy -p 3128:3128 zytedata/zyte-smartproxy-headless-proxy -a $APIKEY -d -x profile=pass -x cookies=disable -x no-bancheck=1 --direct-access-hostpath-regexps=".*?\.(?:txt|json|css|less|js|mjs|cjs|gif|ico|jpe?g|svg|png|webp|mkv|mp4|mpe?g|webm|eot|ttf|woff2?)$" --adblock-list="https://easylist.to/easylist/easylist.txt" --adblock-list="https://easylist.to/easylist/easyprivacy.txt"

Concurrency

There is a limiter on maximal amount of concurrent connections --concurrent-connections. This is required because default Crawlera limits the number of concurrent connections based on the billing plan of the user. If the user exceeds this amount, Crawlera returns a response with status code 429. This can be rather irritating so there is internal limiter which is more friendly to the browsers. You need to set up a number of concurrent connections for your plan and crawlera-headless-proxy will throttle your requests before they will go to Crawlera. It won't send 429 back, it just holds excess requests.

Automatic session management

Crawlera allows using sessions and sessions are natural if we are talking about browsers. Session binds a certain IP to some session ID so all requests will go through the same IP, in the same way as ordinary work with browser looks like. It can slow down your crawl but increase its quality for some websites.

The current implementation of automatic session management is done with the assumption that only one browser is used to access this proxy. There is no clear and simple way how to distinguish the browsers accessing this proxy concurrently.

Basic behavior is here:

  1. If the session is not created, it would be created on the first request.
  2. Until session is known, all other requests are on hold
  3. After session id is known, other requests will start to use that session.
  4. If the session became broken, all requests are set on hold until the new session will be created.
  5. All requests which were failed because of a broken session would be retried with new. If a new session is not ready yet, they will wait until this moment.

Such retries will be done only once because they might potentially block browser for a long time. All retries are also done with 30 seconds timeout.

Adblock list support

crawlera-headless-proxy supports preventive filtering by; adblock-compatible filter lists like EasyList. If you start the tool with such lists, they are going to be downloaded and requests to trackers/advertising platforms will be filtered. This will save you a lot of throughput and requests passed to Crawlera.

If you do not pass any list, such filtering won't be enabled. The list we recommend to use is EasyList (please do not forget to add region-specific lists), EasyPrivacy and Disconnect.

Direct access

Sometimes you want to save a capacity and execute requests bypassing Crawlera. These requests can include some static assets, text files or anything else where Crawlera is not necessary.

You can specify a list of regular expressions which matches host + path parts of URL for direct access from headless proxy, ignoring Crawlera.

TLS keys

Since crawlera-headless-proxy has to inject X-Headers into responses, it works with your browser only by HTTP 1.1. Unfortunately, there is no clear way how to hijack HTTP2 connections. Also, since it is effectively MITM proxy, you need to use its own TLS certificate. This is hardcoded into the binary so you have to download it and apply it to your system. Please consult with manuals of your operating system how to do that.

Link to certificate is Its SHA256 checksum is 100c7dd015814e7b8df16fc9e8689129682841d50f9a1b5a8a804a1eaf36322d.

If you want to have your own certificate, please generate it. The simplest way to do that is to execute the following command:

$ openssl req -x509 -newkey rsa:4096 -keyout private-key.pem -out ca.crt -days 3650 -nodes

This command will generate TLS private key private-key.pem and self-signed certificate ca.crt.

Proxy API

crawlera-headless-proxy has its own HTTP Rest API which is bind to another port. Right now only one endpoint is supported.

GET /stats

This endpoint returns various statistics on the current work of proxy.

Example:

{
  "requests_number": 423,
  "crawlera_requests": 426,
  "crawlera_errors": 0,
  "all_errors": 6,
  "adblocked_requests": 0,
  "sessions_created": 4,
  "clients_connected": 1,
  "clients_serving": 1,
  "traffic": 6326557,
  "overall_times": {
    "average": 0.37859728122037895,
    "minimal": 0.016320158,
    "maxmimal": 6.96558913,
    "median": 0.1117137805,
    "standard_deviation": 1.001460285777158,
    "percentiles": {
      "10": 0.05237131,
      "20": 0.071472272,
      "30": 0.088965026,
      "40": 0.101607119,
      "50": 0.1117137805,
      "60": 0.125672599,
      "70": 0.137716451,
      "75": 0.146478028,
      "80": 0.154273865,
      "85": 0.162262952,
      "90": 0.180582867,
      "95": 3.514414853,
      "99": 3.729193071
    }
  },
  "crawlera_times": {
    "average": 0.30196985308000035,
    "minimal": 3.1394e-05,
    "maxmimal": 3.750836014,
    "median": 0.080731409,
    "standard_deviation": 0.8410949224993787,
    "percentiles": {
      "10": 0.036049358,
      "20": 0.049492537,
      "30": 0.062905696,
      "40": 0.072465399,
      "50": 0.080731409,
      "60": 0.0885455695,
      "70": 0.09895164,
      "75": 0.103160955,
      "80": 0.110743335,
      "85": 0.118884673,
      "90": 0.129430856,
      "95": 3.494953838,
      "99": 3.694614379
    }
  },
  "traffic_times": {
    "average": 15099.18138424821,
    "minimal": 336,
    "maxmimal": 516239,
    "median": 10383,
    "standard_deviation": 31541.145341694657,
    "percentiles": {
      "10": 7511,
      "20": 8441,
      "30": 9230,
      "40": 9833,
      "50": 10383,
      "60": 10889,
      "70": 11398,
      "75": 11853,
      "80": 12327,
      "85": 13121,
      "90": 15153,
      "95": 44791,
      "99": 73846
    }
  },
  "uptime": 123
}

Here is the description of these stats:

  • requests_number - a number of requests managed by headless proxy. This includes all possible requests, not only those which were sent to Crawlera.
  • crawlera_requests - a number of requests which were sent to Crawlera. This also includes retries on session restoration etc.
  • sessions_created - how many sessions were created by headless proxy so far.
  • clients_connected - how many clients (requests) are connected to the headless proxy at this moment.
  • clients_serving - how many clients (requests) are doing requests to Crawlera now.
  • traffic - an amount of traffic sent to clients in bytes. This metric does include headers and body sizes.
  • crawlera_errors - a number of responses where X-Crawlera-Error header is set.
  • all_errors - a number of responses with errors (canceled, timeouts and crawlera_errors).
  • adblocked_requests - a number of requests which were blocked by Adblock lists. *_times describes different time series (overall response time, time spent in crawlera) etc and provide average(mean), min and max values, standard deviation and histogram of percentiles. Time series are done in window mode, tracking only latest 3000 values.

Please pay attention that usually requests_number and crawlera_requests are different. This is because headless proxy filters adblock requests and also retries to recreate sessions which imply additional Crawlera requests. So, depending on the netloc proportion of these numbers can differ.

Also, clients_serving <= clients_connected because of rate limiting. You may consider client_serving as requests which pass rate limiter.

Crawlera X-Headers

Crawlera is configured using the special headers, which usually are called x-headers (they have X- prefix in their name). You can find a full list of them in documentation.

There are 2 different ways of providing these header to the headless proxy:

  1. Use the full name
  2. Use the short version

For example, user wants to use desktop browser profile. Corresponding header is X-Crawlera-Profile. So, user can add following line to the configuration:

[xheaders]
x-crawlera-profile = "desktop"

or pass it via command line:

$ crawlera-headless-proxy ... -x x-crawlera-profile=desktop

but it makes no sense to use X-Crawlera- prefix all the time, please omit it.

[xheaders]
profile = "desktop"

or pass it via command line:

$ crawlera-headless-proxy ... -x profile=desktop

Examples

curl

$ crawlera-headless-proxy -p 3128 -a "$MYAPIKEY" -x profile=desktop
$ curl -x localhost:3128 -sLI https://scrapinghub.com

Selenium (Python)

from selenium import webdriver

CRAWLERA_HEADLESS_PROXY = "localhost:3128"

profile = webdriver.DesiredCapabilities.FIREFOX.copy()
profile["proxy"] = {
    "httpProxy": CRAWLERA_HEADLESS_PROXY,
    "ftpProxy": CRAWLERA_HEADLESS_PROXY,
    "sslProxy": CRAWLERA_HEADLESS_PROXY,
    "noProxy": None,
    "proxyType": "MANUAL",
    "class": "org.openqa.selenium.Proxy",
    "autodetect": False
}

driver = webdriver.Remote("http://localhost:4444/wd/hub", profile)
driver.get("https://scrapinghub.com")