Skip to main content

Skitai App Engine

Project description

At a Glance

Skitai is a Python WSGI/HTTP Server for UNIX (Developing is possible on win32).

And simple to run:

Install,

pip3 install -U skitai rs4 aquests sqlphile

Create and mount your app,

# myservice.py

def app (env, start_response):
  start_response ("200 OK", [("Content-Type", "text/plain")])
  return 'Hello World'

if __name__ == "__main__":
  import skitai

  skitai.mount ('/', app)
  skitai.run (address = "127.0.0.1", port = 5000)

And run.

python3 myservice.py

Your app will work for your thousands or miliions of customers.

Introduce

Skitai is WSGI server and more than that.

Skitai/Atila has purpose for providing complete online runtime environment for Python apps. You can make high concurrency with async sockect works, threading, processing and subprocessing. And you can use them with highly consistent manner.

Skitai handles requests with mainly asynchronous event loop and threading pool and support websocket within WSGI specification not with ASGI nor asyncio (It use asyncore).

Table of Contents

Installation

Requirements

Python 3.5+ PyPy3

Installation

Skitai and other core base dependent libraries is developing on single milestone, install/upgrade all please. Otherwise it is highly possible to meet some errors.

With pip

pip3 install -U skitai

With git

git clone https://gitlab.com/hansroh/skitai.git
cd skitai
pip3 install -e .

You have pla to use database engines or protocols which is supported by Skitai, you install them manually.

pip3 install protobuf # for GRPC
pip3 install redis
pip3 install pymongo
pip3 install psycopg2-binary
# if you use pypy3 psycopg2cffi is better choice,
pip3 install psycopg2cffi

Note

If you have bson related error, PyMongo has own bson module that is incompatible with pypi bson. I cannot exactly figure out this problem. If you really don’t need pypi bson, you can uninstall it:

pip3 uninstall -y bson
pip3 install -U --force pymongo

Or if you don’t have plan using pymongo, uninstall it.

pip3 uninstall -y pymongo

Enginize Your App with Skitai

Here’s a very simple WSGI app,

Basic Usage

Mount Static Directories

Your myproject/app.py,

if __name__ == "__main__":

  import skitai

  skitai.mount ('/', '/home/www')
  skitai.mount ('/uploads', '/var/www/uploads')
  skitai.mount ('/uploads/bigfiles', '/data/www/bifgiles')

  skitai.run (
    address = "127.0.0.1",
    port = 5000
  )

At command line,

python3 app.py

For checking processes,

$ ps -ef | grep skitai

ubuntu   25219     1  0 08:25 ?        00:00:00 skitai(myproject/app): master
ubuntu   25221 25219  1 08:25 ?        00:00:00 skitai(myproject/app): worker #0

Mount WSGI App

#WSGI App

def app (env, start_response):
  start_response ("200 OK", [("Content-Type", "text/plain")])
  return 'Hello World'

app.use_reloader = True
app.debug = True

if __name__ == "__main__":

  import skitai

  skitai.mount ('/', app)
  skitai.run (
    address = "127.0.0.1",
    port = 5000
  )

At now, run this code from console.

python3 app.py

You can access this WSGI app by visiting http://127.0.0.1:5000/.

If you want to allow access to your public IPs, or specify port:

skitai.mount ('/', app)
skitai.run (
  address = "0.0.0.0",
  port = 5000
)

skital.mount () spec is:

mount (mount_point, mount_object, app_name = “app”, pref = None)

  • mount_point

  • mount_object: app, app file path or module object

    skitai.mount ('/', app)
    skitai.mount ('/', 'app_v1/app.py', 'app')
    
    import wissen
    skitai.mount ('/', wissen, 'app')
    skitai.mount ('/', (wissen, 'app_v1.py'), 'app')
    

    In case module object, the module should support skitai exporting spec.

  • app_name: variable name of app

  • pref: run time app config, pref will override app.config

Mount Django App

Basically same as other apps.

Let’s assume your Django app project is ‘/mydjango’ and skitai app engine script is ‘/app.py’.

# and mount static dir used bt Django
skitai.mount ("/static", "mydjango/static")

with skitai.preference () as pref:
  pref.use_reloader = True
  pref.debug = True
  # finally mount django wsgi.py and project root path to append sys.path by path param.
  skitai.mount (
    "/",
    "mydjango/mydjango/wsgi.py",
    "application",
    pref
  )

Note that if app is smae location with django manage.py, you need not path param.

Also note that if you set pref.use_reloader = True, it is possible to replace Django development server (manage,py runserver), But it will work on posix only, because Skitai reloads Django app by restarting worker process, Win32 version doesn’t support.

Logging and Console Displaying For Developing/Debugging

If you do not specify log file path, all logs will be displayed in console, bu specifed all logs will be written into file.

First of all, you should create log directory,

sudo mkdir /var/log/skitai
sudo chown ubuntu:ubuntu

Your request log file willl be placed to: /var/log/skitai/ubuntu/<script path hash>/request.log.

skitai.mount ('/', app)
skitai.enalbe_file_logging ()
skitai.run (
  address = "0.0.0.0",
  port = 5000
)

If you also want to view logs through console for spot developing, you run app.py without option.

python3 app.py

Run with Process Name

If you give ‘name’, process name will be changed.

skitai.mount ('/', app)
skitai.run (name = "myapp")

Your skitai process will be shown as:

ubuntu    9815     1  0 16:04 ?        00:00:00 skitai/myapp: master
ubuntu    9816  9815  0 16:04 ?        00:00:03 skitai/myapp: worker #0

Getting Command Line Options and Arguments

Skitai use short options -d, and long long options starts with “—”, then you SHOULD NOT use these options. Also Skitai use satrt, restart, status, stop in args. then these arguments are removed automatically.

skitai.add_option ('-D', '--dist', 'distribute mode, disable NodeJS proxing')
skitai.add_option (None, '--db=DB_NAME', 'use specified database')
...
active_db = skitai.options.get ('--db', 'testdb')

And if you use ‘–help’, you can see like this:

Usage: apiserve/serve.py [OPTION]... [COMMAND]...
COMMAND can be one of [status|start|stop|restart]

Mandatory arguments to long options are mandatory for short options too.
  -d                      start as daemon, equivalant with using `stop` command
      ---profile          log for performance profiling
      ---gc               enable manual GC
      ---memtrack         show memory status
      --production        run as production mode
      --smtpda            run SMTPDA if not started
      --port=PORT_NUMBER  change port
  -D, --dist              distribute mode, disable NodeJS proxing
      --db=DB_NAME        use specified database

Note that you cannot use below ones:

-d
--help
--smtpda
--port=
all triple hypened options

they are reserved for Skitai.

Run with Threads Pool

Skitai run defaultly multi-threading mode and number of threads are 4. If you want to change number of threads for handling WSGI app:

skitai.mount ('/', app)
skitai.run (
  threads = 8
)

Run with Non-Thread Pool

If you want to run Skitai with entirely main thread only,

skitai.mount ('/', app)
skitai.run (threads = 0)

This features is limited by your WSGI container. If you use Atila container, you can run with single threading mode by using Atila’s async streaming response method. But you don’t and if you have plan to use Skitai ‘was’ requests services, you can’t single threading mode and you SHOULD run with multi-threading mode.

Run with Multiple Workers

Available on posix only

Skitai can run with multiple workers(processes) internally using fork for socket sharing.

skitai.mount ('/', app)
skitai.run (
  port = 5000,
  workers = 4,
  threads = 8
)

Skitai processes are,

$ ps -ef | grep skitai

ubuntu   25219     1    0 08:25 ?        00:00:00 skitai(myproject/app): master
ubuntu   25221 25219  1 08:25 ?        00:00:00 skitai(myproject/app): worker #0
ubuntu   25222 25219  1 08:25 ?        00:00:00 skitai(myproject/app): worker #1
ubuntu   25223 25219  1 08:25 ?        00:00:00 skitai(myproject/app): worker #2
ubuntu   25224 25219  1 08:25 ?        00:00:00 skitai(myproject/app): worker #3

Set Critical Point to Worker Processes

New In Version 0.26.15.2, Available only on posix

You can set parameters for restarting overloaded workers,

skitai.set_worker_critical_point (cpu_percent = 90.0, continuous = 3, interval = 20)

This means if a worker’s CPU usage is 90% for 20 seconds continuously 3 times, Skitai try to kill this worker and start a new worker.

If you do not want to use this, you just do not call set_worker_critical_point () or set interval to zero (0).

But I strongly recommend use this setting especially if you running Sktiai on single CPU processor machine or like AWS t1.x limited computing instances.

Also this is for minimum protection against Skitai’s unexpected bugs.

Mount Multiple WSGI Apps And Static Directories

Skitai can mount multiple WSGI apps.

Independent Apps and Various WSGI Containers

Here’s three WSGI app samples:

# WSGI App

def app (env, start_response):
  start_response ("200 OK", [("Content-Type", "text/plain")])
  return ['Hello World']

app.use_reloader = True
app.debug = True


# OR Flask App
from flask import Flask
app = Flask(__name__)

app.use_reloader = True
app.debug = True

@app.route("/")
def index ():
  return "Hello World"


# OR Atila App
from atila import Atila
app = Atila (__name__)

app.use_reloader = True
app.debug = True

@app.route('/')
def index (was):
  return "Hello World"

Then place this code at bottom of above WSGI app.

if __name__ == "__main__":

  import skitai

  skitai.mount ('/', __file__, 'app')
  skitai.mount ('/', 'static')
  skitai.run ()

Service Versioning

These feature can be used for managing versions.

Let’s assume initial version of app file is app_v1.py.

app = Atila (__name__)

@app.route('/')
def index (was):
  return "Hello World Ver.1"

And in same directory 2nd version of app file is app_v2.py.

app = Atila (__name__)

@app.route('/')
def index (was):
  return "Hello World Ver.2"

Now service.py is like this:

import skitai

skitai.mount ('/', 'static')
skitai.mount ('/v1', 'app_v1')
skitai.mount ('/v2', 'app_v2')
skitai.run ()

Then run with:

python service.py

You can access ver.1 by http://127.0.0.1:5009/v1/ and vwe.2 by http://127.0.0.1:5009/v2/.

Note: Above 3 files is in the same directory and then both share templates directory. If you intend to seperate from app_v1 and app_v2, you should seperate app with directory like this:

service.py

app_v1/app.py
app_v1/templates
app_v1/static

app_v2/app.py
app_v2/templates
app_v2/static

and your service.py:

import skitai

skitai.mount ('/v1', 'app_v1/static'),
skitai.mount ('/v1', 'app_v1/app'),
skitai.mount ('/v2', 'app_v2/static'),
skitai.mount ('/v2', 'app_v2/app')
skitai.run ()

Mounting With Virtual Host

if __name__ == "__main__":

  import skitai
  skitai.mount ('/', 'site1.py', host = 'www.site1.com')
  skitai.mount ('/', 'site2.py', host = 'www.site2.com')
  skitai.run ()

Setting POST Body Size Limitation

For setting 8 Gbytes limitation for POST body size,

import skitai

pref = skitai.pref ()
pref.max_client_body_size = 2 << 32

If you want to set more detaily,

import skitai

with skitai.preference () as pref:
  pref.config.max_post_body_size = 2 << 32
  pref.config.max_multipart_body_size = 2 << 32
  pref.config.max_upload_file_size = 2 << 32

Setting Timeout

Keep alive timeout means seconds gap of each requests. For setting HTTP connection keep alive timeout,

skitai.set_keep_alive (2) # default = 30
skitai.mount ('/', app)
skitai.run ()

If you intend to use skitai as backend application server behind reverse proxy server like Nginx, it is recommended over 300.

Request timeout means seconds gap of data packet recv/sending events,

skitai.set_request_timeout (10) # default = 30
skitai.mount ('/', app)
skitai.run ()

Note that under massive traffic situation, meaning of keep alive timeout become as same as request timeout beacuse a clients requests are delayed by network/HW capability unintensionally.

Anyway, these timeout values are higher, lower response fail rate and longger response time. But if response time is over 10 seconds, you might consider loadbalancing things. Skitai’s default value 30 seconds is for lower failing rate under extreme situation.

New in version 0.26.15

You can set connection timeout for your backends. Basue of Skitai’s ondemend polling feature, it is hard to know disconnected by server side, then Skitai will forcley reconnect if over backend_keep_alive after last interaction. Make sure your backends keep_alive setting value is matched with this value.

skitai.set_backend_keep_alive (1200) # default is 10
skitai.mount ('/', app)
skitai.run ()

Enabling HTTP/HTTPS Proxy

Make sure you really need proxy.

skitai.enable_proxy ()

# tunnel value will be applied to HTTPS proxy
skitai.set_proxy_keep_alive (channel = 60, tunnel = 600)

skitai.run ()

Run as Daemon

Available on posix only

For making a daemon,

python3 app.py start (or -d)

For stopping daemon,

python3 app.py stop (or -s)

Or for restarting daemon,

python3 app.py restart (or -r)

For automatic starting on system start, add a line to /etc/rc.local file like this:

su - ubuntu -c "/usr/bin/python3 /home/ubuntu/app.py -d"

exit 0

Adding Backend Server Alias

Backend server can be defined like this: (alias_type, servers, role = “”, source = “”, ssl = False).

alias_types can be one of these:

  • All of HTTP based services like web, RPC, RESTful API
    • PROTO_HTTP
    • PROTO_HTTPS
  • Websocket
    • PROTO_WS: websocket
    • PROTO_WSS: SSL websocket
  • Database Engines
    • DB_PGSQL
    • DB_SQLITE3
    • DB_REDIS
    • DB_MONGODB
    • DJANGO: mount django database engine of settings.py if database engine is PostgreSQL or SQLite3
  • server: single or server list, server form is [ username : password @ server_address : server_port / database_name weight ]. if your username or password contains “@” characters, you should replace to ‘%40’
  • role (optional): it is valid only when cluster_type is http or https for controlling API access
  • source (optional): comma seperated ipv4/mask
  • ssl (optional): use SSL connection or not, PROTO_HTTPS and PROTO_WSS use SSL defaultly

Some examples,

skitai.alias (
  '@members',
  skitai.PROTO_HTTP,
  [ "username:password@members.example.com:5001" ],
  role = 'admin',
  source = '172.30.1.0/24,192.168.1/24'
)

skitai.alias (
  '@mypostgres',
  skitai.DB_POSTGRESQL,
  [
    "postgres:1234@172.30.0.1:5432/test 20",
    "postgres:1234@172.30.0.2:5432/test 10"
  ]
)

skitai.alias (
  '@mysqlite3',
  skitai.DB_SQLITE3,
  [
    "/var/tmp/db1",
    "/var/tmp/db2"
  ]
)

Run as HTTPS Server

You can get certification from Let’s Encrypt or where you want.

First of all, you make simple script for certbot challenge.

mkdir myservice
mkdir myservice/static
cd myservice

And write serve.py,

#! /usr/bin/env python3

import skitai

skitai.mount ("/", "./static")
skitai.run (port = 80, name = "my-service")

For using port 80, you need root permision,

chmod +x serve.py
sudo ./serve.py

Now on another console,

cd myservice
sudo apt install certbot
sudo certbot certonly --webroot \
  -w ./static \
  -d mydomain.com -d www.mydomain.com

Apply change to serve.py,

#! /usr/bin/env python3

skitai.enable_ssl (
  '/etc/letsencrypt/live/mydomain.com/fullchain.pem',
  '/etc/letsencrypt/live/mydomain.com/privkey.pem'
)
# forward http -> https, www.mydomain.com -> mydomain.com
skitai.enable_forward (80, 443, 'mydomain.com')

skitai.mount ("/", "./static")
skitai.run (port = 443, name = "my-service")

Becasue of you’re using port 80, 443 you need root privileges.

sudo ./serve.py

After binding port 80 and 443 and reading certifications, Skitai will drop root privileges and back to sudo user privileges.

FYI, for automating renew certification, start Skitai as daemon and add cron job,

sudo ./serve.py -d

And add cron job with renew-hook,

sudo crontab -e

# add this line for trying renew twice a day, restart Skitai with user ubuntu privileges if renewed
1 4,16 * * * /usr/bin/certbot renew --renew-hook="su - ubuntu -c 'sudo /PATH/myservice/serve.py restart'"

Note: If you just run with ‘/PATH/myservice/serve.py restart’, server will run with nobody account privileges

Self-Signed certification

To generate self-signed certification file:

; Create the Server Key and Certificate Signing Request
sudo openssl genrsa -des3 -out server.key 2048
sudo openssl req -new -key server.key -out server.csr

; Remove the Passphrase If you need
sudo cp server.key server.key.org
sudo openssl rsa -in server.key.org -out server.key

; Sign your SSL Certificate
sudo openssl x509 -req -days 365 -in server.csr -signkey server.key -out server.crt

Then,

skitai.mount ('/', app)
skitai.enable_ssl ('server.crt', 'server.key', 'your pass phrase')
skitai.run ()

About Mount Point & App Routing

If app is mounted to ‘/flaskapp’,

from flask import Flask
app = Flask (__name__)

@app.route ("/hello")
def hello ():
  return "Hello"

Above /hello can called, http://127.0.0.1:5000/flaskapp/hello

Also app should can handle mount point. In case Flask, it seems ‘url_for’ generate url by joining with env[“SCRIPT_NAME”] and route point, so it’s not problem. Atila can handle obiously. But I don’t know other WSGI containers will work properly.

SMTP Delivery Agent

New in version 0.26

e-Mail sending service is executed seperated system process not threading. Every e-mail is temporary save to file system, e-Mail delivery process check new mail and will send. So there’s possibly some delay time.

You can send e-Mail in your app like this:

from skitai import was

# email delivery service
e = was.email (subject, snd, rcpt)
e.set_smtp ("127.0.0.1:465", "username", "password", ssl = True)
e.add_content ("Hello World<div><img src='cid:ID_A'></div>", "text/html")
e.add_attachment (r"001.png", cid="ID_A")
e.send ()

You can set default SMTP server and you can skip e.set_smtp (…) part.

skitai.set_smtp ("127.0.0.1:465", "username", "password", ssl = True)

For enabling this features,

serve.py --smtpda

All e-mails are saved into /var/temp/skitai/smtpda.

This service will run as system-wide daemon service, and will be not stopped even if app engine is stopped. For stopping it,

skitai smtpda status
skitai smtpda stop

Asccessing File Resources On Startup

Skitai’s working directory is where the script call skitai.run (). Even you run skitai at root directory,

/app/example/app.py -d

Skitai will change working directory to /app/example on startup.

So your file resources exist within skitai run script, you can access them by relative path,

monitor = skital.abspath ('package', 'monitor.py')

Also, you need absolute path on script,

skitai.getswd () # get skitai working directory

Enable Cache File System

If you make massive HTTP requests, you can cache contents by HTTP headers - Cache-Control and Expires. these configures will affect to ‘was’ request services, proxy and reverse proxy.

skitai.enable_cachefs (memmax = 10000000, diskmax = 100000000, path = '/var/tmp/skitai/cache')
skitai.mount ('/', app)
skitai.run ()

Default values are:

  • memmax: 0
  • diskmax: 0
  • path: None

Configure Max Age For Static Files

You can set max-age for static files’ respone header like,

Cache-Control: max-age=300
Expires: Sun, 06 Nov 2017 08:49:37 GMT

If max-age is only set to “/”, applied to all files. But you can specify it to any sub directories.

skitai.mount ('/', 'static')
skitai.set_max_age ("/", 300)
skitai.set_max_age ('/js', 0)
skitai.set_max_age ('/images', 3600)
skitai.run ()

Testing Mounted App

New in version 0.27

For mounted app testing fully network environment,

import skitai

def test_myapp ():
  with skitai.test_client ("./app.py", 6000) as cli:
    resp = cli.get ("/")
    assert "something" in resp.text

    # api call
    stub = cli.api ()
    resp = stub.apis.pets (45).get ()
    assert resp.data ["id"] == 45

Now run pytest.

This test client will start Skitai server on port 6000 with app. app.py shoud have skitai.run ().

Note: Port that skitai.run (port = 5000) will be ignored, app.py will be launched with port 6000 that specified by skitai.test_client for avoiding exist app service.

If your have so many tests, define cli at your conftest.py

import pytest
import skitai

@pytest.fixture (scope = "session")
def cli ():
  c = skitai.test_client ("./app.py", 6000)
  yield c
  c.stop ()

And edit your test script:

import skitai

def test_myapp (cli):
  resp = cli.get ("/")
  assert "something" in resp.text

  # api call
  stub = cli.api ()
  resp = stub.apis.pets (45).get ()
  assert resp.data ["id"] == 45

If you run test server at another console window for watching server error messages, give dry = True parameter.

@pytest.fixture (scope = "session")
def cli ():
  c = skitai.test_client ("./app.py", 5000, dry = True)
  yield c
  c.stop ()

This test client will not start Skitai server but access to port 5000 so you start server manually at another console,

python3 app.py

Inter-Processes State Sharing

New in skitai version 0.26.18

Skitai can run with multiple processes (a.k workers), It is possible matters synchronizing state between workers.

Already mentioned ‘skitai.register_states ()’ can be used for allocating shared memory for inter-process named state.

import skitai

skitai.register_states ("current-user", ...)

Then one process update object by setgs (name, value), the others can be access it by getgs (name).

Note that value type is shoul be integer.

@app.before_request
def before_request (was):
  was.setgs ("current-user", was.getgs ("current-user") + 1)

@app.teardown_request
def teardown_request (was):
  was.setgs ("current-user", was.getgs ("current-user") - 1)

For connecting to event bus,

skitai.register_states ("cluster.num-nodes", "region.somethig", ...)
...

skitai.run ()

Then you can use these,

@app.route ("/nodes", method = ["POST", "DELETE"])
def nodes (was, **nodinfos):
  ...
  was.setgs ("cluster.num-nodes", was.getgs ("cluster.num-nodes") + 1, **nodeinfos)

As a result,

  • cluster.num-nodes state value has been increased
  • “cluster.num-nodes” and **nodeinfos are broadcated to mounted all Atila apps.

A app has interest for this,

@app.on_broadcast ("cluster.num-nodes")
def num_nodes_changed (num_nodes, **nodeinfos):
  ...

But this broadcasting is just within current workers.

All workers has interested in this event, You may add watching routine at app.maintain.

app.config.maintain_interval = 60
app.store ["num_nodes"] = 0

@app.maintain
def maintain_num_nodes (was, now):
  ..
  num_nodes = was.getgs ("cluster.num-nodes")
  if app.store ["num_nodes"] != num_nodes:
    app.store ["num_nodes"] = num_nodes
    app.broadcast ("cluster:num_nodes")

also was.setlu () and was.getlu () is very similar usage which related to track resource updating. And it will be explained Corequest: Caching Result chapter.

Request Logging

Turn Request Logging Off For Specific Path

For turn off request log for specific path,

# turned off starting with
skitai.log_off ('/static/')

# turned off ending with
skitai.log_off ('*.css')

# you can multiple args
skitai.log_off ('*.css', '/static/images/', '/static/js/')

Log Format

Blank seperated items of log line are,

  • log date
  • log time
  • client ip or proxy ip
  • request host: default ‘-‘ if not available
  • request methods
  • request uri
  • request version
  • request body size
  • reply code
  • reply body size
  • global transaction ID: for backtracing request if multiple backends related
  • local transaction ID: for backtracing request if multiple backends related
  • username when HTTP auth: default ‘-‘, wrapped by double quotations if value available
  • bearer token when HTTP bearer auth
  • referer: default ‘-‘, wrapped by double quotations if value available
  • user agent: default ‘-‘, wrapped by double quotations if value available
  • x-forwared-for, real client ip before through proxy
  • Skitai engine’s worker ID like M(Master), W0, W1 (Worker #0, #1,… Posix only)
  • number of active connections when logged, these connections include not only clients but your backend/upstream servers
  • duration ms for request handling
  • duration ms for transfering response data

Skitai with Nginx

If your service is relatvely simple and you may think using Nginx is overkill, it is well enough with Skitai.

Below Nginx features (SSL, HTTP2, forwarding, proxy passing, static file service etc) can be implemetented with Skitai alone.

But if you need some kind of gateway server which control multiple upstreams and Skitai app engine instances, I strongly recommend to use Nginx.

Here’s some helpful sample works with Nginx.

# upstreams with connection keep alive
upstream backend {
  server 127.0.0.1:5000;
  keepalive 100;
}

server {
    listen 80;
    listen [::]:80;
    server_name www.oh-my-jeans.com;
    return 301 https://oh-my-jeans.com$request_uri;
}

server {
    listen 443;
    listen [::]:443;
    server_name www.oh-my-jeans.com;
    return 301 https://oh-my-jeans.com$request_uri;
}

server {
  listen 443 ssl http2;
  listen [::]:443 ssl http2;
  server_name oh-my-jeans.com;

  ssl_certificate /etc/letsencrypt/live/www.oh-my-jeans.com/fullchain.pem;
  ssl_certificate_key /etc/letsencrypt/live/www.oh-my-jeans.com/privkey.pem;
  ssl_session_timeout 5m;
  ssl_protocols TLSv1 TLSv1.1 TLSv1.2;
  ssl_ciphers HIGH:!aNULL:!MD5;
  ssl_prefer_server_ciphers on;

  keepalive_timeout 30s;
  proxy_http_version 1.1;
  proxy_set_header X-NginX-Proxy true;
  proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
  add_header X-Backend "skitai app engine";
  proxy_set_header Host $http_host;

  location / {
    proxy_pass http://backend;
    client_max_body_size 2g;
  }

  location /websocket {
    proxy_pass http://backend;
    proxy_set_header Upgrade $http_upgrade;
    proxy_set_header Connection "Upgrade";
    proxy_read_timeout 86400;
  }

  location /assets/ {
    alias /home/ubuntu/www/statics/assets;
    expires 300;
  }
}

Run As API Gateway Server (Experimental)

Using Skitai’s reverse proxy feature, it can be used as API Gateway Server. All backend API servers can be mounted at gateway server with client authentification and transaction ID logging feature.

def handle_claim (request_handler, request):
  claim = request.claim
  expires = claim.get ("expires", 0)
  if expires and expires < time.time ():
    return request_handler.continue_request (request)
  request_handler.continue_request (request, claim.get ("user"), claim.get ("roles"))

@app.before_mount
def before_mount (wac):
  wac.handler.set_auth_handler (handle_claim)

@app.route ("/")
def index (was):
  return "<h1>Skitai App Engine: API Gateway</h1>"

if __name__ == "__main__":
  import skitai

  skitai.alias (
    '@members', 'https', "members.example.com",
    role = 'admin', source = '172.30.1.0/24,192.168.1/24'
  )
  skitai.alias (
    '@photos', skitai.DB_SQLITE3, ["/var/tmp/db1", "/var/tmp/db2"]
  )
  skitai.mount ('/', app)
  skitai.mount ('/members', '@members')
  skitai.mount ('/photos', '@photos')
  skitai.enable_gateway (True, "8fa06210-e109-11e6-934f-001b216d6e71")
  skitai.run ()

Gateway use only bearer tokens like OAuth2 and JWT(Json Web Token) for authorization. And token issuance is at your own hands. But JWT creation,

from rs4 import jwt

secret_key = b"8fa06210-e109-11e6-934f-001b216d6e71"
token = jwt.gen_token (secret_key, {'user': 'Hans Roh', 'roles': ['user']}, "HS256")

Also Skitai create API Transaction ID for each API call, and this will be explained in Skitai ‘was’ Service chapter.

Run as Win32 Service (Deprecated)

Available on win32 only, New in version 0.26.7

from atila import Atila
from rs4.psutil.win32service import ServiceFramework

class ServiceConfig (ServiceFramework):
  _svc_name_ = "SAE_EXAMPLE"
  _svc_display_name_ = "Skitai Example Service"
  _svc_app_ = __file__
  _svc_python_ = r"c:\python34\python.exe"

app = Atila (__name__)

if __name__ == "__main__":
  skitai.mount ('/', app)
  skitai.set_service (ServiceConfig)
  skitai.run ()

Then at command line,

app.py install # for installing windows service
app.py start
app.py stop
app.py update # when service class is updated
app.py remove # removing from windwos service

Self-Descriptive App

Skitai’s one of philasophy is self-descriptive app. This means that you once make your app, this app can be run without any configuration or config files (at least, if you need own your resources/log files directoring policy). Your app contains all configurations for not only its own app but also Skitai. As a result, you can just install Skitai with pip, and run your app.py immediately.

pip3 install skitai
# if your app has dependencies
pip3 install -Ur requirements.txt
python3 app.py

Export API From Your Module Through Skitai

If your module need export APIs or web pages, you can include app in your module for Skitai App Engine.

Let’s assume your package name is ‘unsub’.

Your app should be located at unsub/export/skitai/__export__.py

Then users uses your module can mount on skitai by like this,

import unsub

with skitai.preference () as pref:
  pref.config.urlfile = skitai.abspath ('resources', 'urllist.txt')
  skitai.mount ("/v1", unsub, "app", pref)
skitai.run ()

If you want to specify filename like app_v1.py for version management,

skitai.mount ("/v1", (unsub, "app_v1.py"), "app", pref)

If your app need bootstraping or capsulizing complicated initialize process from simple user settings, write code to unsub/export/skitai/__init__.py.

 import skitai

 def bootstrap (pref):
   with open (pref.config.urlfile, "r") as f:
     urllist = []
     while 1:
       line = f.readline ().strip ()
       if not line: break
       urllist.append (line.split ("  ", 4))
     pref.config.urllist = urllist

*Important Note:* You should add zip_safe = False flag in your setup.py
because Skitai could access your __export__ script and its sub modules.
setup (
  name = "mymodule",
  ...
  zip_safe = False
)

Extending/Customizing Services

New in version 0.28.15

If you want to customize/extend services, create ‘extends’ directory and mount it to pref.

# extends/apis.py
def __mount__ (app):
  @app.permission_check_handler
  def permission_check_handler (was, perms):
    ...

  @app.route ("")
  def apis_index (was):
    return 'APIS'
# serve.py
import unsub
from extends import apis

with skitai.preference () as pref:
  pref.mount ('/apis', apis)
  pref.config.urlfile = skitai.abspath ('resources', 'urllist.txt')
  skitai.mount ("/v1", unsub, "app", pref)
skitai.run ()

You can access it by /v1/apis.

If you want to mount another services from unpathed, you can specify new path.

Note: This will change sys.path order. You SHOULD do these on your last mount stage.

# serve.py
import unsub

with skitai.preference (path = '../my_service') as pref:
  from services import apis
  pref.mount ('/apis', apis)
  skitai.mount ("/v1", unsub, "app", pref)
skitai.run ()

Examples

Here’re some implementations I made.

HTTP/2.0

New in version 0.16

Skiai supports HTPT2 both ‘h2’ protocl over encrypted TLS and ‘h2c’ for clear text (But now Sep 2016, there is no browser supporting h2c protocol).

Push Promise

Basically you have nothing to do for HTTP2. Client’s browser will handle it except HTTP2 server push.

For using it, you just call was.push (uri) before return response data. It will work only client browser support HTTP2 server push, otherwise will be ignored.

from skitai import was

@app.route ("/promise")
def promise ():

  was.response.push_promise ('/images/A.png')
  was.response.push_promise ('/images/B.png')

  return was.response (
    "200 OK",
    (
      'Promise Sent<br><br>'
      '<img src="/images/A.png">'
      '<img src="/images/B.png">'
    )
  )

HTTP/3.0 (Experimental)

New in version 0.33

Python>=3.6 is required

Testing HTPP3 and QUIC

Skitai has been launched experimetnal HTTP/3 on QUIC with aioquic.

WARNING: DO NOT use this for your production services. You must aware that it is experimetal and unstable yet.

Configuring and Lauching

HTTP3 can be run with https, you need a certification for it.

skitai.enable_ssl (
  '/etc/letsencrypt/live/mydomain.com/fullchain.pem',
  '/etc/letsencrypt/live/mydomain.com/privkey.pem'
)
skitai.mount ("/", "./static")
skitai.run (name = "my-service", port = 443, quic = 4433)

And to avond port permission, you make port forwarding from UDP 443 to 4433.

sudo iptables -A PREROUTING -t nat -i eth0 -p udp --dport 443 -j REDIRECT --to-port 4433

Also you open firewall UDP 443 not 4433.

Note: This MUST BE DONE, becasue:

  • If QUIC port is not 443, it may be ignored by clients’ browsers
  • Secondary, Skitai will bind UDP port 443 per every clients for performance reason, and it need root privileges for running and it is not good idea. Also skitai doesn’t allow keeping root privileges after started

This make both HTTP/2 and HTTP/3 services on TCP/UDP port 443.

And you need sudo for binding TCP port 443 and reading certification on starting.

sudo python3 ./serve.py

After started, Skitai will drop root privileges and fall back to current user’s.

Push Promise

Pushing promise is just same as HTTP/2.0.

Testing HTTP/3 and QUIC

You can test HTTP/3.0 with Chrome Canary build.

You have to run Chrome Canary with command line options,

chrome.exe --enable-quic --quic-version=h3-23

At your browser’s developer window, you can see protocol as http/2+quic/99 during you loading your web page and files.

HTML5 Websocket

New in version 0.11

The HTML5 WebSockets specification defines an API that enables web pages to use the WebSockets protocol for two-way communication with a remote host.

Skitai can be HTML5 websocket server and any WSGI containers can use it.

But I’m not sure my implemetation is right way, so it is experimental and could be changable.

Using Websocket

Use skitai.websocket decorator.

For example with Flask app,

import request

@app.route ("/echo3")
@skitai.websocket (60) # timeout
def echo3 ():
  ws = request.environ ["websocket"]
  while 1:
    message = yield
    if not message:
      return #strop iterating
    yield "ECHO:" + message

I you want to send multiple messages,

yield ['OK', 'Task 1 started', 'Check later, please']
# OR
yield output_iterator ()

Proxying With Atila

It follows WSGI specification as possible as can:

def start_response (environ, start_response):
  ...

Basically, Skitai calls this method on message arriving repeatly. So it is quite ineeficient. If your WSGI framework give a websocket handler object, it will have better performance but it is hard to expect.

Another option is that Sktai provide full usage spec with routing, but I think it is not pretty.

So you can use Atila for websocket service (as websocket proxy) beside your main app. and mount both app on Skitai.

With Atila app, you can use websocket more efficiently, and various options.

@app.route ("/websocket")
@app.websocket (skitai.WS_CHANNEL | skitai.WS_SESSION, 60)
def websocket (was):
  while 1:
    message = yield
    if not message:
      return #strop iterating
    yield "ECHO:" + message

For more about this see Atila Websocket.

WWW-Authenticate

Some browsers do not support WWW-Authenticate on websocket like Safari, then Skitai currently disables WWW-Authenticate for websocket, so you should be careful for requiring secured messages.

Client Side

First of all, see conceptual client side java script for websocket using Vuejs.

<div id="app">
  <ul>
    <li v-for="log in logs" v-html="log.text"></li>
  </ul>
  <input type="Text" v-model="msg" @keyup.enter="push (msg); msg='';">
</div>

<script>
vapp = new Vue({
  el: "#app",
  data: {
    ws_uri: "ws://www.yourserver.com/websocket",
    websocket: null,
    out_buffer: [],
    logs: [],
    msg = '',
  },

  methods: {

    push: function (msg) {
      if (!msg) {
        return
      }
      this.out_buffer.push (msg)
      if (this.websocket == null) {
        this.connect ()
      } else {
        this.send ()
      }
    },

    handle_read: function (evt)  {
      this.log_info(evt.data)
    },

    log_info: function (msg) {
      if (this.logs.length == 10000) {
        this.logs.shift ()
      }
      this.logs.push ({text: msg})
    },

    connect: function () {
      this.log_info ("connecting to " + this.ws_uri)
      this.websocket = new WebSocket(this.ws_uri)
      this.websocket.onopen = this.handle_connect
      this.websocket.onmessage = this.handle_read
      this.websocket.onclose = this.handle_close
      this.websocket.onerror = this.handle_error
    },

    send: function () {
      for (var i = 0; i < this.out_buffer.length; i++ ) {
        this.handle_write (this.out_buffer.shift ())
      }
    },

    handle_write: function (msg) {
      this.log_info ("SEND: " + msg)
      this.websocket.send (msg)
    },

    handle_connect: function () {
      this.log_info ("connected")
      this.send ()
    },

    handle_close: function (evt)  {
      this.websocket.close()
      this.websocket = null
      this.log_info("DISCONNECTED")
    },

    handle_error: function (evt)  {
      this.log_info('ERROR: ' + evt.data)
    },

  },

  mounted: function () {
    this.push ('Hello!')
  },

})

</script>

Send Messages Through Websocket Directly

It needn’t return message, but you can send directly multiple messages through was.websocket,

@app.route ("/websocket/echo")
@was.websocket ("message", 60)
def echo ():
  message = request.args.get ("message")
  request.environ ["websocket"].send ("You said," + message)
  request.environ ["websocket"].send ("I said acknowledge")

Corequest

Skitai handle request connection with asynchronously, also has threads and porcess ass workers. So it works fine with synchronous apps and libraries. You can use standard database client libraries or requests module for API calls.

But Skitai’s main event loop (using asyncore.loop) can be used for not only client’s requests else request to another servers(API, Database engine…) asynchronously.

I think if I don’t use this capabitities, it would be wasting resources. Then, Skitai provide asynchronous request methods for these operations.

Corequest is similar with Python coroutine object, but is is not compatable at all.

  • It is automatically started at creation, no need to call run ()
  • All events are controlled by Skitai main event loop, not by asyncio
  • It is eventually synchronous within current thread. It is desinged for working with multi-threading environment and synchronous code base so it has no differences with synchronous code base, just if you have to consider the most efficient point to call for waiting results
  • It is not a framework nor a library. It is a Skitai native object has specified purpose and usage

Skitai provides some services related with corequests:

  • Concurrent requests (like asyncio or gevent) to your API/Backend and Database engine servers
  • Connection pooling
  • Result caching

These features are just optional, but these might help increase availability of your servers.

For using ‘corequest’, you need to import ‘was’:

from skitai import was

@app.route ("/")
def hello ():
  was.get ("http://...")

Basic

Task

Single corequest object.

API Call
  • was.get ()
  • was.post ()
  • was.put ()
  • was.patch ()
  • was.delete ()
  • was.upload ()

Task will be created by just calling these methods.

task = was.get ('@myapi/v1/some-resources/100')
RPC Call
  • was.xmlrpc ()
  • was.grpc ()
  • was.jsonrpc ()

Task will be created like this,

with was.xmlrpc ('@myrpc/rpc2') as stub:
  task = stub.some_method (arg1, arg2)
Database Call
  • was.db (): PostgreSQL, SQLite3, MongoDB and Redis calls
  • was.transaction (): for RDBMS (PostgreSQL and SQLite3)

Task will be created like this,

# PostgreSQL and SQLite3
with was.db('@mydb') as db:
  task = db.select ('my_table').execute ()

# Redis or MongoDB
with was.db('@mynosql') as db:
  task = db.find ({'city': 'New York'})
Thread/Process Call
  • was.Thread ()
  • was.Process ()
  • was.Subprocess ()

Task will be created like this,

task = was.Thread (my_func, arg1, arg2)
Methods of Task

Task has below core methods:

  • dispatch (timeout)
  • fetch (timeout)
  • one (timeout): should be single lengthed object
  • commit (timeout)
  • returning (data)

Tasks

It is bundle of Tasks.

You can make it by wrapping.

tasks = was.Tasks ([task1, task2])
result1, result2 = tasks.fetch ()

And it has also same methods as Task. But it can be accessed by slicing or indexing for easy handling.

Mask

It is fake of Task(s).

You can make it by wrapping was.Mask (data) if you want to use consistant methods as Task.

task = was.Mask (1)
result = task.fetch () # 1

tasks = was.Mask ([1, 2])
result1, result2 = tasks.fetch () # 1, 2

Long Running Task(s)

corequests is natively a kind of backgound jobs. So you can
create these tasks and return yotur response - usally 202 Accepted.

More explicit way, creating tasks and immediately return 202 response.

return was.post ('@myapi/v1/some-resources').returning (Response ('202 Accepted'))

return was.Thread (func, arg).returning (Response ('202 Accepted'))

Future(s)

Available on Atila only

On Atila, you can hook the callback function with corequest objects.

  • Task can be transformed into Future
  • Tasks can be transformed into Futures

Future/Futures object can be returnable and it has the benefit when your jobs are IO bound and long running time (but reasonably close enough to real time). It returns current thread qucikly, lazy respond when job is done.

Calling API

@app.route (...)
def request (was):
  req = was.get (url)
  resp = req.dispatch (timeout = 3)
  return resp.data

In fact, single request is just like synchronous task at least current thread.

@app.route (...)
def request (was):
  req1 = was.get (url)
  req2 = was.post (url, {"user": "Hans Roh", "comment": "Hello"})
  respones1 = req1.dispatch (timeout = 3)
  response2 = req2.dispatch (timeout = 3)
  return [respones1.data, respones2.data]

Note that req1 and req2 will be executed concurrently.

dispath (timeout = [sec], cache = [sec]) returns response object.

req = was.get (url)
rsponse = req.dispath (5) # timoute
response.status # skitai.STA_NORMAL
response.status_code # 200
response.reason # OK
response.get_header ("Content-Type") # application/json
response.data # {"result": "ok"}

response.status is one of belows:

  • STA_UNSENT
  • STA_REQFAIL
  • STA_TIMEOUT
  • STA_NETERR
  • STA_NORMAL

Note that STA_NORMAL just mean all requesting precess is normally completed, NOT response is. Then you SHOULD check before handle result data.

dispath_or_throw () will raise exception immediatly if status != STA_NORMAL or status_code >= 300.

rsponse = req.dispath_or_throw (5) # timoute

If you want more short hand to result data,

result = req.fetch (5) # timoute and {"result": "ok"}

result = fetch (5) is equivalant with,

rsponse = req.dispath_or_throw (5) # timoute
response = response.data

All supoorted request methods are:

HTTP/API related methods are,

  • was.get ()
  • was.delete ()
  • was.post ()
  • was.put ()
  • was.patch ()
  • was.upload ()
  • was.options ()

Above request type is configured to json. This mean request content type and response accept type is all ‘application/json’.

If you want to change default value, use headers paramter for each request

data = {"Title": "...", "Content": "..."}
headers = [
  ("Content-Type", "application/x-www-form-urlencoded"),
  ("Accept", "text/xml")
]
req = was.post ("@delune/documents", data, headers = headers)

Tasks

Tasks is pack of corequests. It can handle multiple corequests as single one.

@app.route (...)
def request (was):
  reqs = [
    was.get (url),
    was.post (url, {"user": "Hans Roh", "comment": "Hello"})
  ]
  a, b = was.Tasks (reqs, timeout = 3).fetch ()
  return was.API (a = a, b = b)

Tasks is iterable and slicable and returened rs is response object (by dispatch ()). You SHOULD check rs.status and status_code for validating response, or just use fetch () for raising error if invalid.

  • Tasks (reqs, timeout = 10, **meta)
  • Tasks.add (corequest): append corequest or Task object
  • Tasks.merge (corequest): append corequest or Task object, in case Tasks, it will be extracted from inner corequests
  • Tasks.then (callabck): convert Tasks to Futures, available only for Atila app
  • Tasks.dispatch (cache = None, cache_if = (200,), timeout = None)
  • Tasks.wait (timeout = None)
  • Tasks.commit (timeout = None)
  • Tasks.fetch (cache = None, cache_if = (200,), timeout = None)
  • Tasks.one (cache = None, cache_if = (200,), timeout = None)
  • Tasks.meta: dictionary container for user data

Note: If you want to use full asynchronous manner, you can consider Atila’s Futures, but it need to pay some costs.

Calling RPC

@app.route (...)
def request (was):
  with was.xmlrpc ("@myrpc") as stub:
    req = stub.get_version ("skitai")
    return req.fetch () # ["0.29"]

    # or single line
    return stub.get_version ("skitai").fetch ()

was.jsonrpc and was.grpc (Experimental) are also possible.

For gRPC example, calling to tfserver for predicting something with tensorflow model.

from tfserver import cli

@app.route (...)
def predict_grpc (was):
  stub = was.grpc ("http://127.0.0.1:5000/tensorflow.serving.PredictionService")
  fftseq = getone ()
  request = cli.build_request ('model', 'predict', stuff = fftseq)
  req = stub.Predict (request, 10.0)
  resp = req.dispatch ()
  return cli.Response (resp.data).y

RDBMS Querying

Important Note: Async mode you cannot use transaction, and auto commit will be applied.

PostgreSQL query at aquests, First uou alias your database before running Skitai.

skitai.alias ("@mypg", skitai.DB_PGSQL, "user:pass@localhost/mydb")
skitai.alias ("@mylite", skitai.DB_SQLITE3, "./sqlite3.db")
skitai.run ()

Then,

@app.route (...)
def query (was):
  with was.db ("@mypg") as db:
    req = db.excute ("SELECT city, t_high, t_low FROM weather;")
    resp = req.dispatch (timeout = 2)
    if resp.status != 200:
      raise HTTPError ("500 Server Error")
  for row in rows:
    row.city, row.t_high, row.t_low

For consistency handling response of API calls, response.status_code will be set 200 if any error does not occure, otherwise set 500.

Basically Skitai handle as same for all kind of external requests.

@app.route (...)
def query (was):
  with was.db ("@mypg") as db:
    req = db.excute ("SELECT city, t_high, t_low FROM weather;")
    rows = req.fetch (2)
  for row in rows:
    row.city, row.t_high, row.t_low

If you needn’t returned data and just wait for completing query,

db.execute ("INSERT INTO CITIES VALUES ('New York');").commit (timeout = 2)

If failed, exception will be raised.

In case database querying, you can use one () method.

@app.route (...)
def query (was):
  with was.db ("@mypg") as db:
    hispet = db.excute ("SELECT ... FROM pets").one (timeout = 2)

If result record count is not 1 (zero or more than 1), raise HTTP 410 error.

With PostgreSQL you can also raise HTTP 409 using returning caluse.

@app.route (...)
def query (was):
  with was.db ("@mypg") as db:
    hispet = db.excute ("INSERT INTO pets ... RETURNING id").one (timeout = 2)

If primary key or unique key is duplicated, psycopg2 raises IntegrityError then Skitai raise HTTP 409 Conflict error

CAUTION: DO NOT even think your statements will be executed ordered sequencially.

@app.route (...)
def query (was):
  with was.db ("@mypg") as db:
    reqs = [
      db.excute ("INSERT INTO weather (id, 'New York', 9, 25);"),
      db.excute ("SELECT city, t_high, t_low FROM weather order by id desc limit 1 ;")
    ]
    Tasks (reqs) [1].fetch () # No guarantee it is New York or something new

Execute and wait or use transaction.

@app.route (...)
def query (was):
  with was.db ("@mypg") as db:
    db.excute ("INSERT INTO weather (id, 'New York', 9, 25);").commit ()
    latest = db.excute ("SELECT city, t_high, t_low FROM weather order by id desc limit 1 ;").fetch (2)
    # latest  is New York

Using Database Transaction

If you want use asynchronous database transaction,
you can use asynchronous drivers.

Also Skitai provide PostgreSQL connection with connection pool. And SQLite connection without pool.

@app.route ("/")
def index (was):
    with was.transaction ("@mypg") as tx:
        tx.execute ('INSERT ...')
        tx.execute ('UPDATE ...')
        tx.execute ('SELECT ...')
        tx.fetch () # equivlant to fetchall () but list of dict type
        tx.commit ()

With context manager, connection will return back to the pool automatically else you SHOULD call tx.putback () manually.

In transaction mode, standard DBAPI - rollback (), fetchall (), fetchone () and fetchmany () are also possible but caching is not.

was.transaction has second paramter ‘auto_putback’. If it is False, transaction object does not return to the pool automatically.

# models.py
from skitai import was

def update (...):
    with was.transaction ("@mypg", False) as tx:
        tx.execute ('INSERT ...')
        tx.execute ('UPDATE ...')
        tx.execute ('SELECT ...')
        return tx

        tx.fetch () # equivlant to fetchall () but list of dict type

# app.py
import models

@app.route (...)
def update (was):
  tx = models.update (...)
  rows = tx.fetch ()
  tx.commit ()

Note that you MUST call commit/rollback finally, if not connection pool will be exhausted very soon and entire threads will be blocked.

Using SQLPhile for Querying

Actullay, was.db and was.transaction are fully intergrated with SQLPhile.

You can write with raw SQL,

with was.db ("@mydb") as db:
  rows = db.execute (
    "SELECT a.id, b.name, c.phone "
    "FROM user a, profile b, contact c "
    "WHERE b.name like '%{name}%'"
    "ORDER BY a.id desc"
    "LIMIT {limit}".format (name = name, limit = limit)
  ).fetch ()

But also can use SQLPhile style,

with was.db ("@mydb") as db:
  rows = (db.get ("a.id, b.name, c.phone")
          .select ("user a, profile b, contact c")
          .filter (b__name__contains = name)
          .order_by ("-a.id") [:limit]
          .execute ().fetch ())

It may be not very helpful because of my laziness of documentation, however SQLPhile can provide some other benefits using SQL I recommend read it instantly.

NoSQL Querying

skitai.alias ("@mymongo", skitai.DB_MONGODB, "localhost/mycollection")
skitai.alias ("@myredis", skitai.DB_REDIS, "localhost/0")
skitai.run ()

Then,

@app.route (...)
def query (was):
  with was.db ("@mymongo") as db:
    documents = db.find ({'city': 'New York'}).fetch (2)

  with was.db ("@myredis") as db:
    db.set('foo', 'bar').wait ()
    db.get('foo').fetch () # bar

Request As Many You Need

For getting concurrent tasks advantages, you request at
once as many as possible.
@app.route (...)
def query (was):
  reqs = was.post ("@pypi/upload...", {data: ...})
  reqs = was.get ("@pypi/somethong..."})
  with was.db ("@mypg") as db:
    reqs.append (db.excute ("SELECT ..."))
    reqs.append (db.excute ("SELECT ..."))

  with was.jsonrpc ("@pypi/pypi") as stub:
    reqs.append (stub.get_version ("skitai"))
    reqs.append (stub.get_version ("atila"))

  contents = []
  for rs in Tasks (reqs, 3):
    if rs.status_code != 200:
      contents.append ("Error")
    else:
      contents.append (str (rs.data))
  return contents

Intermezzo

For creating corequest object,

  • HTTP based request: was.get (alias), .post (alias), ….
  • Database request: as.db (alias).execute (…), .find (), set (), … other MongoDB and Redis methods
  • Tasks: bundle of corequests

Corequest object has main 5 methods.

  • dispatch (): it returns Result object contains data (or text/content) and request status information
  • wait (): it returns Result object contains request status information
  • fetch (): it returns records list. if request failed raise exception
  • one (): it returns one record if query result length is exactly one otherwise raise 410 or 409 HTTP error. if request failed raise exception
  • commit (): it wait finishing non-select query, if request failed raise exception

Result object is mainly used for checking status and handling error to individual corequest, and Result object also has fetch (), one () and commit ().

Please DO remember. If ou call dispatch, fetch, … to corequest object, it immediatly act as synchronous task. But already created another corequests are still has concurrency.

Load-Balancing

Skitai support load-balancing requests.

If server members are pre defined, skitai choose one automatically per each request supporting fail-over.

Then let’s request XMLRPC result to one of mysearch members.

@app.route ("/search")
def search (was, keyword = "Mozart"):
  with was.jsonrpc.lb ("@mysearch/rpc2") as stub:
    s = stub.search (keyword)
    results = s.dispatch (timeout = 5)
    return result.data

    # or short hand
    return stub.search (keyword).fetch (5)

if __name__ == "__main__":
  import skitai

  skitai.alias (
    '@mysearch',
     skitai.PROTO_HTTPS,
     ["s1.myserver.com", "s2.myserver.com"]
  )
  skitia.mount ("/", app)
  skitai.run ()

It just small change from was.jsonrpc () to was.jsonrpc.lb ()

Note: If @mysearch member is only one, was.get.lb (“@mydb”) is equal to was.get (“@mydb”).

Note2: You can mount cluster @mysearch to specific path as proxypass like this:

if __name__ == "__main__":
  import skitai

  skitai.alias (
    '@mysearch',
     skitai.PROTO_HTTPS,
     ["s1.myserver.com", "s2.myserver.com:443"]
  )
  skitia.mount ("/", app)
  skitia.mount ("/search", '@mysearch')
  skitai.run ()

It can be accessed from http://127.0.0.1:5000/search, and handled as load-balanced proxypass. And it will be remapped to http://s1.myserver.com/.

If you mount like this,

skitia.mount ("/search", '@mysearch/search')

It can be accessed from same URL, but it will be remapped to http://s1.myserver.com/search.

This sample is to show loadbalanced querying database. Add mydb members to config file.

@app.route ("/query")
def query (was, keyword):
  with was.db.lb ("@mydb") as dbo:
    req = dbo.execute ("SELECT * FROM CITIES;")
    result = req.dispatch (timeout = 2)

 if __name__ == "__main__":
  import skitai

  skitai.alias (
    '@mydb',
     skitai.PGSQL,
     [
       "s1.yourserver.com:5432/mydb/user/passwd",
       "s2.yourserver.com:5432/mydb/user/passwd"
     ]
  )
  skitia.mount ("/", app)
  skitai.run ()

Map-Reducing

Basically same with load_balancing except Skitai requests to all members per each request.

@app.route ("/search")
def search (was, keyword = "Mozart"):
  with was.rpc.map ("@mysearch/rpc2") as stub:
    req = stub.search (keyword)
    results = req.dispatch (timeout = 2)

  all_results = []
  for result in results:
     all_results.extend (result.data)
  return all_results

There are 2 changes:

  1. from was.rpc.lb () to was.rpc.map ()
  2. results is iterable

You can use Dataabse, API calls same way.

Caching Result

By default, all HTTP requests keep server’s cache policy given by HTTP response header (Cache-Control, Expire etc). But you can control cache as your own terms including even database query results.

Every results returned by dispatch() can cache.

s = was.rpc.lb ("@mysearch/rpc2").getinfo ()
result = s.dispatch (60, timeout = 2) # cache seconds
result.data

s = was.rpc.map ("@mysearch/rpc2").getinfo ()
results = s.dispatch (60, timeout = 2)

Cahing when just only Although code == 200 alredy implies status == STA_NORMAL.

New in version 0.15.28

You can control number of caches by your system memory before running app.

skitai.set_max_rcache (300)
skitai.mount ('/', app)
skitai.run ()

For expiring cached result by updating new data:

refreshed = False
if was.request.method == "POST":
  ...
  refreshed = True

s = was.rpc.lb (
  "@mysearch/rpc2",
  use_cache = not refreshed and True or False
).getinfo ()
result = s.fetch (2, 60)

If you want cache for another status_code,

s = was.rpc.lb (
  "@mysearch/rpc2",
  use_cache = not refreshed and True or False
).getinfo ()
result = s.dispatch (60, (200, 201), timeout = 2)

More About Cache Control: Model Synchronized Cache

New in version 0.26.15

You can efficient cache with explicit model mutation time.

  • when your model is changed, call was.setlu (“model-state-name”)
  • when query your model, add parameter - was.getlu (“model-state-name”), for deciding if use cache or not

Note that it is useful only if your model make regular and controlled mutation by single or a few producer (any of computer, machine or human). Otherwise you could consider NoSQL things for your cache system, and Skitai corequest support MongoDB and Redis.

Corequest’s use_cache parameter value can be True, False or last updated time of base object. If last updated is greater than cached time, cache will be expired immediately and begin new query/request.

You can integrate your models changing and cache control.

First of all, you should set all cache control keys to Skitai for sharing model state beetween worker processes.

skitai.register_states ('tables.users', 'table.photos')

These key names are might be related your database model names nor table names. In general cases, key names are fine if you easy to recognize.

These key names are not mutable and you cannot add new key after calling skitai.run ().

Also it can be used as decorator for clarency.

import skitai

@skitai.register_states ('tables.users')
class User:
  ...


@skitai.register_states ('tables.users', 'table.photos')
def __mount__ (app):
  @app.route (...)
  def index (...):
     ...

Then you can use setlu () and getlu (),

app = Atila (__name__)

@app.route ("/update")
def update (was):
  # update users tabale
  was.db ('@mydb').execute (...)
  # update last update time by key string
  was.setlu ('tables.users')

@app.route ("/query1")
def query1 (was):
  # determine if use cache or not by last update information 'users'
  was.db ('@mydb', use_cache = was.getlu ('tables.users')).execute (...)

@app.route ("/query2")
def query2 (was):
  # determine if use cache or not by last update information 'users'
  was.db ('@mydb', use_cache = was.getlu ('tables.users')).execute (...)

It makes helping to reduce the needs for building or managing caches. And the values by setlu() are synchronized between Skitai workers by multiprocessing.Array.

If your query related with multiple models,

use_cache = was.getlu ("myapp.models.User", "myapp.models.Photo")

was.getlu () returns most recent update time stamp of given models.

Available on Python 3.5+

Also was.setlu () emits ‘model-changed’ events. You can handle event if you need. But this event system only available on Atila middle-ware.

app = Atila (__name__)

@app.route ("/update")
def update (was):
  # update users tabale
  was.db ('@mydb').execute (...)
  # update last update time by key string
  was.setlu ('tables.users', something...)

@app.on_broadcast ("model-changed:tables.users")
def on_broadcast (was, *args, **kargs):
  # your code

Note: if @app.on_broadcast is located in mount function at services directory, even app.use_reloader is True, it is not applied to app when component file is changed. In this case you should manually reload app by resaving app file.

Corequest Based Model

Here’s an model example with RDBMS.

Alias Your Database

First of all, alias your database to Skitai.

# serve.py
...
skitai.alias ("@blog", skitai.DB_PGSQL, "postgres:password@localhost/blog")
...
skitai.run (port = 5000)

Create Model Classes

I think all public model methods maybe return corequest object or None.

# services/models.py

from skitai import was
import skitai
from sqlphile import Q
from datetime import datetime

class BlogPost:
  EXCLUDES = Q (share = 'test')

  @classmethod
  def search (cls, keyword = None, period = None, offset = 0, limit = 10, fields = "*"):
      with was.db ("@blog") as db:
          stem = (db.select ("blogpost")
                   .get (fields)
                   .exclude (cls.EXCLUDES)
                   .filter (posted_at__between = period)
                   .filter (Q (title__contains = keyword) | Q (content__contains = keyword)))

          reqs = [
              stem.branch ().get ("count (*) as total").execute (),
              (stem.branch ()
                  .order_by ("-posted_at").offset (offset).limit (limit)
                  .execute ())
          ]
          return was.Tasks (reqs)

  @classmethod
  def get (cls, id, fields = "*"):
      with was.db ("@blog") as db:
          return (db.select ("blogpost")
                      .get (fields)
                      .filter (id = id).execute ())

  @classmethod
  def delete (cls, id):
      # example for transaction deletion
      was.setlu (STATE_POST)
      with was.transaction ("@blog") as db:
          (db.delete ("blogcomment")
                      .filter (post_id = id).execute ())
          (db.delete ("blogpost")
                      .filter (id = id).execute ())
          db.commit ()

  @classmethod
  def add (cls, post):
      was.setlu (STATE_POST)
      with was.db ("@blog") as db:
          return (db.insert ("blogpost")
                      .data (post)
                      .returning ("id").execute ())

  @classmethod
  def update (cls, id, post):
      was.setlu (STATE_POST)
      post ["updated_at"] = datetime.now ()
      with was.db ("@blog") as db:
          return (db.update ("blogpost")
                      .data (post)
                      .filter (id = id).execute ())

  @classmethod
  def get_comments (cls, id, offset = 0, limit = 10):
      with was.db ("@blog") as db:
          return (db.select ("blogcomment")
                    .filter (post_id = id)
                    .offset (offset).limit (limit)
                    .execute ())

  @classmethod
  def get_stat (cls, dateunit = 'year'):
      with was.db ("@blog") as db:
          return (db.select ("blog")
                  .get (f"date_part('{dateunit}', created_at) as year, count (*) as cnt")
                  .group_by ("year")
                  .execute ())

Using Models

Finally, you can use this models.py.

# services/blog.py
from . models import BlogPost

@app.route ("/posts/", methods = ["GET", "POST"])
def posts (was, offset = 0, limit = 10, **payload):
  if was.request.method == "GET":
    stat, posts = BlogPost.search (offset = int (offset), limit = int (limit)).fetch ()
    return was.API (posts = posts, total = stat [0].total)

  new_post = BlogPost.add (payload).one ()
  return was.API ("201 Created", id = new_post.id)

@app.route ("/posts/<int:id>", methods = ["GET", "PATCH", "DELETE", "OPTIONS"])
def post (was, id, num_comments = 0):
  if was.request.method == "GET":
    comments_ = BlogPost.get_comments (id, 0, int (num_comments))
    post = BlogPost.get (id).one ()
    post.comments = comments_.fetch ()
    return was.API (post = post)

  if was.request.method == "DELETE":
    BlogPost.delete (id)
    return was.API ("204 No Content")
  ...

@app.route ("/posts/int:id>/comments", methods = ["GET", "PATCH", "DELETE", "OPTIONS"])
def comments (was, id, offset = 0, limit = 10):
  if was.request.method == "GET":
    comments = BlogPost.get_comments (id, int (offset), int (limit)).fetch ()
    return was.API (comments = comments)
  ...

Conclusion

Above example pattern is just one of my implemetation with async models.

It can be extended and changed into NoSQL or even RESTful/RPC with any Skitai corequest object which has same 5 methods - dispatch, wait, fetch, one and commit.

Background Tasks

Skitai integrated async/sync concurrents. They have also very same usage and methods like fetch, one, dispatch etc.

Task(s) object is natively async corequests. It creates backgorund async jobs and can be responded immediately.

@app.route ('...')
def foo ():
  req = was.get ("@myupstream/something")
  return  req.returning (
    Response ('', 202, headers = {'Content-Location': "..."})
  )

Tasks is also available,

@app.route ('...')
def foo ():
  reqs = [
    was.get ("@myupstream/something"),
    was.post ("@myupstream/something", {})
  ]
  return was.Tasks (reqs).returning (
    Response ('', 202, headers = {'Content-Location': "..."})
  )

Note: With Atila, you can add callback for late response.

Process / Thread is very same as Task.

Skitai will create thread/process pool as you use it at once. If you do’t use this, pool will not be created for resource saving. Pool size is your number of CPUs.

You can just use multi processing with pool instantly.

def side_job (a, b):
  ...

@app.route ('...')
def foo ():
  ps = was.Process (job2, 1000, -1000)
  ...
  result = ps.fetch () # wait for finishing
  return Response (result, 200, headers = {'Content-Type': "application/vnd-..."})

Also you can create async jobs for long run process.

@app.route ('...')
def foo ():
  return was.Process (job2, 1000, -1000).returning (
    Response ('', 202, headers = {'Content-Location': "..."})
  )

was.Thread () and was.Subprocess () are also available.

  • was.Thread (target, *args, **kargs): return wrapper of concurrent.futures.Future
  • was.Process (target, *args, **kargs): return wrapper of concurrent.futures.Future
  • was.Subprocess (command, timeout = 300): return wrapper of subprocess.Popen

Note: With Atila, you can add callback for late response.

Miscellaneous

API Transaction ID

New in version 0.21

For tracing REST API call, Skitai use global/local transaction IDs.

If a client call a API first, global transaction ID (gtxnid) is assigned automatically like ‘GTID-C4676-R67’ and local transaction ID (ltxnid) is ‘1000’.

You call was.get (), was.post () or etc, both IDs will be forwarded via HTTP request header. Most important thinng is that gtxnid is never changed by client call, but ltxnid will be changed per API call.

when client calls gateway API or HTML, ltxnid is 1000. And if it calls APIs internally, ltxnid will increase to 2001, 2002. If ltxnid 2001 API calls internal sub API, ltxnid will increase to 3002, and ltxnid 2002 to 3003. Briefly 1st digit is call depth and rest digits are sequence of API calls.

This IDs is logged to Skitai request log file like this.

2016.12.30 18:05:06 [info] 127.0.0.1:1778 127.0.0.1:5000 GET / \
HTTP/1.1 200 0 32970 \
GTID-C3-R8 1000 - - \
"Mozilla/5.0 (Windows NT 6.1;) Gecko/20100101 Firefox/50.0" \
4ms 3ms

Focus 3rd line above log message. Then you can trace a series of API calls from each Skitai instance’s log files for finding some kind of problems.

In next chapters’ features of ‘was’ are only available for Atila WSGI container. So if you have no plan to use Atila, just skip.

Utility Methods of ‘was’

This chapter’s ‘was’ services are also avaliable for all WSGI middelwares.

  • was.status () # HTML formatted status information
  • was.get_lock (name = “__main__”) # getting process lock
  • was.gentemp () # return temp file name with full path
  • was.restart () # Restart Skitai App Engine Server, but this only works when processes is 1 else just applied to current worker process.
  • was.shutdown () # Shutdown Skitai App Engine Server, but this only works when processes is 1 else just applied to current worker process.

Change Log

  • 0.32 (Oct 2019)
    • initiate HTTP3+QUIC, you can test HTTP/3 with Chrome Canary
  • 0.31 (Sep 2019)
    • change handling command line options, required rs4>=0.2.5.0
    • add skitai.set_smtp ()
    • remove protobuf, redis, pymongo and psycopg2 from requirements, if you need these, install them maually
    • skitai.preference () can be used with context
    • fix http/2 response delaying when body is not exist
    • skitai.enable_forward () can forward to single domain
    • add dropping root privileges when Skitai run with sudo for using under 1024 ports etc.
    • refix: master process does not drop root privileges for clean resources
    • fix reloading for file mounted apps
    • confirmed to work on PyPy3
  • 0.30 (Sep 2019)
    • skitai.websocket spec changed, lower version compatable
  • 0.29 (Aug 2019)
    • add was.Subprocess
    • add handlers for Range, If-Range, If-Unmodified-Since, If-Match headers
    • asyncore and asynchat are vendored as rs4.asyncore and chat, because they will be exsanguinated from standard Python library. Mr. Rossum has been listed up on my mortal enemy list
    • deprecated: was.Future and was.Futures, it doesn’t need. for using returning (), use corequest.returning () and was.Tasks.returning ()
    • new corequest.pth package
    • over 100 unit tests
  • 0.28 (Feb 2019)
    • fix auto reloading bug in case multiple apps are mounted
    • add was.Thread () and was.Process ()
    • add @skitai.states () decorator
    • rename skitai.deflu () => skitai.register_states ()
    • add corequest object explaination and corequest based model example
    • drop SQLAlchemy query statement object
    • fix https proxypass, and add proxypass remapping
    • add was.transaction ()
    • update psycopg2 connection parameter: async => async_ for Py3.7 compatablity
    • replace from data_or_thow (), one_or_throw () to fetch (), one ()
    • fix HTTP2 server push and add was.push ()
    • getwait () and getswait () are integrated into dispatch ()
    • add data_or_throw () and one_or_throw ()
    • was.promise has been deprecated, use was.futures: see Atila documentation
    • reinstate gc.collect () schedule
    • fix GTXID
    • fix app reloader
    • remove gc.collect () schedule
    • support SQLAlchemy query statement object
    • removed sugar methods: was.getjson, getxml, postjson, …, instead use headers parameter or app.config.default_request_type
    • skitai.win32service has been moved to rs4.psutil.win32service
    • improve ‘was’ magic method search speed
    • seperate skitai.saddle into atila
  • 0.27.6 (Jan 2019)
    • rename directory decorative to services
    • change from skital.saddle.contrib.decorative to skital.saddle.contrib.services
  • 0.27.3 (May 2018)
    • remove -v option from skitai and smtpda
    • add script: skitai
    • remove scripts: skitai-smtpda and skitai-cron
    • remove skitai.enable_smtpda (), skitai.cron ()
  • 0.27.2 (May 2018)
    • add was.request.get_real_ip () and was.request.is_private_ip ()
    • fix CORS preflight
  • 0.27.1 (May 2018)
    • sqlphile bug fixed and change requirements
  • 0.27 (Apr 2018)
    • add app.setup_sqlphile ()
    • add @app.mounted_or_reloaded decorator
    • removed @app.auth_required, added @app.authorization_required (auth_type)
    • rename @app.preworks -> @app.run_before and @app.postworks -> @app.run_after
    • add @app.bearer_handler
    • add was.mkjwt and was.dejwt
    • add was.timestamp amd was.uniqid
    • renamed was.token -> was.mktoken
    • renamed api -> API, for_api -> Fault
    • skitai.use_django_models has been deprecated, use skitai.alias
    • functions are integrated skitai.mount_django into skitai.mount, skitai.alias_django into skitai.alias
    • fix empty payload posting
    • add was.partial and was.basepath
    • raise NameError when non-exists funtion name to was.ap
    • fix default arg is missing on was.ab
    • add skitai.launch and saddle.make_client for unittest

0.26 (May 2017)

  • 0.26.18 (Jan 2018)
    • fix HTTP2 trailers
    • fix HTTP2 flow control window
    • remove was.response.traceback(), use was.response.for_ap (traceback = True)
    • rename was.sqlmap to was.sql
    • add @app.auth_required and @app.auth_not_required decorator
    • change default export script to __export__.py
    • remove app reloading progress:
      • before:
        • before_umount (was)
        • umounted (wac)
        • before_remount (wac): deprecated
        • remounted (was): deprecated
      • now:
        • before_reload (was)
        • reloaded (was)
    • change app.model_signal () to app.redirect_signal (), add @app.on_signal ()
    • change skitai.addlu to skitai.deflu (args, …)
    • add @app.if_file_modified
    • add @app.preworks and @app.postworks
    • fix HTTP/2 remote flow control window
    • fix app.before_mount decorator exxcute point
    • add was.gentemp () for generating temp file name
    • add was.response.throw (), was.response.for_api() and was.response.traceback()
    • add @app.websocket_config (spec, timeout, onopen_func, onclose_func, encoding)
    • was.request.get_remote_addr considers X-Forwarded-For header value if exists
    • add param keep param to was.csrf_verify()
    • add and changed app life cycle decorators:
      • before_mount (wac)
      • mounted (was)
      • before_remount (wac)
      • remounted (was)
      • before_umount (was)
      • umounted (wac)
    • add skitai.saddle.contrib.django,auth for integrating Django authorization
    • change was.token(),was.detoken(), was.rmtoken()
    • add jsonrpc executor
    • add some methods to was.djnago: login (), logout (), authenticate () and update_session_auth_hash ()
    • add app.testpass_required decorator
    • add decorative concept
  • 0.26.17 (Dec 2017)
    • can run SMTP Delivery Agent and Task Scheduler with config file
    • add error_handler (prev errorhandler) decorator
    • add default_error_handler (prev defaulterrorhandler) decorator
    • add login_handler, login_required decorator
    • add permission_handler, permission_required decorator
    • add app events emitting
    • add was.csrf_token_input, was.csrf_token and was.csrf_verify()
    • make session iterable
    • prevent changing function spec by decorator
    • change params of use_django_models: (settings_path, alias), skitai.mount_django (point, wsgi_path, pref = pref (True), dbalias = None, host = “default”)
  • 0.26.16 (Oct 2017)
    • add app.sqlmaps
    • add use_django_models (settings_path), skitai.mount_django (point, wsgi_path, pref = pref (True), host = “default”)
    • fix mbox, add app.max_client_body_size
    • add skitai.addlu (args, …)
    • fix promise and proxing was objects
    • change method name from skitai.set_network_timeout to set_erquest_timeout
    • fix getwait, getswait. get timeout mis-working
    • fix backend_keep_alive default value from 10 to 1200
    • fix dbi reraise on error
    • JSON as arguments
  • 0.26.15
    • added request.form () and request.dict ()
    • support Django auto reload by restarting workers
    • change DNS query default protocol from TCP to UDP (posix only)
    • add skitai.set_proxy_keep_alive (channel = 60, tunnel = 600) and change default proxy keep alive to same values
    • increase https tunnel keep alive timeout to 600 sec.
    • fix broad event bus
    • add getjson, deletejson, this request automatically add header ‘Accept: application/json’
    • change default request content-type from json to form data, if you post/put json data, you should change postjson/putjson
    • add skitai.trackers (args,…) that is equivalant to skitai.lukeys ([args])
    • fix mounting module
    • app.storage had been remove officially, I cannot find any usage. but unoficially it will be remains by some day
    • add skitai.lukeys () and fix inconsistency of was.setlu & was.getlu between multi workers
    • was.storage had been remove
    • add skitai.set_worker_critical_point ()
    • fix result object caching
    • add app.model_signal (), was.setlu () and was.getlu ()
  • 0.26.14
    • add app.storage and was.storage
    • removed wac._backend and wac._upstream, use @app.mounted and @app.umount
    • replaced app.listen by app.on_broadcast
  • 0.26.13
    • add skitai.log_off (path,…)
    • add reply content-type to request log, and change log format
    • change posix process display name
  • 0.26.12
    • change event decorator: @app.listen -> @app.on_broadcast
    • adaptation to h2 3.0.1
    • fix http2 flow controling
    • fix errorhandler and add defaulterrorhandler
    • fix WSGI response handler
    • fix cross app URL building
    • Django can be mounted
    • fix smtpda & default var directory
    • optimize HTTP/2 response data
    • fix HTTP/2 logging when empty response body
    • http_response.outgoing is replaced by deque
    • change default mime-type from text/plain to application/octet-stream in response header
    • HTTP response optimized
  • 0.26.10
    • start making pytest scripts
    • add was-wide broadcast event bus: @app.listen (event), was.broadcast (event, args…) and @was.broadcast_after (event)
    • add app-wide event bus: @app.on (event), was.emit (event, args…) and @was.emit_after (event)
    • remove @app.listento (event) and was.emit (event, args…)
  • 0.26.9
    • add event bus: @app.listento (event) and was.emit (event, args…)
  • 0.26.8
    • fix websocket GROUPCHAT
    • add was.apps
    • was.ab works between apps are mounted seperatly
  • 0.26.7
    • add custom error template on Saddle
    • add win32 service tools
    • change class method name from make_request () to backend ()
    • retry once if database is disconnected by keep-live timeout
    • drop wac.make_dbo () and wac.make_stub ()
  • 0.26.6
    • add wac.make_dbo (), wac.make_stub () and wac.make_request ()
    • wac.ajob () has been removed
    • change repr name from wasc to wac
    • websocket design spec, WEBSOCKET_DEDICATE_THREADSAFE has been removed and WEBSOCKET_THREADSAFE is added
    • fix websocket, http2, https proxy tunnel timeout, related set_network_timeout () is recently added
  • 0.26.4.1: add set_network_timeout (timoutout = 30) and change default keep alive timeout from 2 to 30
  • 0.26.4: fix incomplete sending when resuested with connection: close header
  • 0.26.3.7: enforce response to HTTP version 1.1 for 1.0 CONNECT with 1.0 request
  • 0.26.3.5: revert multiworkers
  • 0.26.3.2: fix multiworkers
  • 0.26.3.1: update making for self-signing certification
  • 0.26.3: add skitai.enable_forward
  • 0.26.2.1: remove was.promise.render_all (), change method name from was.promise.push () to send ()
  • 0.26.2: change name from was.aresponse to was.promise
  • 0.26.1.1: add skitai.abspath (*args)
  • 0.26.1: fix proxy & proxypass, add was.request.scheme and update examples
  • change development status to Beta
  • fix Saddlery routing
  • disable WWW-Authenticate on websocket protocol
  • support CORS (Cross Origin Resource Sharing)
  • support PATCH method
  • runtime app preferences and add __init__.bootstrap (preference)
  • fix route caching
  • auto reload sub modules in package directory, if app.use_reloader = True
  • new was.request.json ()
  • integrated with skitaid package, single app file can contain all configure options
  • level down developement status to alpha
  • fix sqlite3 closing

0.25 (Feb 2017)

  • 0.25.7: fix fancy url, non content-type header post/put request
  • 0.25.6: add Chameleon template engine
  • 0.25.5: app.jinja_overlay ()’s default args become jinja2 default
  • 0.25.4.8: fix proxy retrying
  • 0.25.4 license changed from BSD to MIT, fix websocket init at single thread
  • 0.25.3 handler of promise args spec changed, class name is changed from AsyncResponse to Promise
  • 0.25.2 fix promise exception handling, promise can send streaming chunk data
  • 0.25.1 change app.jinja_overlay () default values and number of args, remove raw line statement
  • project name chnaged: Skitai Library => Skitai App Engine

0.24 (Jan 2017)

  • 0.24.9 bearer token handler spec changed
  • 0.24.8 add async response, fix await_fifo bug
  • 0.24.7 fix websocket shutdown
  • 0.24.5 eliminate client arg from websocket config
  • 0.24.5 eliminate event arg from websocket config
  • fix proxy tunnel
  • fix websocket cleanup
  • change websocket initializing, not lower version compatible
  • WEBSOCKET_MULTICAST deprecated, and new WEBSOCKET_GROUPCHAT does not create new thread any more

0.23 (Jan 2017)

  • ready_producer_fifo only activated when proxy or reverse proxy is enabled, default deque will be used
  • encoding argument was eliminated from REST call
  • changed RPC, DBO request spec
  • added gRPC as server and client
  • support static files with http2
  • fix POST method on reverse proxying

0.22 (Jan 2017)

  • 0.22.7 fix was.upload(), was.post*()
  • 0.22.5 fix xml-rpc service
  • 0.22.4 fix proxy
  • 0.22.3
    • fix https REST, XML-RPC call
    • fix DB pool
  • 0.22
    • Skitai REST/RPC call now uses HTTP2 if possible
    • Fix HTTP2 opening with POST method
    • Add logging on disconnecting of Websocket, HTTP2, Proxy Tunnel channels
    • See News

0.21 (Dec 2016)

  • 0.21.17 - fix JWT base64 padding problem
  • 0.21.8 - connected with MongoDB asynchronously
  • 0.21.3 - add JWT (JSON Web Token) handler, see Skitai WSGI App Engine Daemon
  • 0.21.2 - applied global/local-transaction-ID to app logging: was.log (msg, logtype), was.traceback ()
  • 0.21 - change request log format, add global/local-transaction-ID to log file for backtrace

0.20 (Dec 2016)

  • 0.20.15 - minor optimize asynconnect, I wish
  • 0.20.14 - fix Redis connector’s threading related error
  • 0.20.4 - add Redis connector
  • 0.20 - add API Gateway access handler

0.19 (Dec 2016)

  • Reengineering was.request methods, fix disk caching

0.18 (Dec 2016)

  • 0.18.11 - default content-type of was.post(), was.put() has been changed from ‘application/x-www-form-urlencoded’ to ‘application/json’. if you use this method currently, you SHOULD change method name to was.postform()
  • 0.18.7 - response contents caching has been applied to all was.request services (except websocket requests).

0.17 (Oct 2016)

0.16 (Sep 2016)

  • 0.16.20 fix SSL proxy and divide into package for proxy & websocket_handler
  • 0.16.19 fix HTTP2 cookie
  • 0.16.18 fix handle large request body
  • 0.16.13 fix thread locking for h2.Connection
  • 0.16.11 fix pushing promise and response on Firefox
  • 0.16.8 fix pushing promise and response
  • 0.16.6 add several configs to was.app.config for limiting post body size from client
  • 0.16.5 add method: was.response.hint_promise (uri) for sending HTP/2 PUSH PROMISE frame
  • 0.16.3 fix flow control window
  • 0.16.2 fix HTTP/2 Uprading for “http” URIs (RFC 7540 Section 3.2)
  • 0.16 HTTP/2.0 implemented with hyper-h2

0.15 (Mar 2016)

  • fixed fancy URL <path> routing
  • add Websocket design spec: WEBSOCKET_DEDICATE_THREADSAFE
  • fixed Websocket keep-alive timeout
  • fixed fancy URL routing
  • ‘was.cookie.set()’ method prototype has been changed.
  • added Named Session & Messaging Box
  • fix select error when closed socket, thanks to spam-proxy-bots
  • add mimetypes for .css .js
  • fix debug output
  • fix asynconnect.maintern
  • fix loosing end of compressed content
  • fix app reloading, @shutdown
  • fix XMLRPC response and POST length
  • add was.mbox.search (), change spec was.mbox.get ()
  • fix routing bugs & was.ab()
  • add saddle.Saddlery class for app packaging
  • @app.startup, @app.onreload, @app.shutdown arguments has been changed

0.14 (Feb 2016)

  • fix proxy occupies CPU on POST method failing
  • was.log(), was.traceback() added
  • fix valid time in message box
  • changed @failed_request arguments and can return custom error page
  • changed skitaid.py command line options, see ‘skitaid.py –help’
  • batch task scheduler added
  • e-mail sending fixed
  • was.session.getv () added
  • was.response spec. changed
  • SQLite3 DB connection added

0.13 (Feb 2016)

  • was.mbox, was.g, was.redirect, was.render added
  • SQLite3 DB connection added

0.12 (Jan 2016) - Re-engineering ‘was’ networking, PostgreSQL & proxy modules

0.11 (Jan 2016) - Websocket implemeted

0.10 (Dec 2015) - WSGI support

Release history Release notifications

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for skitai, version 0.32.8b9
Filename, size File type Python version Upload date Hashes
Filename, size skitai-0.32.8b9.tar.gz (192.2 kB) File type Source Python version None Upload date Hashes View hashes

Supported by

Elastic Elastic Search Pingdom Pingdom Monitoring Google Google BigQuery Sentry Sentry Error logging AWS AWS Cloud computing DataDog DataDog Monitoring Fastly Fastly CDN SignalFx SignalFx Supporter DigiCert DigiCert EV certificate StatusPage StatusPage Status page