Permalink
Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
257 lines (182 sloc) 13.4 KB

Introduction

Attention!

triggerFS has been launched and is now available with all features enabled. This means that all signups are made with the free-tier with no limitations and all features enabled. Enjoy!

Welcome to triggerFS

triggerFS is a distributed, realtime message passing and trigger system. triggerFS enables you to build distributed systems and do realtime messaging in a service-oriented fashion.

It is made up of four modules:

  • worker
  • cli
  • client
  • fs

Features

  • firewall-friendly (only outbound connections being made by workers)
  • build lambda functions on your own servers
  • build a network of workers and services
  • make use of various plugins
  • fast, reliable and service-oriented networking
  • high-speed, low latency and asynchronous messaging
  • realtime stdout output
  • cluster-enabled services
  • different message passing algorithms on services: roundrobin, serial, mirror (parallel)
  • messaging via regular files with the fs module (triggers)
  • the fs module is available on every device (distributed, synced FUSE filesystem)
  • write your own plugin
  • invite others to your team and share resources with each other
  • join other teams and share resources with each other
  • make services public so everybody can use them

How it works

You deploy the worker module on your servers and make a request with the client module. The worker will take that request, pass it to a plugin and send the output back to the client. By specifying the plugin on the client-side you can control what plugin shall be used by the worker. You can then go ahead and group all workers by binding them to services.

These services act like a proxy to your workers and enable you to do the client requests in a load-balanced, serial or parallelized way. Services guarantee high-availability by doing a fail-over to other available workers in a group if one or more workers should fail within that group.

Here is a small list of things a plugin can do:

  • compute something
  • filter, grep, grok, sort or manipulate data
  • write to a file
  • log to a file
  • forward a message
  • read logs
  • collect metrics (eg. send metrics to graphite)
  • just echo back the received message
  • monitor the server the worker is running on
  • execute something on a remote machine via ssh
  • run a chef/puppet/fabric recipe or ansible playbook
  • run a bash script
  • send an SMS via twillio
  • send a text message to telegram
  • flip a switch (0/1 flipflops)
  • start/stop/restart a service
  • kill a process
  • chain requests by writing to another trigger file (this is fun)
  • ...

To learn more about Plugins please have a look at Plugins.

Roadmap

  • marketplace api for the plugin ecosystem
  • marketplace web UI
  • marketplace integration into cli module
  • streaming services for the worker module (long-running plugins/services)
  • listening feature for the client module (for streaming services)
  • service broadcasting feature
  • HTTP/broker gateway for making requests via HTTP
  • periodic tasks via HTTP/db
  • team mailboxes in cli for notifications from broker and triggerfs.io
  • log tables for storing output of plugins. (history of stdouts)
  • encrypted communication (messaging) channels (no content encryption)
  • and much more...

Modules

worker

A worker is or has a service and receives messages (requests) from one or more clients in a safe and concurrent way and executes them by using a plugin. Deploy a worker to a bunch of servers and connect them together, build a network of services for high-speed messaging, to distribute workload or create clusters or groups of workers.

Every worker has an identity and a token to authenticate against the broker. A worker can be attached to a service. If more than one worker is attached to the same service, the service acts as a load-balancer.

You can specify other algorithms beside load-balacing. Eg. serial (distribution of tasks to workers in order) or mirror (parallelized execution of a task on all workers behind that service).

Note

After successfully authenticated, a worker's identity is referred to as a service. We do not distinguish between them. However, a service you define in the cli is a sort of reverse proxy or alias for a worker (a real service). The identity of a worker is really only a string acting as an identifier.

Just remember, that once a worker is online, it's identity becomes a service. Hence the flag -service in the client module. You can pass both a real service or an identity of a worker to it.

cli

The cli module is an interactive cli-based management console for managing everything on your triggerFS platform. With the cli module you can:

  • create new users
  • create new workers
  • create new services
  • make services public (public services feature)
  • select the message dispatching algorithm (loadbalancing and more)
  • create triggers
  • make requests to your services in realtime
  • and much more

The triggerFS cli module is the central place for managing, orchestrating and configuring your triggerFS environment.

Suppose we wanted to do the following: If we know that there will be a service called demo with a command plugin and three workers behind that service (announcing that service), we could create a trigger named command, attach it to the demo service, add the demo service to all of our three workers and set a timeout of - let's say - 10s. Now we have created a trigger which is represented as a file in our fs module.

client

A client sends messages (requests) to a service and gets back a response. The client can specify which service it wants to talk to and what plugin shall be used. The client module is simple but powerful.

fs

The fs module is a module for mapping the above mentioned triggers to files with the help of FUSE. It was built to go a step further than just sending messages back and forth. It enables machine-to-machine communication. Mounting files is cheap and doing socket communication by using files makes this module attractive to embedded devices or small computers like the Raspberry Pi™.

Create a directory and define a trigger in that directory in your cli. Now, if you mount the fs module to a place on your filesystem, you end up with a file in that directory within that mountpoint. Every write to that file (with the content being the data written to that file) will send a request to the workers behind the above defined service with all the predefined set of rules we configured ealier. The fs module aims to make triggerFS app-friendly in such a way that other applications can use files as their way to send a message to your services.

For example: A trigger file could be defined in such a way that the result would be a logging of the request being sent to a service. Now we could tell Nginx to log into our trigger file instead of /var/log/nginx/*. Now everytime Nginx wants to log someting, it makes a syscall (write) to our file which would result in a message being sent. Our service would then write it to eg. a central NFS server of the company which is located on the machine where the worker is running.

Another example would be a raspberry pi which collects weather data and sends it to a central server (service) by writing into the trigger-files it mounted on its filesystem. Either scripted or syscalled. A simple echo 'somedata 31F;10°;3.2' > /mnt/triggerfs/weatherstation/rpi/station1 is enough to send your data.

What we just did is, we triggered an action by writing to a file. Hence the name trigger.

Note

You can't create regular files in your mountpoint. The fs module only supports trigger-files. The only allowed operations are:

  • mkdir to create directories
  • mv to rename trigger-files
  • chmod to set unix permissions on trigger-files

There is one more module called broker. This is the broker we maintain and operate in the cloud (the service behind triggerfs.io). The client/worker communication happens to be routed via the broker. The broker is the main coordinator for every message. It takes the request from the client and dispatches it to the services accordingly.

Security

Communication Flow

Beside our RESTful HTTP (JSON) API for database access, we use ZeroMQ for the communication between the client/worker and the broker. Every authenticated request to our API is done by using a JSON Web Token (JWT).

The communication/networking between clients and workers (services) are as follows:

  • client <==> broker <==> service (worker or service)

The central broker in the cloud (we, the triggerFS team) is responsible for routing the messages back and forth. A client cannot reach a worker without the broker and vise versa.

The broker exclusively uses the JWT of the client and/or worker if it has to make some operations on behalf of either part. This means that the JWT is also being sent when a message is sent. It is part of the message.

Since a JWT in triggerFS does not include sensitive data (only metadata) it is acceptable to send a JWT over the wire. However, in future releases we want to implement channel encryption on top of the SSL/TLS HTTP API calls, so that even the zmq channels we use to communicate are also encrypted in the future.

Database

Our database is powered by postgreSQL and here is a listing of what will be stored in our db:

  • users with their identity and secret (we use pgcrypto and bcrypt the password/secret before it is inserted into the db)
  • workers with their identities and tokens (we use uuid_v4 for the token of a worker)
  • teams with their configuration settings
  • services with their configuration settings
  • triggers with their configuration settings

This is everything which is stored in our db. The only sensitive data is the identity/secret of a user and identity/token of a worker and we make sure to use cryptography to secure those.

In future there will be a log table for storing the output of a plugin into the db. Which will possibly hold sensitive data. We will think about how to store those in a secure way.

Maintenances

If triggerfs.io schedules maintenances and/or the broker has to be shut down, all workers/services will automatically get notified and will reconnect as soon as the broker is up and running again.

In the future we will also notify the team with a notification message to their mailboxes (future feature), which they can access within their cli.

Pricing

triggerFS will have a three-tier plan. Here is an overview of the pricing plan:

  Free Basic Advanced
  1 team 1 team 2 teams
  2 workers 25 workers 100 workers
  1 service 5 services 20 services
  3 triggers 26 triggers 51 triggers
  2 users/team 9 users/team 25 users/team
  ✔ Unlimited access to marketplace
  ✔ Join other teams
  ✘ public services ✔ All features enabled
Price free $x/month $x/month
  • worker, service and trigger limits are per team
  • the free tier will always be free

Note

We are in the beta-testing phase. We can't exactly tell the price, but we will update it once we know how we want to charge our customers. Our goal is to launch this application in a beta-testing stage so we can estimate which resources will cost us how much. Based on that calculation we will try to offer a fair price to our customers.

If you have been using this application for a while, we would like to hear your feedback. You can reach us at feedback@triggerfs.io. Thank you.

Target Group

We think that devops and system administrators will love to use triggerFS due to the way it simplifies building tools such as automation systems and communication of services.

We see DCs (data centers) in general also as a target group. For example: A triggerfs-worker as a top-of-the-rack (tor) worker which is responsible for the systems in a rack to handle deployments, automation, triggering of jobs, etc. is one of the scenarios triggerFS can fit into.

Systemadministrators can use triggerFS for maintenance purposes or devops engineers can build whole clusters for various deployment scenarios.

In the end, it will be the massive amount of plugins which will enable triggerFS to become something useful for any possibly imaginable task.

Of course everybody is welcome to try out triggerFS (there is a free-tier subscription. Go try it out!)