phorge-phorge/src/docs/user/userguide/almanac.diviner

@title Almanac User Guide
@group userguide

Using Almanac to manage devices and services.

Overview
========

Almanac is a device and service inventory application. It allows you to create
lists of //devices// and //services// that humans and other applications can
use to keep track of what is running where.

Almanac is an infrastructure application that will normally be used by
administrators to configure advanced Phabricator features. In most cases,
normal users will very rarely interact with Almanac directly.

At a very high level, Almanac can be thought of as a bit like a DNS server.
Callers ask it for information about services, and it responds with details
about which devices host those services. However, it can respond to a broader
range of queries and provide more detailed responses than DNS alone can.

Today, the primary use cases for Almanac are internal to Phabricator:

  - Providing a list of build servers to Drydock so it can run build and
    integration tasks.
  - Configuring Phabricator to operate in a cluster setup.

Beyond internal uses, Almanac is a general-purpose service and device inventory
application and can be used to configure and manage other types of service and
hardware inventories, but these use cases are currently considered experimental
and you should be exercise caution in pursuing them.


Example: Drydock Build Pool
================================

Here's a quick example of how you might configure Almanac to solve a real-world
problem. This section describes configuration at a high level to give you an
introduction to Almanac concepts and a better idea of how the pieces fit
together.

In this scenario, we want to use Drydock to run some sort of build process. To
do this, Drydock needs hardware to run on. We're going to use Almanac to give
Drydock a list of hosts it should use.

In this scenario, Almanac will work a bit like a DNS server. When we're done,
Drydock will be able to query Almanac for information about a service (like
`build.mycompany.com`) and get back information about which hosts are part of
that service and which addresses/ports it should connect to.

Before getting started, we need to create a **network**. For simplicity, let's
suppose everything will be connected through the public internet. If you
haven't already, you'd create a "Public Internet" network first.

Once we have a network, we create the actual physical or virtual hosts by
launching instances in EC2, or racking and powering on some servers, or already
having some hardware on hand we want to use. We set the hosts up normally and
connect them to the internet (or another network).

After the hosts exist, we add them to Almanac as **devices**, like
`build001.mycompany.com`, `build002.mycompany.com`, and so on. In Almanac,
devices are usually physical or virtual hosts, although you could also use it
to inventory other types of devices and hardware.

For each **device**, we add an **interface**. This is just an address and port
on a particular network. Since we're going to connect to these hosts over
SSH, we'll add interfaces on the standard SSH port 22. An example configuration
might look a little bit like this:

| Device | Network | Address | Port |
|--------|---------|---------|------|
| `build001.mycompany.com` | Public Internet | 58.8.9.10 | 22
| `build002.mycompany.com` | Public Internet | 58.8.9.11 | 22
| ... | Public Internet | ... | 22

Now, we create the **service**. This is what we'll tell Drydock about, and
it can query for information about this service to find connected devices.
Here, we'll call it `build.mycompany.com`.

After creating the service, add **bindings** to the interfaces we configured
above. This will tell Drydock where it should actually connect to.

Once this is complete, we're done in Almanac and can continue configuration in
Drydock, which is outside the scope of this example. Once everything is fully
configured, this is how Almanac will be used by Drydock:

  - Drydock will query information about `build.mycompany.com` from Almanac.
  - Drydock will get back a list of bound interfaces, among other data.
  - The interfaces provide information about addresses and ports that Drydock
    can use to connect to the actual devices.

You can now add and remove devices to the pool by binding them and unbinding
them from the service.


Concepts
========

The major concepts in Almanac are **devices**, **interfaces**, **services**,
**bindings**, **networks**, and **namespaces**.

**Devices**: Almanac devices represent physical or virtual devices.
Usually, they are hosts (like `web001.mycompany.net`), although you could
use devices to keep inventory of any other kind of device or physical asset
(like phones, laptops, or office chairs).

Each device has a name, and may have properties and interfaces.

**Interfaces**: Interfaces are listening address/port combinations on devices.
For example, if you have a webserver host device named `web001.mycompany.net`,
you might add an interface on port `80`.

Interfaces tell users and applications where they should connect to to access
services and devices.

**Services**: These are named services like `build.mycompany.net` that work
a bit like DNS. Humans or other applications can look up a service to find
configuration information and learn which devices are hosting the service.

Each service has a name, and may have properties and bindings.

**Bindings**: Bindings are connections between services and interfaces. They
tell callers which devices host a named service.

**Networks**: Networks allow Almanac to distingiush between addresses on
different networks, like VPNs vs the public internet.

If you have hosts in different VPNs or on private networks, you might have
multiple devices which share the same IP address (like `10.0.0.3`). Networks
allow Almanac to distinguish between devices with the same address on different
sections of the network.

**Namespaces**: Namespaces let you control who is permitted to create devices
and services with particular names. For example, the namespace `mycompany.com`
controls who can create services with names like `a.mycompany.com` and
`b.mycompany.com`.


Namespaces
==========

Almanac namespaces allow you to control who can create services and devices
with certain names.

If you keep a list of cattle as devices with names like
`cow001.herd.myranch.com`, `cow002.herd.myranch.moo`, you might have some
applications which query for all devices in `*.herd.myranch.moo`, and thus
want to limit who can create devices there in order to prevent mistakes.

If a namespace like `herd.myranch.moo` exists, users must have permission to
edit the namespace in order to create new services, devices, or namespaces
within it. For example, a user can not create `cow003.herd.myranch.moo` if
they do not have edit permission on the `herd.myranch.moo` namespace.

When you try to create a `cow003.herd.myranch.moo` service (or rename an
existing service to have that name), Almanac looks for these namespaces, then
checks the policy of the first one it finds:

| Namespace |
|----|-----
| `cow003.herd.ranch.moo` | //"Nearest" namespace, considered first.//
| `herd.ranch.moo` | |
| `ranch.moo` | |
| `moo` | //"Farthest" namespace, considered last.//

Note that namespaces treat names as lists of domain parts, not as strict
substrings, so the namespace `herd.myranch.moo` does not prevent
someone from creating `goatherd.myranch.moo` or `goat001.goatherd.myranch.moo`.
The name `goatherd.myranch.moo` is not part of the `herd.myranch.moo` namespace
because the initial subdomain differs.

If a name belongs to multiple namespaces, the policy of the nearest namespace
is controlling. For example, if `myranch.moo` has a very restrictive edit
policy but `shed.myranch.moo` has a more open one, users can create devices and
services like `rake.shed.myranch.moo` as long as they can pass the policy check
for `shed.myranch.moo`, even if they do not have permission under the policy
for `myranch.moo`.

Users can edit services and devices within a namespace if they have edit
permission on the service or device itself, as long as they don't try to rename
the service or device to move it into a namespace they don't have permission
to access.