Skip to content

Commit

Permalink
WIP (3) v2 content (still hidden)
Browse files Browse the repository at this point in the history
  • Loading branch information
@ssorj
ssorj committed Oct 14, 2024
1 parent 7d877f1 commit f47093d
Show file tree
Hide file tree
Showing 2 changed files with 58 additions and 109 deletions.
96 changes: 31 additions & 65 deletions docs/v2/index.html
View file
Original file line number Diff line number Diff line change
Expand Up @@ -100,8 +100,6 @@ <h2 id="the-move-to-custom-resources">The move to custom resources</h2>
something v1 did not require. This is an advantage for some of our
users, but a disadvantage for others. We believe that custom
resources are, on balance, the right choice.</p>
<!-- ref the adr -->

<h2 id="a-uniform-declarative-api">A uniform declarative API</h2>
<p>V2 has a new, uniform API for site configuration, site linking, and
service exposure. In v2, all of Skupper's interfaces and platforms
Expand Down Expand Up @@ -134,16 +132,10 @@ <h2 id="a-uniform-declarative-api">A uniform declarative API</h2>
<p>Service exposure in particular sees a change in v2. In v1, service
exposure is implicit: exposing a service in one site by default
exposed it in all the linked sites. In v2, service exposure is
instead explicit. A connector binds a local workload to a routing
instead <em>explicit</em>. A connector binds a local workload to a routing
key. In another site, a listener with a matching routing key makes it
available for application connections. Only those sites with a
listener can connect to the service.</p>
<!-- V2 also has a new approach to exposing pods in another namespace. -->

<!-- AttachedConnector and AttachedConnectorAnchor. A better security -->

<!-- model. -->

matching listener can connect to the service.</p>
<h2 id="a-new-controller-and-cli">A new controller and CLI</h2>
<p>The new controller combines the previous service and site controllers
into one that can be deployed at cluster or namespace scope. The
Expand All @@ -152,39 +144,27 @@ <h2 id="a-new-controller-and-cli">A new controller and CLI</h2>
configuration changes without requiring re-creation of the site.
Notably, you can reconfigure your site without losing existing
site-to-site links.</p>
<p>The new CLI closely follows the API, and indeed the CLI is really just
a thin layer on top of the API. To simplify its use, the CLI blocks
until operations are complete.</p>
<!-- XXX use of CLI to generate YAML -->

<!-- ## Router improvements -->

<!-- XXX -->

<!-- A new, faster TCP adaptor. -->

<!-- Cut through. Input IO thread directly to output IO thread, without going through a third coordinating thread. -->

<!-- Protocol observers built in to said adaptor. -->

<!-- Lower latency, and lower CPU utilization. -->

<!-- No more dedicated HTTP adaptors. This simplifies our work and makes the code easier to maintain. -->
<p>The new CLI closely follows the API. Indeed, in v2 the CLI is really
just a thin layer on top of the API. To simplify its use, the CLI
blocks until operations are complete.</p>
<h2 id="router-improvements">Router improvements</h2>
<p>The router in v2 has a new, faster TCP adaptor with improved buffer
handling and reduced threading overhead. The new TCP adaptor
incorporates lightweight protocol observers for capturing HTTP traffic
metrics. Together these reduce application latency and router CPU
utilization.</p>
<!-- In v1, HA for routers was -->

<!-- HA router configuration -->

<!-- ## Docker, Podman, and Systemd sites -->

<!-- XXX -->

<!-- Simpler and more uniform. -->

<!-- Multiple sites per single user -->

<!-- Uses the standard resources and API. -->

<!-- Gateways go away. -->
<!-- - HA routers! -->

<h2 id="nonkube-sites">Non-Kube sites</h2>
<p>Skupper is not just for Kubernetes. Skupper sites can run on Docker,
Podman, VMs, or bare metal. In v2, we've made the support for
non-Kube sites simpler and more uniform. They use the same YAML
resources as Kube sites. One codebase implements support for all of
the non-Kube platforms.</p>
<!-- ## The observability components stand apart -->

<!-- Deployment is separate from that of sites. -->
Expand All @@ -195,47 +175,33 @@ <h2 id="a-new-controller-and-cli">A new controller and CLI</h2>

<!-- OpenShift site console plugin -->

<!-- - CRDs! -->

<!-- - Uniform model. Declarative API. Everything goes through the CRDs. -->

<!-- - New CLI that follows the new model. The CLI isn't doing anything clever. It's just a convenient tool for producing CRs. -->

<!-- - Same model and CRDs across site types (Kubernetes, Docker, Podman, and Systemd sites) -->

<!-- - GitOps -->

<!-- - Integrations -->

<!-- - A uniform model and API across platforms and interfaces. -->

<!-- - Service exposure model! -->

<!-- - (?) Attached connectors - Tracking pods in namespaces other than that of the site -->

<!-- - No more service sync -->
<!-- - Gordon's preso -->

<!-- - A new controller impl. -->
<!-- - My planning docs -->

<!-- - Combine site and service controllers. -->
<!-- - Observability decoupled - flexible deployment -->

<!-- - Avoiding site recreation! -->
<!-- ## Important to know -->

<!-- - HA routers! -->
<!-- Gateways go away. -->

<!-- - Faster routers! -->
<!-- 1.x is _not_ backward compatible with 2. -->

<!-- - Gordon's preso -->
<!-- We are developing tooling to migrate 1.x config to 2.x config. -->

<!-- - My planning docs -->
<!-- stateful sets! -->

<!-- - Observability decoupled -->
<!-- Multiple sites per single user -->

<!-- ## Important to know -->
<!-- V2 also has a new approach to exposing pods in another namespace. -->

<!-- 1.x is _not_ backward compatible with 2. -->
<!-- AttachedConnector and AttachedConnectorAnchor. A better security -->

<!-- We are developing tooling to migrate 1.x config to 2.x config. -->
<!-- model. -->

<!-- | 1.x | 2.x | -->

Expand Down
71 changes: 27 additions & 44 deletions input/v2/index.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -32,8 +32,6 @@ something v1 did not require. This is an advantage for some of our
users, but a disadvantage for others. We believe that custom
resources are, on balance, the right choice.

<!-- ref the adr -->

[custom-resources]: https://kubernetes.io/docs/concepts/extend-kubernetes/api-extension/custom-resources/
[rbac]: https://kubernetes.io/docs/reference/access-authn-authz/rbac/

Expand Down Expand Up @@ -64,14 +62,10 @@ tools and to provide a foundation for third-party integrations.
Service exposure in particular sees a change in v2. In v1, service
exposure is implicit: exposing a service in one site by default
exposed it in all the linked sites. In v2, service exposure is
instead explicit. A connector binds a local workload to a routing
instead *explicit*. A connector binds a local workload to a routing
key. In another site, a listener with a matching routing key makes it
available for application connections. Only those sites with a
listener can connect to the service.

<!-- V2 also has a new approach to exposing pods in another namespace. -->
<!-- AttachedConnector and AttachedConnectorAnchor. A better security -->
<!-- model. -->
matching listener can connect to the service.

## A new controller and CLI

Expand All @@ -84,32 +78,29 @@ configuration changes without requiring re-creation of the site.
Notably, you can reconfigure your site without losing existing
site-to-site links.

The new CLI closely follows the API, and indeed the CLI is really just
a thin layer on top of the API. To simplify its use, the CLI blocks
until operations are complete.
The new CLI closely follows the API. Indeed, in v2 the CLI is really
just a thin layer on top of the API. To simplify its use, the CLI
blocks until operations are complete.

<!-- XXX use of CLI to generate YAML -->
## Router improvements

<!-- ## Router improvements -->

<!-- XXX -->
<!-- A new, faster TCP adaptor. -->
<!-- Cut through. Input IO thread directly to output IO thread, without going through a third coordinating thread. -->
<!-- Protocol observers built in to said adaptor. -->
<!-- Lower latency, and lower CPU utilization. -->

<!-- No more dedicated HTTP adaptors. This simplifies our work and makes the code easier to maintain. -->
The router in v2 has a new, faster TCP adaptor with improved buffer
handling and reduced threading overhead. The new TCP adaptor
incorporates lightweight protocol observers for capturing HTTP traffic
metrics. Together these reduce application latency and router CPU
utilization.

<!-- In v1, HA for routers was -->
<!-- HA router configuration -->
<!-- - HA routers! -->

<!-- ## Docker, Podman, and Systemd sites -->

<!-- XXX -->
<!-- Simpler and more uniform. -->
<!-- Multiple sites per single user -->
<!-- Uses the standard resources and API. -->
## Non-Kube sites

<!-- Gateways go away. -->
Skupper is not just for Kubernetes. Skupper sites can run on Docker,
Podman, VMs, or bare metal. In v2, we've made the support for
non-Kube sites simpler and more uniform. They use the same YAML
resources as Kube sites. One codebase implements support for all of
the non-Kube platforms.

<!-- ## The observability components stand apart -->

Expand All @@ -120,34 +111,26 @@ until operations are complete.
<!-- Cert reloading -->
<!-- OpenShift site console plugin -->

<!-- - CRDs! -->
<!-- - Uniform model. Declarative API. Everything goes through the CRDs. -->
<!-- - New CLI that follows the new model. The CLI isn't doing anything clever. It's just a convenient tool for producing CRs. -->
<!-- - Same model and CRDs across site types (Kubernetes, Docker, Podman, and Systemd sites) -->
<!-- - GitOps -->
<!-- - Integrations -->
<!-- - A uniform model and API across platforms and interfaces. -->

<!-- - Service exposure model! -->
<!-- - (?) Attached connectors - Tracking pods in namespaces other than that of the site -->
<!-- - No more service sync -->

<!-- - A new controller impl. -->
<!-- - Combine site and service controllers. -->
<!-- - Avoiding site recreation! -->

<!-- - HA routers! -->
<!-- - Faster routers! -->

<!-- - Gordon's preso -->
<!-- - My planning docs -->

<!-- - Observability decoupled -->
<!-- - Observability decoupled - flexible deployment -->

<!-- ## Important to know -->

<!-- Gateways go away. -->
<!-- 1.x is _not_ backward compatible with 2. -->
<!-- We are developing tooling to migrate 1.x config to 2.x config. -->
<!-- stateful sets! -->

<!-- Multiple sites per single user -->
<!-- V2 also has a new approach to exposing pods in another namespace. -->
<!-- AttachedConnector and AttachedConnectorAnchor. A better security -->
<!-- model. -->

<!-- | 1.x | 2.x | -->
<!-- |-|-| -->
Expand Down

0 comments on commit f47093d

Please sign in to comment.