linkerd2/CHANGES.md

v0.4.4

Conduit v0.4.4 continues to improve production suitability and sets up internals for the upcoming v0.5.0 release.

• Production Readiness
  • The destination service has been mostly-rewritten to improve safety and correctness, especially during controller initialization.
  • Readiness and Liveness checks have been added for some controller components.
  • RBAC settings have been expanded so that Prometheus can access node-level metrics.
• User Interface
  • Ad blockers like uBlock prevented the Conduit dashboard from fetching API data. This has been fixed.
  • The UI now highlights pods that have failed to start a proxy.
• Internals
  • Various dependency upgrades, including Rust 1.26.2.
  • TLS testing continues to bear fruit, precipitating stability improvements to dependencies like Rustls.

Special thanks to @alenkacz for improving docker build times!

v0.4.3

Conduit v0.4.3 continues progress towards production readiness. It features a new latency-aware load balancer.

• Production Readiness
  • The proxy now uses a latency-aware load balancer for outbound requests. This implementation is based on Finagle's Peak-EWMA balancer, which has been proven to significantly reduce tail latencies. This is the same load balancing strategy used by Linkerd.
• User Interface
  • conduit stat is now slightly more predictable in the way it outputs things, especially for commands like watch conduit stat all --all-namespaces.
  • Failed and completed pods are no longer shown in stat summary results.
• Internals
  • The proxy now supports some TLS configuration, though these features remain disabled and undocumented pending further testing and instrumentation.

Special thanks to @ihcsim for contributing his first PR to the project and to @roanta for discussing the Peak-EWMA load balancing algorithm with us.

v0.4.2

Conduit v0.4.2 is a major step towards production readiness. It features a wide array of fixes and improvements for long-running proxies, and several new telemetry features. It also lays the groundwork for upcoming releases that introduce mutual TLS everywhere.

• Production Readiness
  • The proxy now drops metrics that do not update for 10 minutes, preventing unbounded memory growth for long-running processes.
  • The proxy now constrains the number of services that a node can route to simultaneously (default: 100). This protects long-running proxies from consuming unbounded resources by tearing down the longest-idle clients when the capacity is reached.
  • The proxy now properly honors HTTP/2 request cancellation.
  • The proxy could incorrectly handle requests in the face of some connection errors. This has been fixed.
  • The proxy now honors DNS TTLs.
  • conduit inject now works with statefulset resources.
• Telemetry
  • New conduit stat now supports the all Kubernetes resource, which shows traffic stats for all Kubernetes resources in a namespace.
  • New the Conduit web UI has been reorganized to provide namespace overviews.
  • Fix a bug in Tap that prevented the proxy from simultaneously satisfying more than one Tap request.
  • Fix a bug that could prevent stats from being reported for some TCP streams in failure conditions.
  • The proxy now measures response latency as time-to-first-byte.
• Internals
  • The proxy now supports user-friendly time values (e.g. 10s) from environment configuration.
  • The control plane now uses client for Kubernetes 1.10.2.
  • Much richer proxy debug logging, including socket and stream metadata.
  • The proxy internals have been changed substantially in preparation for TLS support.

Special thanks to @carllhw, @kichristensen, & @sfroment for contributing to this release!

Upgrading from v0.4.1

When upgrading from v0.4.1, we suggest that the control plane be upgraded to v0.4.2 before injecting application pods to use v0.4.2 proxies.

v0.4.1

Conduit 0.4.1 builds on the telemetry work from 0.4.0, providing rich, Kubernetes-aware observability and debugging.

• Web UI
  • New Automatically-configured Grafana dashboards for Services, Pods, ReplicationControllers, and Conduit mesh health.
  • New conduit dashboard Pod and ReplicationController views.
• Command-line interface
  • Breaking change conduit tap now operates on most Kubernetes resources.
  • conduit stat and conduit tap now both support kubectl-style resource strings (deploy, deploy/web, and deploy web), specifically:
    • namespaces
    • deployments
    • replicationcontrollers
    • services
    • pods
• Telemetry
  • New Tap support for filtering by and exporting destination metadata. Now you can sample requests from A to B, where A and B are any resource or group of resources.
  • New TCP-level stats, including connection counts and durations, and throughput, wired through to Grafana dashboards.
• Service Discovery
  • The proxy now uses the trust-dns DNS resolver. This fixes a number of DNS correctness issues.
  • The Destination service could sometimes return incorrect, stale, labels for an endpoint. This has been fixed!

v0.4.0

Conduit 0.4.0 overhauls Conduit's telemetry system and improves service discovery reliability.

• Web UI
  • New automatically-configured Grafana dashboards for all Deployments.
• Command-line interface
  • conduit stat has been completely rewritten to accept arguments like kubectl get. The --to and --from filters can be used to filter traffic by destination and source, respectively. conduit stat currently can operate on Namespace and Deployment Kubernetes resources. More resource types will be added in the next release!
• Proxy (data plane)
  • New Prometheus-formatted metrics are now exposed on :4191/metrics, including rich destination labeling for outbound HTTP requests. The proxy no longer pushes metrics to the control plane.
  • The proxy now handles SIGINT or SIGTERM, gracefully draining requests until all are complete or SIGQUIT is received.
  • SMTP and MySQL (ports 25 and 3306) are now treated as opaque TCP by default. You should no longer have to specify --skip-outbound-ports to communicate with such services.
  • When the proxy reconnected to the controller, it could continue to send requests to old endpoints. Now, when the proxy reconnects to the controller, it properly removes invalid endpoints.
  • A bug impacting some HTTP/2 reset scenarios has been fixed.
• Service Discovery
  • Previously, the proxy failed to resolve some domain names that could be misinterpreted as a Kubernetes Service name. This has been fixed by extending the Destination API with a negative acknowledgement response.
• Control Plane
  • The Telemetry service and associated APIs have been removed.
• Documentation
  • Updated Roadmap

Special thanks to @ahume, @alenkacz, & @xiaods for contributing to this release!

Upgrading from v0.3.1

When upgrading from v0.3.1, it's important to upgrade proxies before upgrading the controller. As you upgrade proxies, the controller will lose visibility into some data plane stats. Once all proxies are updated, conduit install |kubectl apply -f - can be run to upgrade the controller without causing any data plane disruptions. Once the controller has been restarted, traffic stats should become available.

v0.3.1

Conduit 0.3.1 improves Conduit's resilience and transparency.

• Proxy (data plane)
  • The proxy now makes fewer changes to requests and responses being proxied. In particular, requests and responses without bodies or with empty bodies are better supported.
  • HTTP/1 requests with different Host header fields are no longer sent on the same HTTP/1 connection even when those hostnames resolve to the same IP address.
  • A connection leak during proxying of non-HTTP TCP connections was fixed.
  • The proxy now handles unavailable services more gracefully by timing out while waiting for an endpoint to become available for the service.
• Command-line interface
  • $KUBECONFIG with multiple paths is now supported. (PR #482 by @hypnoglow).
  • conduit check now checks for the availability of a Conduit update. (PR #460 by @ahume).
• Service Discovery
  • Kubernetes services with type ExternalName are now supported.
• Control Plane
  • The proxy is injected into the control plane during installation to improve the control plane's resilience and to "dogfood" the proxy.
  • The control plane is now more resilient regarding networking failures.
• Documentation
  • The markdown source for the documentation published at https://conduit.io/docs/ is now open source at https://github.com/runconduit/conduit/tree/master/doc.

v0.3.0

Conduit 0.3 focused heavily on production hardening of Conduit's telemetry system. Conduit 0.3 should "just work" for most apps on Kubernetes 1.8 or 1.9 without configuration, and should support Kubernetes clusters with hundreds of services, thousands of instances, and hundreds of RPS per instance.

With this release, Conduit also moves from experimental to alpha---meaning that we're ready for some serious testing and vetting from you. As part of this, we've published the Conduit roadmap, and we've also launched some new mailing lists: conduit-users, conduit-dev, and conduit-announce.

• CLI
  • CLI commands no longer depend on kubectl
  • conduit dashboard now runs on an ephemeral port, removing port 8001 conflicts
  • conduit inject now skips pods with hostNetwork=true
  • CLI commands now have friendlier error messages, and support a --verbose flag for debugging
• Web UI
  • All displayed metrics are now instantaneous snapshots rather than aggregated over 10 minutes
  • The sidebar can now be collapsed
  • UX refinements and bug fixes
• Conduit proxy (data plane)
  • Proxy does load-aware (P2C + least-loaded) L7 balancing for HTTP
  • Proxy can now route to external DNS names
  • Proxy now properly sheds load in some pathological cases when it cannot route
• Telemetry system
  • Many optimizations and refinements to support scale goals
  • Per-path and per-pod metrics have been removed temporarily to improve scalability and stability; they will be reintroduced in Conduit 0.4 (#405)
• Build improvements
  • The Conduit docker images are now much smaller.
  • Dockerfiles have been changed to leverage caching, improving build times substantially

Known Issues:

• Some DNS lookups to external domains fail (#62, #155, #392)
• Applications that use WebSockets, HTTP tunneling/proxying, or protocols such as MySQL and SMTP, require additional configuration (#339)

v0.2.0

This is a big milestone! With this release, Conduit adds support for HTTP/1.x and raw TCP traffic, meaning it should "just work" for most applications that are running on Kubernetes without additional configuration.

• Data plane
  • Conduit now transparently proxies all TCP traffic, including HTTP/1.x and HTTP/2. (See caveats below.)
• Command-line interface
  • Improved error handling for the tap command
  • tap also now works with HTTP/1.x traffic
• Dashboard
  • Minor UI appearance tweaks
  • Deployments now searchable from the dashboard sidebar

Caveats:

• Conduit will automatically work for most protocols. However, applications that use WebSockets, HTTP tunneling/proxying, or protocols such as MySQL and SMTP, will require some additional configuration. See the documentation for details.
• Conduit doesn't yet support external DNS lookups. These will be addressed in an upcoming release.
• There are known issues with Conduit's telemetry pipeline that prevent it from scaling beyond a few nodes. These will be addressed in an upcoming release.
• Conduit is still in alpha! Please help us by filing issues and contributing pull requests.

v0.1.3

• This is a minor bugfix for some web dashboard UI elements that were not rendering correctly.

v0.1.2

Conduit 0.1.2 continues down the path of increasing usability and improving debugging and introspection of the service mesh itself.

• Conduit CLI
  • New conduit check command reports on the health of your Conduit installation.
  • New conduit completion command provides shell completion.
• Dashboard
  • Added per-path metrics to the deployment detail pages.
  • Added animations to line graphs indicating server activity.
  • More descriptive CSS variable names. (Thanks @natemurthy!)
  • A variety of other minor UI bugfixes and improvements
• Fixes
  • Fixed Prometheus config when using RBAC. (Thanks @FaKod!)
  • Fixed tap failure when pods do not belong to a deployment. (Thanks @FaKod!)

v0.1.1

Conduit 0.1.1 is focused on making it easier to get started with Conduit.

• Conduit can now be installed on Kubernetes clusters that use RBAC.
• The conduit inject command now supports a --skip-outbound-ports flag that directs Conduit to bypass proxying for specific outbound ports, making Conduit easier to use with non-gRPC or HTTP/2 protocols.
• The conduit tap command output has been reformatted to be line-oriented, making it easier to parse with common UNIX command line utilities.
• Conduit now supports routing of non-fully qualified domain names.
• The web UI has improved support for large deployments and deployments that don't have any inbound/outbound traffic.

v0.1.0

Conduit 0.1.0 is the first public release of Conduit.

• This release supports services that communicate via gRPC only. non-gRPC HTTP/2 services should work. More complete HTTP support, including HTTP/1.0 and HTTP/1.1 and non-gRPC HTTP/2, will be added in an upcoming release.
• Kubernetes 1.8.0 or later is required.
• kubectl 1.8.0 or later is required. conduit dashboard will not work with earlier versions of kubectl.
• When deploying to Minikube, Minikube 0.23 or 0.24.1 or later are required. Earlier versions will not work.
• This release has been tested using Google Kubernetes Engine and Minikube. Upcoming releases will be tested on additional providers too.
• Configuration settings and protocols are not stable yet.
• Services written in Go must use grpc-go 1.3 or later to avoid grpc-go bug #1120.