skip to Main Content

Today we’re happy to release Linkerd 0.9.0, our best release yet! This release is jam packed with internal efficiency upgrades and major improvements to the admin dashboard. We also took this opportunity to make some backwards incompatible changes to simplify Linkerd configuration. See the bottom of this post for a detailed guide on what changes you’ll need to make to your config to upgrade from 0.8.* to 0.9.0.

You can download Linkerd 0.9.0 from github or find images on dockerhub. To learn about all the great stuff included in 0.9.0, read on!


We’ve added some bar charts and stats to the admin dashboard to give you further visibility into the performance of your services:

  • Added a retries stat to the router’s summary, so you can see at a glance if something is wrong with an underlying service, causing Linkerd to retry requests.
  • Added a retries bar chart per router which shows the percentage of your configured retry budget that has been used. (The default budget is 20%.)
  • Added a client pool health bar, showing the ratio of live endpoints to total endpoints.

We’ve also made some appearance tweaks to make the dashboard easier to consume:

  • Clients now have a colored border to make them easier to distinguish.
  • Long transformer prefixes have been hidden! To see the full label, click on the client name.
  • Collapsing a client will hide it from the top client requests graph.


Naming and routing are some of the most complex aspects to configuring Linkerd, especially for new users. To simplify this, we’re changing the default identifier to produce names like /svc/foo rather than /http/1.1/GET/foo. These shorter names are easier to understand and to write dtabs for.

We recommend updating your dtabs to use the simpler /svc style names. If you don’t want to do that immediately, the previous default io.l5d.methodAndHost identifier can still be configured explicitly.


We’ve done a lot of work to change the Prometheus metrics output to take advantage of Prometheus’s tags, and to better fit Prometheus metrics naming conventions. As part of this change, the /admin/metrics/prometheus endpoint is no longer provided by default. To get this endpoint you need to add the io.l5d.prometheus telemeter to your config.


Configuration-specific data has moved from metric names to labels. This should minimize the number of metric names while still providing granular breakdowns of metrics across various configurations. For example:



Prior to 0.9.0, Linkerd exported histograms as collections of gauges with a stat label. Linkerd now exports histograms as Prometheus summaries. For example:



To continue our never-ending quest to improve Linkerd, we need a broad picture of how users are running it. To this end, 0.9.0 includes some basic anonymized usage reporting. We’ve been careful to capture only non-identifying information, and to make it easy for you to disable this feature.

Linkerd captures:

  • How it is configured (kinds of namers, initializers, identifiers, transformers, protocols, & interpreters used)
  • What environments it is running in (OS, orchestrator, etc),
  • Performance metrics

It do not capture the labels of namers/routers, designated service addresses or directories, dtabs, or any request or response data.

To review the payload reported at any point, visit :9990/admin/metrics/usage. To disable reporting, simply set enabled: false in your Linkerd config under the top-level usage: section:

You can also optionally provide an organization ID string that will help us to identify your organization if you so choose:


Follow these steps to upgrade an 0.8.* config into one that will work with 0.9.0:

  • The dtabBase field to has been renamed to just dtab.
  • The io.l5d.commonMetrics telemeter no longer exists and should be removed from configs. Metrics will continue to be served on /admin/metrics.json without requiring that they be enabled via the io.l5d.commonMetrics telemeter.
  • The tracers section has been removed in favor of the telemetry section. The only tracer that was previously provided was the io.l5d.zipkin tracer. That configuration can be moved directly to the telemetry section; none of its configuration options have changed.
  • The default dstPrefix has changed from the protocol name (e.g. /http/thrift, etc.) to simply /svc. To get the old behavior you’ll need to manually set thedstPrefix to the protocol name. E.g. dstPrefix: /http.

    Alternatively, update your dtab to expect names starting with /svc instead of with the protocol name. E.g. replace /http/foo => ... with /svc/foo => ....

  • The default HTTP identifier has changed from io.l5d.methodAndHost to io.l5d.header.token. To get the old behavior you’ll need to manually set the identifier to io.l5d.methodAndHost. Alternatively, update your dtab to expect names of the form /svc/<host> instead of /http/1.1/<method>/<host>.

    See the section above on Simpler Logical Names to learn about the motivation for these two changes.



Thank you to all of our users, the wonderful Linkerd community, and an extra special thank you to Borys PierovMark Eijsermans, and Don Petersen for their contributions to this release.

If you run into any issues whatsoever porting your Linkerd setup to use 0.9.0, don’t hesitate to hop into the Linkerd community Slack, and we’ll help you get it sorted out.