# Since 0.12.0

# New

# Upstream load balancing

With the v0.13 release, routes may contain multiple to URLs, and Pomerium will load balance between the endpoints. This allows Pomerium to fill the role of an edge proxy without the need for additional HTTP load balancers.

Active health checks and passive outlier detection
Configurable load balancing policies
Configurable load balancing weight

See Load Balancing for more information on using this feature set.

# Dynamic certificate updates

With the v0.13 release, all TLS files referenced from Pomerium's configuration are reloaded automatically when updating. This improves availability in environments which automate short lived TLS certificate rotation via certbot (opens new window) or similar tools.

# Proxy Protocol support

The Pomerium HTTP listener now supports HAPROXY's proxy protocol (opens new window) to update X-Forwarded-For accurately when behind another proxy service.

# Breaking

# Sign-out endpoint requires CSRF Token

The frontchannel-logout endpoint will now require a CSRF token for both GET and POST requests.

# User impersonation removed

Prior to the v0.13 release, it was possible for an administrative user to temporarily impersonate another user. This was done by adding an additional set of claims to that user's session token. Having additional identity state stored client-side significantly expands the attack surface of Pomerium and complicates policy enforcement by having multiple sources of truth for identity. User impersonation was removed to shrink that attack surface and simplify policy enforcement. Pomerium now stores all identity state server-side and encrypted in the databroker.

# Administrators option removed

The administrators configuration option has been removed.

# Since 0.11.0

# New

# TCP Proxying

Pomerium can now be used for non-HTTP services. See documentation for more details.

# Datadog Tracing

Datadog has been added as a natively supported tracing backend

# Since 0.10.0

# Breaking

# User impersonation disabled by default

With the v0.11.0 release, the ability to do user user impersonation is disabled by default. To enable user impersonation, set enable_user_impersonation to true in the configuration options.

# `cache_service_url` has been renamed to `databroker_service_url`

The cache_service_url parameter has been deprecated since v0.10.0 and is now removed. Please replace it with databroker_service_url in your yaml configuration, or DATABROKER_SERVICE_URL as an environment variable.

# New

# Docker Multi-Arch Images

With the v0.11.0 release, Pomerium docker images are multi-arch for arm64 and amd64. Individual images for each architecture will continue to be published.

# Since 0.9.0

# Breaking

# Service accounts required for groups and directory data

With the v0.10.0 release, Pomerium now queries group information asynchronously using a service account. While a service account was already required for a few identity providers like Google's GSuite, an Identity Provider Service Account is now required for all other providers as well. The format of this field varies and is specified in each identity provider's documentation.

WARNING

If no Identity Provider Service Account is supplied, policies using groups (e.g. allowed_groups will not work).

# Cache service builds stateful context

With the v0.10 release, Pomerium now asynchronously fetches associated authorization context (e.g. identity provider directory context, groups, user-data, session data, etc) in the cache service. In previous versions, Pomerium used session cookies to associated identity state which authorization policy was evaluated against. While using session tokens had the advantage of making Pomerium a relatively stateless application, that approach has many shortcomings which is more extensively covered in the data storage docs.

There are two storage backend types available: memory or redis. You can see the existing storage backend configuration settings in the docs.

# Memory Storage Backend

For memory storage, restarting the cache service will result in all users having to re-login. Code for the in-memory database used by the cache service can be found here: internal/databroker/memory (opens new window).

WARNING

Running more than one instance of the memory type cache service is not supported.

# Redis Storage Backend

In production deployments, we recommend using the redis storage backend. Unlike the memory backend, redis can be used for persistent data.

# Implementing your own storage backend

Please see the following interfaces for reference to implement your storage backend interface.

# Identity headers

With this release, pomerium will not insert identity headers (X-Pomerium-Jwt-Asserttion/X-Pomerium-Claim-*) by default. To get pre 0.9.0 behavior, you can set pass_identity_headers to true on a per-policy basis.

# Since 0.8.0

# Breaking

# Default log level

With this release, default log level has been changed to INFO.

# HTTP 1.0

HTTP 1.0 (not to be confused with HTTP 1.1) is not supported anymore. If you relied on it make sure to upgrade to HTTP 1.1 or higher.

Example for HAProxy health check, in pre 0.9.0:

shell script option httpchk GET /ping

In 0.9.0:

option httpchk GET /ping HTTP/1.1\r\nHost:pomerium

# preserve_host_header option

With this release, Pomerium uses an embedded envoy proxy instead hand-written one. Thus, we defer the preserve host header functionality to envoys auto_host_rewrite (opens new window), which does not affect if the policy routes to a static IP.

To preserve 0.8.x behavior, you can use the set_request_headers option to explicitly set the Host header.

# Unsupported platforms

With this release we now use an embedded envoy (opens new window) binary as our proxy server. Due to this change we now only build and support Linux and MacOS binaries with the AMD64 architecture. We plan on supporting more platforms and architectures in future releases.

# Observability

The service label on metrics and tracing no longer reflects the Services configuration option directly. pomerium will be used for all-in-one mode, and pomerium-[service] will be used for distributed services

# Tracing

Jaeger tracing support is no longer end-to-end in the proxy service. We recommend updating to the Zipkin provider for proper tracing support. Jaeger will continue to work but will not have coverage in the data plane.
Option tracing_debug is no longer supported. Use tracing_sampling_rate instead. Details (opens new window).

# Metrics

With this release we now use an embedded envoy (opens new window) binary as our proxy server.

Due to this change, data plane metric names and labels have changed to adopt envoy's internal data model. Details (opens new window)

# Since 0.7.0

# Breaking

# Using paths in from URLs

Although it's unlikely anyone ever used it, prior to 0.8.0 the policy configuration allowed you to specify a from field with a path component:

policy:
  - from: "https://example.com/some/path"

The proxy and authorization server would simply ignore the path and route/authorize based on the host name.

With the introduction of prefix, path and regex fields to the policy route configuration, we decided not to support using a path in the from url, since the behavior was somewhat ambiguous and better handled by the explicit fields.

To avoid future confusion, the application will now declare any configuration which contains a from field with a path as invalid, with this error message:

config: policy source url (%s) contains a path, but it should be set using the path field instead

If you see this error you can fix it by simply removing the path from the from field and moving it to a prefix field.

In other words, this configuration:

policy:
  - from: "http://example.com/some/path"

Should be written like this:

policy:
  - from: "http://example.com"
    prefix: "/some/path"

# Since 0.6.0

# Breaking

# Getting user's identity

WARNING

This changed was partially reverted in v0.7.2. Session details like user, email, and groups can still be explicitly extracted by setting the jwt_claims_header configuration option.

User detail headers ( x-pomerium-authenticated-user-id / x-pomerium-authenticated-user-email / x-pomerium-authenticated-user-groups) have been removed in favor of using the more secure, more data rich attestation jwt header (x-pomerium-jwt-assertion).

If you still rely on individual claim headers, please see the jwt_claims_headers option here (opens new window).

# Non-standard port users

Non-standard port users (e.g. those not using 443/80 where the port would be part of the client's request) will have to clear their user's session before upgrading. Starting with version v0.7.0, audience (aud) and issuer (iss) claims will be port specific.

# Since 0.5.0

# Breaking

# New cache service

A back-end cache service was added to support session refreshing from single-page-apps (opens new window).

For all-in-one deployments, no changes are required. The cache will be embedded in the binary. By default, autocache an in-memory LRU cache will be used to temporarily store user session data. If you wish to persist session data, it's also possible to use bolt or redis.
For split-service deployments, you will need to deploy an additional service called cache. By default, pomerium will use autocache as a distributed, automatically managed cache. It is also possible to use redis as backend in this mode.

For a concrete example of the required changes, consider the following changes for those running split service mode,:

...
  pomerium-authenticate:
    environment:
      - SERVICES=authenticate
+      - CACHE_SERVICE_URL=http://pomerium-cache:443
...
+  pomerium-cache:
+    image: pomerium/pomerium
+    environment:
+      - SERVICES=cache
+    volumes:
+      - ../config/config.example.yaml:/pomerium/config.yaml:ro
+    expose:
+      - 443

Please see the updated examples, and cache service docs as a reference and for the available cache stores. For more details as to why this was necessary, please see PR438 (opens new window) and PR457 (opens new window).

# Since 0.4.0

# Breaking

# Subdomain requirement dropped

Pomerium services and managed routes are no longer required to be on the same domain-tree root. Access can be delegated to any route, on any domain (that you have access to, of course).

# Azure AD

Azure Active Directory now uses the globally unique and immutableID instead of group name to attest a user's group membership (opens new window). Please update your policies to use group ID instead of group name.

# Okta

Okta no longer uses tokens to retrieve group membership. Group membership (opens new window) is now fetched using Okta's API.
Okta's group membership is now determined by the globally unique and immutable ID field. Please update your policies to use group ID instead of group name.
Okta now requires an additional set of credentials to be used to query for group membership set as a service account (opens new window).

# OneLogin

OneLogin group membership (opens new window) is now determined by the globally unique and immutable ID field. Please update your policies to use group ID instead of group name.

# Force Refresh Removed

Force refresh has been removed from the dashboard. Logging out and back in again should have the equivalent desired effect.

# Programmatic Access API changed

Previous programmatic authentication endpoints (/api/v1/token) has been removed and has been replaced by a per-route, oauth2 based auth flow. Please see updated programmatic documentation (opens new window) how to use the new programmatic access api.

# Forward-auth route change

Previously, routes were verified by taking the downstream applications hostname in the form of a path (e.g. ${forwardauth}/.pomerium/verify/verify.some.example) variable. The new method for verifying a route using forward authentication is to pass the entire requested url in the form of a query string (e.g. ${forwardauth}/.pomerium/verify?url=https://verify.some.example) where the routed domain is the value of the uri key.

Note that the verification URL is no longer nested under the .pomerium endpoint.

For example, in nginx this would look like:

-    nginx.ingress.kubernetes.io/auth-url: https://forwardauth.corp.example.com/.pomerium/verify/verify.corp.example.com?no_redirect=true
-    nginx.ingress.kubernetes.io/auth-signin: https://forwardauth.corp.example.com/.pomerium/verify/verify.corp.example.com
+    nginx.ingress.kubernetes.io/auth-url: https://forwardauth.corp.example.com/verify?uri=$scheme://$host$request_uri
+    nginx.ingress.kubernetes.io/auth-signin: https://forwardauth.corp.example.com?uri=$scheme://$host$request_uri

# Since 0.3.0

# Breaking

# Authorize Service URL no longer used in all-in-one mode

Pomerium no longer handles both gRPC and HTTPS traffic from the same network listener (port). As a result, all-in-one mode configurations will default to serving gRPC traffic over loopback on port 5443 and will serve HTTPS traffic as before on port 443. In previous versions, it was recommended to configure authorize in this mode which will now break. The error will typically look something like:

rpc error: code = DeadlineExceeded desc = latest connection error: connection closed

To upgrade, simply remove the AUTHORIZE_SERVICE_URL setting.

# Removed Authenticate Internal URL

The authenticate service no longer uses gRPC to do back channel communication. As a result, AUTHENTICATE_INTERNAL_URL/authenticate_internal_url is no longer required.

# No default certificate location

In previous versions, if no explicit certificate pair (in base64 or file form) was set, Pomerium would make a last ditch effort to check for certificate files (cert.key/privkey.pem) in the root directory. With the introduction of insecure server configuration, we've removed that functionality. If there settings for certificates and insecure server mode are unset, pomerium will give a appropriate error instead of a failed to find/open certificate error.

# Authorize service health-check is non-http

The Authorize service will no longer respond to HTTP-based healthcheck queries when run as a distinct service (vs all-in-one). As an alternative, you can used on TCP based checks. For example, if using Kubernetes (opens new window):

---
readinessProbe:
  tcpSocket:
    port: 443
  initialDelaySeconds: 5
  periodSeconds: 10
livenessProbe:
  tcpSocket:
    port: 443
  initialDelaySeconds: 15
  periodSeconds: 20

# Non-breaking changes

# All-in-one

If service mode (SERVICES/services) is set to all, gRPC communication with the authorize service will by default occur over localhost, on port :5443.

# Since 0.2.0

Pomerium v0.3.0 has no known breaking changes compared to v0.2.0.

# Since 0.1.0

Pomerium v0.2.0 has no known breaking changes compared to v0.1.0.

# Since 0.0.5

This page contains the list of deprecations and important or breaking changes for pomerium v0.1.0 compared to v0.0.5. Please read it carefully.

# Semantic versioning changes

Starting with v0.1.0 we've changed our releases (opens new window) are versioned (MAJOR.MINOR.PATCH+GITHASH). Planned, monthly releases will now bump MINOR and any security or stability releases required prior will bump PATCH.

Please note however that we are still pre 1.0.0 so breaking changes can and will happen at any release though we will do our best to document them.

# Breaking: Policy must be valid URLs

Previously, it was allowable to define a policy without a schema (e.g. http/https). Starting with version v0.1.0 all to and from policy URLS must contain valid schema and host-names. For example:

policy:
  - from: verify.corp.domain.example
    to: http://verify
    allowed_domains:
      - pomerium.io
  - from: external-verify.corp.domain.example
    to: https://verify.pomerium.com
    allow_public_unauthenticated_access: true

Should now be:

policy:
  - from: https://verify.corp.domain.example
    to: http://verify
    allowed_domains:
      - pomerium.io
  - from: https://external-verify.corp.domain.example
    to: https://verify.pomerium.com
    allow_public_unauthenticated_access: true

# Since 0.0.4

This page contains the list of deprecations and important or breaking changes for pomerium v0.0.5 compared to v0.0.4. Please read it carefully.

# Breaking: POLICY_FILE removed

Usage of the POLICY_FILE envvar is no longer supported. Support for file based policy configuration has been shifted into the new unified config file.

# Important: Configuration file support added

Pomerium now supports an optional -config flag. This flag specifies a file from which to read all configuration options. It supports yaml, json, toml and properties formats.
All options which can be specified via MY_SETTING style envvars can now be specified within your configuration file as key/value. The key is generally the same as the envvar name, but lower cased. See Reference Documentation for exact names.
Options precedence is environmental variables > configuration file > defaults

The options file supports a policy key, which contains policy in the same format as POLICY_FILE. To convert an existing policy.yaml into a config.yaml, just move your policy under a policy key.

Old:

- from: verify.localhost.pomerium.io
  to: http://verify
  allowed_domains:
    - pomerium.io
  cors_allow_preflight: true
  timeout: 30s

New:

policy:
  - from: verify.localhost.pomerium.io
    to: http://verify
    allowed_domains:
      - pomerium.io
    cors_allow_preflight: true
    timeout: 30s

# Authenticate Internal Service Address

The configuration variable Authenticate Internal Service URL must now be a valid URL (opens new window) type and contain both a hostname and valid https schema.

← FAQ Changelog →

# Since 0.12.0

# New

# Upstream load balancing

# Dynamic certificate updates

# Proxy Protocol support

# Breaking

# Sign-out endpoint requires CSRF Token

# User impersonation removed

# Administrators option removed

# Since 0.11.0

# New

# TCP Proxying

# Datadog Tracing

# Since 0.10.0

# Breaking

# User impersonation disabled by default

# cache_service_url has been renamed to databroker_service_url

# New

# Docker Multi-Arch Images

# Since 0.9.0

# Breaking

# Service accounts required for groups and directory data

# Cache service builds stateful context

# Memory Storage Backend

# Redis Storage Backend

# Implementing your own storage backend

# Identity headers

# Since 0.8.0

# Breaking

# Default log level

# HTTP 1.0

# preserve_host_header option

# Unsupported platforms

# Observability

# Tracing

# Metrics

# Since 0.7.0

# Breaking

# Using paths in from URLs

# Since 0.6.0

# Breaking

# Getting user's identity

# Non-standard port users

# Since 0.5.0

# Breaking

# New cache service

# Since 0.4.0

# Breaking

# Subdomain requirement dropped

# Azure AD

# Okta

# OneLogin

# Force Refresh Removed

# Programmatic Access API changed

# Forward-auth route change

# Since 0.3.0

# Breaking

# Authorize Service URL no longer used in all-in-one mode

# Removed Authenticate Internal URL

# No default certificate location

# Authorize service health-check is non-http

# Non-breaking changes

# All-in-one

# Since 0.2.0

# Since 0.1.0

# Since 0.0.5

# Semantic versioning changes

# Breaking: Policy must be valid URLs

# Since 0.0.4

# Breaking: POLICY_FILE removed

# Important: Configuration file support added

# Authenticate Internal Service Address

# `cache_service_url` has been renamed to `databroker_service_url`