v1.40.2

🔒 Security

Apply limits.http_max_request_bytes on streaming request body decompression (PR #4759)

This release fixes a Denial-of-Service (DoS) type vulnerability which exists in affected versions of the Router according to our published security advistory. The fix changes the evaluation of the limits.http_max_request_bytes configuration to take place on a stream of bytes, allowing it to be applied to compressed HTTP payloads, prior to decompression. Previously, the limit was only being applied after the entirety of the compressed payload was decompressed, which could result in significant memory consumption which exceeded configured expectations while compressed payloads were expanded.

🐛 Fixes

Re-activate the macOS Intel builder (PR #4723)

We have re-activated macOS Intel (x86) builds in CircleCI, despite their upcoming deprecation, while we take a different approach to solving this and maintaining Intel support for the time-being. This became necessary since cross-compiling the router from ARM to x86 resulted in issues with V8 snapshots and runtime issues on the macOS Intel binaries produced by those Apple Silicon build machines.

By @Geal in #4723

v1.40.2-rc.0

1.40.2-rc.0

v1.41.0-rc.2

1.41.0-rc.2

v1.41.0-rc.1

1.41.0-rc.1

v1.41.0-rc.0

1.41.0-rc.0

v1.40.1

🐛 Fixes

Propagate tracing headers even when not sampling a trace (Issue #4544)

When the router was configured to sample only a portion of the trace, either through a ratio or using parent based sampling, and when trace propagation was configured, if a trace was not sampled, the router did not send the propagation headers to the subgraph. The subgraph was then unable to decide whether to record the trace or not. Now we make sure that trace headers will be sent even when a trace is not sampled.

By @Geal in #4609

Load TLS native certificate store once (Issue #4491)

When TLS was not configured for subgraphs, the OS-provided list of certificates was being parsed once per subgraph, which resulted in long loading times on macOS. With this change, the native root store is generated once and then reused across subgraphs, resolving the long loading times.

By @Geal in #4616

Apollo library dependency updates (Issue #4525, Issue #413)

Updates to our own dependencies of apollo-rs and apollo-federation bring in upstream fixes for operation validation including adjustments to field merging and enum input values for Rust-based validation.

By @goto-bus-stop in #4510

v1.40.1-rc.1

1.40.1-rc.1

v1.40.1-rc.0

1.40.1-rc.0

v1.40.0

🚀 Features

GraphOS entity caching (Issue #4478)

⚠️ This is a preview for an Enterprise feature of the Apollo Router. It requires an organization with a GraphOS Enterprise plan.

If your organization doesn't currently have an Enterprise plan, you can test out this functionality by signing up for a free Enterprise trial.

The Apollo Router can now cache fine-grained subgraph responses at the entity level, which are reusable between requests.

Caching federated GraphQL responses can be done at the HTTP level, but it's inefficient because a lot of data can be shared between different requests. The Apollo Router now contains an entity cache that works at the subgraph level: it caches subgraph responses, splits them by entities, and reuses entities across subgraph requests.
Along with reducing the cache size, the router's entity cache brings more flexibility in how and what to cache. It allows the router to store different parts of a response with different expiration dates, and link the cache with the authorization context to avoid serving stale, unauthorized data.

As a preview feature, it's subject to our Preview launch stage expectations. It doesn't support cache invalidation. We're making it available to test and gather feedback.

By @Geal in #4195

Graduate distributed query plan caching from experimental (Issue #4575)

[Distributed query plan caching] (https://www.apollographql.com/docs/router/configuration/distributed-caching#distributed-query-plan-caching) has been validated in production deployments and is now a fully supported, non-experimental Enterprise feature of the Apollo Router.

To migrate your router configuration, replace supergraph.query_planning.experimental_cache with supergraph.query_planning.cache.

This release also adds improvements to the distributed cache:
1. The . separator is replaced with : in the Redis cache key to align with conventions.
2. The cache key length is reduced.
3. A federation version is added to the cache key to prevent confusion when routers with different federation versions (and potentially different ways to generate a query plan) target the same cache.
4. Cache insertion is moved to a parallel task. Once the query plan is created, this allows a request to be processed immediately instead of waiting for cache insertion to finish. This improvement has also been applied to the APQ cache.

By @Geal in #4583

Replace selector to extract body elements from subgraph responses via JSONPath (Issue #4443)

The subgraph_response_body selector has been deprecated and replaced with selectors for a response body's constituent elements: subgraph_response_data and subgraph_response_errors.

When configuring subgraph_response_data and subgraph_response_errors, both use a JSONPath expression to fetch data or errors from a subgraph response.

An example configuration:

telemetry:
  instrumentation:
    spans:
      subgraph:
        attributes:
          "my_attribute":
            subgraph_response_data: "$.productName"
            subgraph_response_errors: "$.[0].message"

By @bnjjj in #4579

Add a .remove method for headers in Rhai

The router supports a new .remove method that enables users to remove headers in a Rhai script.

For example:

fn supergraph_service(service) {
    print("registering callbacks for operation timing");

    const request_callback = Fn("process_request");
    service.map_request(request_callback);

    const response_callback = Fn("process_response");
    service.map_response(response_callback);
}

fn process_request(request) {
    request.context["request_start"] = Router.APOLLO_START.elapsed;
}

fn process_response(response) {
    response.headers.remove("x-custom-header")
}

By @lrlna in #4632

Helm update to allow a list of gateways to VirtualService (Issue #4464)

Configuration of the router's Helm chart has been updated to allow multiple gateways. This enables configuration of multiple gateways in an Istio VirtualService.

The previous configuration for a single virtualservice.gatewayName has been deprecated in favor of a configuration for an array of virtualservice.gatewayNames.

By @marcantoine-bibeau in #4520

Configure logging format automatically based on terminal (Issue #4369)

You can configure the logging output format when running with an interactive shell.

If both format and tty_format are configured, then the format used depends on how the router is run:

• If running with an interactive shell, then tty_format takes precedence.
• If running with a non-interactive shell, then format takes precedence.

You can explicitly set the format in router.yaml with telemetry.exporters.logging.stdout.tty_format:

telemetry:
  exporters:
     logging:
       stdout:
         enabled: true
         format: json
         tty_format: text

By @bnjjj in #4567

Add configurable histogram buckets per metric (Issue #4543)

The router supports overriding instrument settings for metrics with OpenTelemetry views. You can use views to override default histogram buckets.

Configure views with the views option. For example:

telemetry:
  exporters:
    metrics:
      common:
        service_name: apollo-router
        views:
          - name: apollo_router_http_request_duration_seconds # Instrument name you want to edit. You can use wildcard in names. If you want to target all instruments just use '*'
            unit: "ms" # (Optional) override the unit
            description: "my new description of this metric" # (Optional) override the description
            aggregation: # (Optional)
              histogram:
                buckets: # Override default buckets configured for this histogram
                - 1
                - 2
                - 3
                - 4
                - 5
            allowed_attribute_keys: # (Optional) Keep only listed attributes on the metric
            - status

By @bnjjj in #4572

🐛 Fixes

Fix active_session_count when future is dropped (Issue #4601)

Fixes an issue where apollo_router_session_count_active would increase indefinitely due
to the request future getting dropped before a counter could be decremented.

By @xuorig in #4619

v1.40.0-rc.1

1.40.0-rc.1