🚀 Features
Add support for DHAT-based heap profiling (PR #1829)
The dhat-rs crate provides DHAT-style heap profiling. We have added two compile-time features, dhat-heap and dhat-ad-hoc, which leverage this ability.
Add trace_id in logs to correlate entries from the same request (Issue #1981)
A trace_id is now added to each log line to help correlate log entries to specific requests. The value for this property will be automatically inherited from any enabled distributed tracing headers, such as those listed in our Tracing propagation header documentation (e.g., Jaeger, Zipkin, Datadog, etc.).
In the event that a trace_id was not inherited from a propagated header, the Router will originate a trace_id and also propagate that value to subgraphs to enable tracing in subgraphs.
Here is an example of the trace_id appearing in plain-text log output:
2022-10-21T15:17:45.562553Z ERROR [trace_id=5e6a6bda8d0dca26e5aec14dafa6d96f] apollo_router::services::subgraph_service: fetch_error="hyper::Error(Connect, ConnectError(\"tcp connect error\", Os { code: 111, kind: ConnectionRefused, message: \"Connection refused\" }))"
2022-10-21T15:17:45.565768Z ERROR [trace_id=5e6a6bda8d0dca26e5aec14dafa6d96f] apollo_router::query_planner::execution: Fetch error: HTTP fetch failed from 'accounts': HTTP fetch failed from 'accounts': error trying to connect: tcp connect error: Connection refused (os error 111)
And an exmaple of the trace_id appearing in JSON-formatted log output in a similar scenario:
{"timestamp":"2022-10-26T15:39:01.078260Z","level":"ERROR","fetch_error":"hyper::Error(Connect, ConnectError(\"tcp connect error\", Os { code: 111, kind: ConnectionRefused, message: \"Connection refused\" }))","target":"apollo_router::services::subgraph_service","filename":"apollo-router/src/services/subgraph_service.rs","line_number":182,"span":{"name":"subgraph"},"spans":[{"trace_id":"5e6a6bda8d0dca26e5aec14dafa6d96f","name":"request"},{"name":"supergraph"},{"name":"execution"},{"name":"parallel"},{"name":"fetch"},{"name":"subgraph"}]}
{"timestamp":"2022-10-26T15:39:01.080259Z","level":"ERROR","message":"Fetch error: HTTP fetch failed from 'accounts': HTTP fetch failed from 'accounts': error trying to connect: tcp connect error: Connection refused (os error 111)","target":"apollo_router::query_planner::execution","filename":"apollo-router/src/query_planner/execution.rs","line_number":188,"span":{"name":"parallel"},"spans":[{"trace_id":"5e6a6bda8d0dca26e5aec14dafa6d96f","name":"request"},{"name":"supergraph"},{"name":"execution"},{"name":"parallel"}]}Reload configuration when receiving the SIGHUP signal (Issue #35)
The Router will now reload its configuration when receiving the SIGHUP signal. This signal is only supported on *nix platforms,
and only when a configuration file was passed to the Router initially at startup.
🐛 Fixes
Fix the deduplication logic in deduplication caching (Issue #1984)
Under load, we found it was possible to break the router de-duplication logic and leave orphaned entries in the waiter map. This fixes the de-duplication logic to prevent this from occurring.
Follow back-off instructions from Studio Uplink (Issue #1494 Issue #1539)
When operating in a Managed Federation configuration and fetching the supergraph from Apollo Uplink, the Router will now react differently depending on the response from Apollo Uplink, rather than retrying incessantly:
- Not attempt to retry when met with unrecoverable conditions (e.g., a Graph that does not exist).
- Back-off on retries when the infrastructure asks for a longer retry interval.
Fix the rhai SDL print function (Issue #2005)
Fixes the print function exposed to rhai which was broken due to a recent change that was made in the way we pass SDL (schema definition language) to plugins.
By @fernando-apollo in #2007
Export router_factory::Endpoint (PR #2007)
We now export the router_factory::Endpoint struct that was inadvertently unexposed. Without access to this struct, it was not possible to implement the web_endpoints trait in plugins.
By @scottdouglas1989 in #2007
Validate default values for input object fields (Issue #1979)
When validating variables, the Router now uses graph-specified default values for object fields, if applicable.
Address regression when sending gRPC to localhost (Issue #2036)
We again support sending unencrypted gRPC tracing and metrics data to localhost. This follows-up on a regression which occurred in the previous release which addressed a limitation which prevented sending gRPC to TLS-secured endpoints.
Applying a proper fix was complicated by an upstream issue (opentelemetry-rust#908) which incorrectly assumes https in the absence of a more-specific protocol/schema, contrary to the OpenTelmetry specification which indicates otherwise.
The Router will now detect and work-around this upstream issue by explicitly setting the full, correct endpoint URLs when not specified in config.
In addition:
- Basic TLS-encyrption will be enabled when the endpoint scheme is explicitly
https. - A warning will be emitted if the endpoint port is 443 but no TLS config is specified since most traffic on port 443 is expected to be encrypted.
By @BrynCooke in https://github.com/apollographql/router/pull/#2048
🛠 Maintenance
Apply Tower best-practice to "inner" Service cloning (PR #2030)
We found our Service readiness checks could be improved by following the Tower project's recommendations for cloning inner Services.
Split the configuration file implementation into modules (Issue #1790)
The internals of the implementation for the configuration have been modularized to facilitate on-going development. There should be no impact to end-users who are only using YAML to configure their Router.
Apply traffic-shaping directly to supergraph and subgraph (PR #2034)
The plugin infrastructure works on BoxService instances and makes no guarantee on plugin ordering. The traffic shaping plugin needs a clonable inner service, and should run right before calling the underlying service. We'e changed the traffic plugin application so it can work directly on the underlying service. The configuration remains the same since this is still implemented as a plugin.
📚 Documentation
Remove references to Git submodules from DEVELOPMENT.md (Issue #2012)
We've removed the instructions from our development documentation which guide users to familiarize themselves with and clone Git submodules when working on the Router source itself. This follows-up on the removal of the modules themselves in PR #1856.