New Version of Envoy With Minor Modifications

Just a quarter of a year after the release of Envoy 1.18.0, here comes Envoy 1.19.0 with an array of minor changes and a template for the imminent Envoy 1.20.0. Included in this new version are big fixes leading the way for potential large-scale architecture deployment—very productive days from the developers of Envoy proxy.

Originally developed at Lyft, the distributed service has evolved into a massively adopted C++ service for extensive service mesh infrastructure that offers improved reliability, observability, and security.

Although Envoy itself is an independent service, it also serves as a vital component of the Istio service mesh; hence these changes will hold meaning to Istio users.

Envoy 1.19.0 involves a couple of significant differing behavior changes that might require changes because of incompatibility. The grpc_bridge_filter has been exonerated from grpc stats; this could be reversed with the aid of a runtime key. Another significant incompatibility change is that Apache Skywalker tracer support is now incompatible with data collect protocol 8.3.0. It is, however, compatible with Apache SkyWalking tracer 8.3.0.

Another set of users might experience incompatibility owing to minor changes, including the abridgment of surplus quotes on metadata string values in the access_log. This could also experience reversal from a manually set runtime key. Requests to AWS services are now automatically diminished so that signatures that include the payload hash may be computed, making the filter interoperable with the vast majority of AWS services. See the change list on the Envoy project's website for more information.

This release features previously ignored bug fixes, including the cluster stats histograms, properly configuring the number of bytes sent, handling retries in the process. The infamous erroneous handling of invalid nghttp2 frames has also been fixed. Envoy 1.19.0 also includes a new response indicator for overload manager termination in the access_log triggered when the overload manager ends the http stream.

The cluster manager modification is the dns_resolution_config which sums all DNS resolver configuration into a unit. Also in the cluster manager is the wait_for_warm_on_init, which prevents interruptions between cluster readiness and cluster warm-up. This is, however, only limited to DNS-based clusters.

In this release, crash support restores crash context when running requests or answers resulting from a prompt that directly executes a filter.