proxy: add idle timeout

[wasaga]

Summary

• Expose idle timeout https://www.envoyproxy.io/docs/envoy/latest/api-v3/config/route/v3/route_components.proto#envoy-v3-api-field-config-route-v3-routeaction-idle-timeout
• make both timeout and idle_timeout optional pointers
• reset timeout if unset and idle_timeout is set
• ensure backwards compatibility

Related issues

Fixes https://github.com/pomerium/internal/issues/445

Checklist

• reference any related issues
• updated docs
• updated unit tests
• updated UPGRADING.md
• add appropriate tag (improvement / bug / etc)
• ready for review

[codeclimate]

Code Climate has analyzed commit 677f681 and detected 0 issues on this pull request.

View more on Code Climate.

[coveralls]

Coverage decreased (-0.05%) to 64.654% when pulling 677f681 on wasaga/route-idle-timeout into 9bc5227 on master.

[desimone]

How do these policy timeout settings interact with the option timeouts?

pomerium/config/options.go

Lines 108 to 113 in 41a2622

	// Timeout settings : https://github.com/pomerium/pomerium/issues/40
	ReadTimeout time.Duration `mapstructure:"timeout_read" yaml:"timeout_read,omitempty"`
	WriteTimeout time.Duration `mapstructure:"timeout_write" yaml:"timeout_write,omitempty"`
	IdleTimeout time.Duration `mapstructure:"timeout_idle" yaml:"timeout_idle,omitempty"`

In particular, IdleTimeout?

It looks like WriteTimeout and ReadTimeout are no longer used as originally designed (as part of the old underlying net.Conn for proxy)?

[desimone]

same declaration is seperated in ToProto. Personally I prefer splitting it, and keeping the variable close to the assignment as in ToProto but I think we should be consistent either way.

edit: removed nit. Let's be consistent.

[wasaga]

How do these policy timeout settings interact with the option timeouts?

my reading of https://www.envoyproxy.io/docs/envoy/latest/api-v3/config/route/v3/route_components.proto#envoy-v3-api-field-config-route-v3-routeaction-idle-timeout: per-route idle_timeout relates to HTTP2 stream while the one you're referring controls timeout for TCP HTTP2 connection as a whole.

read_timeout is mapped to request_timeout

and write_timeout is mapped to max_stream_duration but only if it's set - currently it defaults to zero which disables it.