John Howard
John Howard
> Can we make a flag that loads protos on-demand maybe? You could add descriptors as files next to the binary, and then have a flag to load them up...
> Not sure if it's safe to release - there'll be significant impact on the size of xDS since JSONs are inefficient in proto. Technically, the user can simply rewrite...
> The fallback to typed struct will increase the xDS size significantly - e.g. adding a filter to every listener will be O(10KB) instead of O(1KB). I guess I was...
``` No EF: 47.8kb Proto: 49.1kb TypedStruct: 49.8kb ``` For this config: ``` apiVersion: networking.istio.io/v1alpha3 kind: EnvoyFilter metadata: name: reviews-lua spec: configPatches: # The first patch adds the lua filter...
> Maybe it's OK then? The overhead is changing the field number to a field name, so as long as there are not too many fields used, it might be...
> Do we plan to continuously update the protos(atleast the main ones) so that fallback is for few niche use cases like contrib? The current list was manually aggregated based...
Not 100% sure I follow. What is the expectation? That at least one field under outlierDetection is set before its enabled?
`outlierDetection: {}` seemed like a reasonable workaround, is there any issue with that? Maybe you don't want to set it on a bunch of DRs?
The intent behind this is that without outlierDetection, we cannot detect failing workloads, so will never failover at all. IE its not "failover" just "only send to same zone"