xds: mesh gateway CDS requests are now allowed to receive an empty CDS reply

[rboyer]

This is the rest of the fix for #6543 that was incompletely fixed in #6576.

[rboyer]

This is restoring the original length check which actually didn't have to get changed in the first PR.

[codecov]

Codecov Report

Merging #6787 into master will decrease coverage by <.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master    #6787      +/-   ##
==========================================
- Coverage   65.82%   65.82%   -0.01%     
==========================================
  Files         431      431              
  Lines       52292    52285       -7     
==========================================
- Hits        34421    34415       -6     
- Misses      13747    13750       +3     
+ Partials     4124     4120       -4

Impacted Files	Coverage Δ
agent/xds/server.go	78.1% <100%> (-0.74%)	⬇️
agent/proxycfg/snapshot.go	73.17% <100%> (ø)	⬆️
agent/consul/raft_rpc.go	78.72% <0%> (-4.26%)	⬇️
command/agent/agent.go	46.64% <0%> (-2.38%)	⬇️
agent/consul/rpc.go	78.57% <0%> (-1.13%)	⬇️
agent/consul/leader_connect.go	71.45% <0%> (-1.09%)	⬇️
agent/consul/connect_ca_endpoint.go	56.52% <0%> (-0.67%)	⬇️
command/debug/debug.go	65.97% <0%> (-0.6%)	⬇️
agent/config/builder.go	83.48% <0%> (+0.08%)	⬆️
... and 6 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 40df8be...cbde438. Read the comment docs.

[s-mang]

Looks fine, thanks for making that change.

[banks]

LGTM.

I'm still not 100% sure if the empty results filter is actually worth having at all. I added it originally thinking that was never valid. Then removed it to fix the "no healthy instances make sidecar hang" bug, then brought it back just because I realised there was a more semantically correct fix for that as noted in the comment.

At this point we've found another example of why it's valid so I'm totally fine with this as a conservative approach for now, but I feel like if we hit more reasonable cases where sending an empty response is correct and valid we should probably just ditch the check completely - it's one of those cases where "empty isn't necessarily the same as not-set-yet".

[ghost]

Hey there,

This issue has been automatically locked because it is closed and there hasn't been any activity for at least 30 days.

If you are still experiencing problems, or still have questions, feel free to open a new one 👍.