xds: omit node in subsequent discovery requests in the gRPC call #7860
Labels
Milestone
Comments
|
Yes, this should be done if we're not already. I thought we were already doing this. |
htuch
added
enhancement
|
I'm pretty sure it's not happening right now. Taking a cursory look at go-control-plane, I see that it does not cache Node from the first request, but it still works. More importantly, this is under-specified in xDS specification. |
This was referenced Aug 8, 2019
htuch
pushed a commit
that referenced
this issue
Aug 13, 2019
Omit the node identifier from subsequent discovery requests on the same stream. Restricted to non-incremental xDS for tractability. Risk Level: low, affects xDS protocol but guarded by an option Testing: Unit/integration tests are updated Docs Changes: xDS spec clarification Release Notes: omit the node identifier from subsequent discovery requests Fixes: #7860 Signed-off-by: Kuat Yessenov <kuat@google.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
xDS requests include a node proto with an arbitrarily large metadata field. This causes a problem if this metadata is a large JSON since it is sent for every EDS request. The proposal is to omit
nodefield in a gRPC stream after the first request. Most gRPC servers can save some state per stream, so they can store the node in the local scope. In fact, this optimization is already applied for ALS and metrics services, so it seems like a gap in xDS.cc @htuch
The text was updated successfully, but these errors were encountered: