4 / 4 API Gateway, Proxies & Service Mesh

Six Hops Later

After rolling out a sidecar-based mesh, a user request that fans through a chain of four services (three internal calls) sees a clear p99 latency increase, while CPU and error rates look normal. What is the most likely dominant cause?

A

Sidecars disable connection reuse between hops, so each request in the chain pays a fresh TCP handshake on both sides of the call

B

Every request performs a fresh mTLS handshake between the two sidecars, and the asymmetric cryptography in that handshake dominates the added time on each hop

C

The control plane is consulted for a routing and policy decision on each call, adding a network round trip to the control plane before every hop in the chain

D

Each internal call now traverses the caller's sidecar and the callee's sidecar, so three calls became six extra proxy traversals along the chain