server-zd v0.1.16

Breaking Changes

None. Configuration and persistent formats are unchanged.

server-zd v0.1.16

Fixes

• Take-back after a step down (#1182, fixes #1174). A process that steps down and then takes its own lease back now serves instead of exiting.
  • A step down ends public calls still running after the lease release with UNAVAILABLE ("the server process stopped serving during the call; retry it"). These calls are safe to retry.
  • Calls that finish before the release keep their answers.
  • The stepped-down process waits up to 60 s for its old local store to close before it reopens the store or competes for the lease. If a stuck task still holds the store, the process exits with status 1 without taking the lease.
• One gRPC server instance accepts connections (#1182). Across startup, takeover and step down, only the newest gRPC server instance accepts new connections. A client no longer sees "the node is starting" after readiness reports SERVING, or an h2 GoAway error on its first call.

Known issues

• After a take-back, readiness can report SERVING for up to about 250 ms on the old, stepped-down server instance (#1210). This was already present in v0.1.15.

Upgrade

• Upgrade every process, including standbys.
• Before server-zd 1.0, rolling a node back to an older 0.x binary after a newer one has run is not supported.

Artifacts

• Container image dex-server-zd:0.1.16 for linux/arm64 and linux/amd64.
• dex-server-zd binaries for Linux amd64, Linux arm64 (glibc 2.28 or newer), macOS arm64 and macOS amd64, with checksums.txt.