Repository navigation
[P3] XDS versionCount resets to 1 on router restart — config changes may not reach Envoy #617
Description
Severity: P3 (edge case; only affects first restart after exactly one prior snapshot)
Component: Network Plane — cmd/atenet/internal/router/xds.go
Audit ID: NET-9
Summary
The XDS snapshot version counter is an in-memory int64 starting at 0. Each router
restart begins versioning at 1. If Envoy previously ACKed snapshot version "1" and
the router restarts pushing a new config as version "1" again (same string), the
go-control-plane ADS server may not re-push the config (depending on its same-version
handling). This creates a small window where Envoy's config reflects the pre-restart
state rather than the new state.
Root Cause
File: cmd/atenet/internal/router/xds.go lines 99–100
var versionCount int64 = 0 // reset to 0 on every process start
func nextVersion() string {
return strconv.FormatInt(atomic.AddInt64(&versionCount, 1), 10)
// First call after restart always returns "1"
}The go-control-plane cache's behavior when a client reconnects with version_info="1"
and the server's current version is also "1" depends on the implementation:
if the cache treats it as "client already has this version," it may not push an update.
Steps to Reproduce
# Initial state: router running, Envoy has ACKed version "1" (just started)
# Confirm via Envoy admin
curl http://localhost:9901/config_dump | jq '.configs[] | select(.["@type"] | contains("ScopedRoutesConfig")) | .version_info'
# "1"
# Change router config and restart
kubectl set env -n ate-system deployment/atenet-router SOME_PARAM=new_value
kubectl rollout restart -n ate-system deployment/atenet-router
# New router pushes version "1" again (reset counter)
# Envoy may not receive the updated config if it reconnects and sends version_info="1"
curl http://localhost:9901/config_dump | jq '...'
# May still show old configSuggested Fix
Initialize versionCount with a monotonically increasing base that survives restarts:
// Use Unix timestamp as base to ensure monotonic increase across restarts
var versionCount int64 = time.Now().Unix()Or use a UUID-based version string that is always different from any prior version,
eliminating the same-version collision entirely:
func nextVersion() string {
return fmt.Sprintf("%d-%d", startupTimestamp, atomic.AddInt64(&versionCount, 1))
}
Activity
[-]XDS versionCount resets to 1 on router restart — config changes may not reach Envoy[/-][+][P3] XDS versionCount resets to 1 on router restart — config changes may not reach Envoy[/+]fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique
fix(atenet): make xDS snapshot versions restart-unique