Branding: bulk rename across 50+ files (Go, JS, CSS, HTML, YAML, JSON, shell scripts). Fix Helm chart description and git repo URLs. Fix test helper taglines. Admin surface: strip 14 gutted tabs (providers, models, personas, roles, knowledge, memory, tasks, health, routing, capabilities, channels, dashboard, usage, stats). Collapse to 4 categories: People, Workflows, System, Monitoring. Settings surface: strip 11 gutted tabs (models, personas, providers, roles, knowledge, memory, usage, workflows, tasks, data, gitkeys). Keep: general, appearance, profile, teams, connections, notifications. Team-admin surface: strip 5 gutted tabs (personas, providers, knowledge, tasks, usage). Keep: members, groups, connections, workflows, settings, activity. Components: delete chat-pane/ (13 files, 1938 lines) and notes-pane/ (10 files, 2381 lines). main.go: remove stale Health.Prune maintenance goroutine. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
27 lines
923 B
YAML
27 lines
923 B
YAML
# k8s/middleware-retry.yaml
|
|
# ============================================
|
|
# Switchboard Core - Traefik Retry Middleware
|
|
# ============================================
|
|
# Retries failed requests (502, connection reset) once before giving up.
|
|
# Applies to API + WebSocket paths via the Ingress annotation.
|
|
#
|
|
# Why: After a heavy provider model sync (100+ DB upserts), the Go
|
|
# process may briefly delay responses. Traefik sees a stale keepalive
|
|
# connection or a slow initial byte and returns 502. A single retry
|
|
# on a fresh connection resolves it transparently.
|
|
#
|
|
# This is applied per-environment (${DEPLOY_SUFFIX} ensures isolation).
|
|
# ============================================
|
|
apiVersion: traefik.io/v1alpha1
|
|
kind: Middleware
|
|
metadata:
|
|
name: switchboard-retry${DEPLOY_SUFFIX}
|
|
namespace: ${NAMESPACE}
|
|
labels:
|
|
app: switchboard
|
|
env: ${ENVIRONMENT}
|
|
spec:
|
|
retry:
|
|
attempts: 2
|
|
initialInterval: 100ms
|