vpp-maglev/docs/config-guide.md

maglevd Configuration Guide

Overview

maglevd consumes a YAML configuration file of a specific format. Validation is performed in two stages:

1. Structural parsing: the YAML is unmarshalled into typed Go structs. Unknown fields and type mismatches are rejected immediately.
2. Semantic validation: cross-field and cross-object rules are enforced, for example ensuring that every backend referenced by a frontend exists, that address families are consistent within a frontend, and that IP source addresses are the correct family.

If you want to get started quickly, take a look at the example config.

Basic structure

The YAML configuration file has the following top-level structure:

maglev:
  healthchecker:
    [ Global health checker settings ]

  vpp:
    lb:
      [ VPP load-balancer integration settings ]

  healthchecks:
    my-check:
      [ Health check definition ]

  backends:
    my-backend:
      [ Backend definition ]

  frontends:
    my-frontend:
      [ Frontend (VIP) definition ]

All five sections live under the top-level maglev: key. The healthchecks, backends, and frontends sections are maps keyed by an arbitrary name of your choosing. Names must be unique within their section and are case-sensitive. The vpp section is required when maglevd has a working VPP connection — its lb.ipv4-src-address and lb.ipv6-src-address fields are mandatory and maglevd will refuse to start without them.

---

healthchecker

Global settings for the health checker engine.

• transition-history: An integer >= 1 that controls how many state transitions are retained per backend for display via the gRPC API. Defaults to 5.

• netns: The name of a Linux network namespace in which probes are executed. When empty or omitted, probes run in the current (default) network namespace. Useful when backends are reachable only through a dedicated dataplane namespace.

  Capability requirement: setting this field makes maglevd call setns(CLONE_NEWNET) on the probe thread before each probe, which the kernel only permits to processes holding CAP_SYS_ADMIN in the target namespace's user namespace (setns(2)). The Debian systemd unit (vpp-maglev.service) already grants this capability; if you run maglevd by hand under a non-root user make sure the binary has CAP_SYS_ADMIN via setcap cap_net_raw,cap_sys_admin=eip /usr/sbin/maglevd or equivalent, otherwise every probe fails with enter netns "<name>": operation not permitted and all backends transition to down on their first probe.

  Also make sure the named namespace is mounted under /var/run/netns/ (which is where ip netns add puts it) and that it is readable by the user maglevd runs as — the default mode from ip netns add is 0644, which is fine for any user.

Example:

maglev:
  healthchecker:
    transition-history: 10
    netns: dataplane

---

vpp

Settings controlling the integration with a locally running VPP instance. The vpp section is a map with a single sub-section, lb. Both lb.ipv4-src-address and lb.ipv6-src-address are required — maglevd --check exits with a semantic error and the daemon refuses to start when either is missing, because VPP's GRE encap needs a source address and every VIP maglevd programs uses GRE.

• lb.ipv4-src-address: Required. The IPv4 source address VPP uses when encapsulating IPv4 traffic into GRE4 tunnels to application servers. Must be a valid IPv4 address. No default.
• lb.ipv6-src-address: Required. The IPv6 source address VPP uses when encapsulating IPv6 traffic into GRE6 tunnels. Must be a valid IPv6 address. No default.
• lb.sync-interval: A positive Go duration (e.g. 30s, 1m) controlling how often maglevd reconciles the VPP load-balancer dataplane against its running configuration. On startup, an immediate full sync runs; subsequent syncs fire at this interval as long as the VPP connection is up. Defaults to 30s. The purpose is to catch drift — for example, a VIP added to VPP by hand — and bring VPP back in line with the maglev config.
• lb.sticky-buckets-per-core: The number of buckets per worker thread in the established-flow table. Must be a power of 2. Defaults to 65536 (64k).
• lb.flow-timeout: Idle time after which an established flow is removed from the table. Must be a whole number of seconds between 1s and 120s inclusive. Defaults to 40s.
• lb.startup-min-delay: Absolute hands-off window at the start of the maglevd process. For the first startup-min-delay seconds of the process's life, no VPP LB sync of any kind is issued — neither the periodic SyncLBStateAll loop nor the per-transition SyncLBStateVIP path from the reconciler touches VPP. This makes a maglevd restart dataplane-neutral: without this gate, the first sync would fire before any probes had completed, every backend would still be in StateUnknown, and every AS would be reprogrammed to weight 0 until the rise counters caught up — producing a visible black-hole window of several seconds on every restart. A non-negative Go duration. Must be <= startup-max-delay. Defaults to 5s. Set to 0s (together with startup-max-delay: 0s) to disable the warmup entirely and sync VPP immediately on startup.
• lb.startup-max-delay: Watchdog for the per-VIP release phase that follows startup-min-delay. Between min-delay and max-delay, each frontend is released individually as soon as every backend it references has reached a non-Unknown state, and a single SyncLBStateVIP runs against the newly-released frontend. At max-delay the warmup driver unconditionally runs SyncLBStateAll, marking the warmup phase complete regardless of whether any backends are still StateUnknown — stragglers get programmed at their current effective weight at that point, which for a StateUnknown backend means weight 0. A non-negative Go duration. Must be >= startup-min-delay. Defaults to 30s. Set to 0s together with startup-min-delay: 0s to disable the warmup entirely.

These values are pushed to VPP via lb_conf when maglevd connects to VPP and again after every config reload (whenever they change). A log line vpp-lb-conf-set records the effective values. The startup-* settings are latched at the first successful VPP connect and are not re-read on subsequent config reloads — a reload that changes them only takes effect on the next process start.

Example:

maglev:
  vpp:
    lb:
      sync-interval: 60s
      ipv4-src-address: 10.0.0.1
      ipv6-src-address: 2001:db8::1
      sticky-buckets-per-core: 65536
      flow-timeout: 40s
      startup-min-delay: 5s
      startup-max-delay: 30s

---

healthchecks

A named map of health check definitions. Each health check describes how to probe a backend. Backends reference health checks by name. The same health check can be reused across any number of backends; each backend is probed exactly once regardless of how many frontends reference it.

Common fields (all types):

• type: Required. One of icmp, tcp, http, or https.
• port: The destination port to probe. Required for tcp, http, and https. Must be omitted for icmp.
• probe-ipv4-src: An optional IPv4 source address used when probing IPv4 backends. Must be an IPv4 address. When omitted, the OS chooses the source address.
• probe-ipv6-src: An optional IPv6 source address used when probing IPv6 backends. Must be an IPv6 address. When omitted, the OS chooses the source address.
• interval: Required. A positive Go duration string (e.g. 2s, 500ms) controlling how often a probe is sent when the backend is fully healthy (counter at maximum).
• fast-interval: Optional. A positive duration used instead of interval while the backend's health counter is degraded (between down and up) or in unknown state. When omitted, interval is used.
• down-interval: Optional. A positive duration used instead of interval while the backend is fully down (counter at zero). When omitted, interval is used. Setting this to a longer value reduces probe traffic to backends that are known to be offline.
• timeout: Required. A positive duration after which an in-flight probe is abandoned and counted as a failure.
• rise: The number of consecutive successes required to transition from down to up. Defaults to 2. Must be >= 1.
• fall: The number of consecutive failures required to transition from up to down. Defaults to 3. Must be >= 1.

type: icmp

Sends an ICMP echo request (ping) to the backend address. Requires CAP_NET_RAW. No port may be specified. No params block is used.

healthchecks:
  ping:
    type: icmp
    probe-ipv4-src: 10.0.0.1
    probe-ipv6-src: 2001:db8::1
    interval: 2s
    timeout: 1s
    rise: 2
    fall: 3

type: tcp

Opens a TCP connection to the backend and immediately closes it upon success. Use params to optionally wrap the connection in TLS.

• params.ssl: A boolean. When true, a TLS handshake is performed after the TCP connection is established. Defaults to false.
• params.server-name: The TLS SNI hostname sent during the handshake. When omitted, the backend IP address is used.
• params.insecure-skip-verify: A boolean. When true, the TLS certificate presented by the server is not verified. Defaults to false.

healthchecks:
  imaps-check:
    type: tcp
    port: 993
    params:
      ssl: true
      server-name: imaps.example.com
    interval: 5s
    timeout: 3s
    rise: 2
    fall: 3

type: http / https

Opens a TCP (or TLS for https) connection, sends an HTTP request, and evaluates the response code. An optional regexp can additionally match against the response body.

• params.path: Required. The HTTP request path, e.g. /healthz.
• params.host: The Host header value sent in the request. When omitted, the backend IP address is used.
• params.response-code: The expected HTTP response code. Can be a single value ("200") or an inclusive range ("200-299"). Defaults to "200".
• params.response-regexp: An optional Go regular expression matched against the response body. If specified, the body must match for the probe to succeed.
• params.server-name: The TLS SNI hostname (https only). Defaults to the value of params.host if not set.
• params.insecure-skip-verify: A boolean. Skip TLS certificate verification (https only). Defaults to false.

healthchecks:
  nginx-http:
    type: http
    port: 80
    params:
      path: /healthz
      host: nginx.example.com
      response-code: "200-204"
    interval: 2s
    fast-interval: 500ms
    down-interval: 30s
    timeout: 3s
    rise: 2
    fall: 3

  nginx-https:
    type: https
    port: 443
    params:
      path: /healthz
      host: nginx.example.com
      server-name: nginx.example.com
      insecure-skip-verify: false
    interval: 5s
    timeout: 3s

---

backends

A named map of individual backend servers. Each backend has a single IP address and optionally references a health check by name. Backends are probed exactly once, even if they appear in multiple frontends.

• address: Required. The IPv4 or IPv6 address of this backend server.
• healthcheck: The name of a health check defined in the healthchecks section. When empty or omitted, the backend is static: no probing is performed and the backend enters StateUp immediately on startup (via a synthetic pass, rise/fall forced to 1/1). This is useful for backends that are always available or managed by other means. See healthchecks.md for details on the static-backend behavior.
• enabled: A boolean controlling whether this backend participates in any frontend. When false, the backend is excluded entirely and no probe goroutine is started. Defaults to true.

Examples:

backends:
  nginx0-ams:
    address: 198.51.100.10
    healthcheck: nginx-http
  nginx0-lon:
    address: 198.51.100.11
    healthcheck: nginx-http
  nginx0-draining:
    address: 198.51.100.12
    healthcheck: nginx-http
    enabled: false
  static-backend:
    address: 198.51.100.20
    # no healthcheck: assumed always healthy

---

frontends

A named map of virtual IPs (VIPs). Each frontend ties together a listener address with an ordered list of backend pools. The gRPC API exposes frontends by name.

• description: An optional free-text string for documentation purposes.
• address: Required. The IPv4 or IPv6 address of the VIP.
• protocol: The IP protocol, either tcp or udp. When omitted, the frontend matches all traffic to the VIP address regardless of protocol. If port is specified, protocol must also be set.
• port: The destination port of the VIP, an integer between 1 and 65535. Requires protocol to be set. When omitted, the frontend matches all ports. Note that the frontend port is independent of the healthcheck port: a frontend on port 443 may use a healthcheck that probes port 80.
• pools: Required. A non-empty ordered list of pool objects. Pools express priority: the first pool is preferred; subsequent pools act as fallbacks. When every backend in pool[0] leaves StateUp (down, paused, disabled, or not yet probed), pool[1] is automatically promoted — its up backends take over serving traffic. The promotion cascades across further tiers. See healthchecks.md for the full failover semantics. All backends across all pools in a frontend must have addresses of the same address family (all IPv4 or all IPv6).
• src-ip-sticky: Boolean, default false. When true, the VPP load-balancer programs this VIP with source-IP-based stickiness — all flows from the same client source IP hash to the same backend (subject to the Maglev consistent-hash bucket assignment). Use this for protocols that require session affinity at the L3 level, or when clients open many short flows that should land on one backend. Changing this field in a running config and reloading causes maglevd to tear down the VIP (all application servers are deleted with flush, then the VIP itself is deleted) and recreate it with the new value; VPP has no API to mutate src_ip_sticky on an existing VIP, and existing flow state cannot be preserved across the flip.
• flush-on-down: Boolean, default true. Controls what happens to existing flows pinned to a backend that transitions to down. With it true, maglevd issues lb_as_set_weight(is_flush=true) on the down transition, clearing VPP's flow-table entries for that backend so existing connections are torn down immediately and reshuffle onto healthy backends. With it false, the weight drops to 0 (drain only): new flows skip the dead backend via the Maglev lookup table, but existing sticky flows keep being steered at the dead IP until the client retries, producing visible "connection refused" oscillations during an outage. The default is true because the healthcheck's rise / fall counters already absorb single-probe flaps, so a fall-counted down is almost always a real outage where immediate session teardown is the safer behaviour. Set it to false per frontend only when existing sessions are expensive enough that you'd rather keep them pinned to a potentially-dead backend than reset them (e.g. long-lived WebSocket connections with expensive reconnect logic). The disabled state always flushes regardless of this flag — disabled is an explicit operator signal that the backend is going away.

Each pool has:

• name: Required. A non-empty string identifying the pool (e.g. primary, fallback).
• backends: A map of backend names to per-pool backend options. Every name must refer to an existing entry in the backends section.

Per-pool backend options:

• weight: An integer between 0 and 100 (inclusive) expressing the relative weight of this backend within the pool. 0 keeps the backend in the pool but assigns it no traffic. Defaults to 100. Weight is per-pool, not global — the same backend can appear with different weights in different frontends.

Examples:

frontends:
  nginx-v4-http:
    description: "IPv4 HTTP VIP with fallback"
    address: 198.51.100.1
    protocol: tcp
    port: 80
    pools:
      - name: primary
        backends:
          nginx0-ams: { weight: 10 }
          nginx0-lon: {}
      - name: fallback
        backends:
          nginx0-fra: {}

  maildrop-imaps:
    description: "IMAPS VIP"
    address: 2001:db8::1
    protocol: tcp
    port: 993
    src-ip-sticky: true
    pools:
      - name: primary
        backends:
          maildrop0-ams: {}
          maildrop0-lon: {}

---

For a detailed description of the health state machine, probe intervals, and all transition events, see healthchecks.md. For a user guide on how to use the maglev daemon and client, see the user-guide.md.