224167ce396c4441070dc3705e74579933c16e3a
7 Commits
| Author | SHA1 | Message | Date | |
|---|---|---|---|---|
Pim van Pelt
|
224167ce39 |
Dataplane reconcile fixes; LB counters cleanup; SPA scope cookie
Checker / reload:
- Reload's update-in-place branch now mirrors b.Address onto the
runtime health.Backend. Without this, GetBackend kept returning
the pre-reload address indefinitely after a config edit that
touched addresses but not healthcheck settings — the VPP sync
path reads cfg.Backends directly so the dataplane moved on
while the gRPC and SPA view stayed wedged on the old IPv4/IPv6.
Sync (internal/vpp/lbsync.go):
- reconcileVIP now detects encap mismatch in addition to
src-ip-sticky mismatch and takes the full tear-down / re-add
path via a new shared recreateVIP helper. Triggered when every
backend flips address family (gre4 <-> gre6) and the existing
VIP can no longer accept new ASes — previously the sync wedged
with 'Invalid address family' until a full maglevd restart.
- setASWeight is issued whenever the state machine requests
flush (a.Flush=true), not only on the weight-value transition
edge. Fixes the case where a backend reached StateDisabled
after its effective weight had already been drained to 0 by
pool failover — the sticky-cache entries pointing at it were
previously never cleared.
maglev-frontend:
- signal.Ignore(SIGHUP) so a controlling-terminal disconnect
doesn't kill the daemon.
- debian/vpp-maglev.service grants CAP_SYS_ADMIN in addition to
CAP_NET_RAW so setns(CLONE_NEWNET) can join the healthcheck
netns. Comment documents the 'operation not permitted' symptom
and notes the knob can be dropped if the deployment doesn't use
the 'netns:' healthcheck option.
LB plugin counters (internal/vpp/lbstats.go + friends):
- Fix the VIP counter regex: the LB plugin registers
vlib_simple_counter_main_t names without a leading '/'
(vlib_validate_simple_counter in counter.c:50 uses cm->name
verbatim; only entries that set cm->stat_segment_name get a
slash). first/next/untracked/no-server now read through as
live values instead of zero.
- Drop the per-backend FIB counter block end-to-end (proto,
grpcapi, metrics, vpp.Client, lbstats, maglevc). Traced from
lb/node.c:558 into ip{4,6}_forward.h:141 — the LB plugin
forwards by writing adj_index[VLIB_TX] directly and bypassing
ip{4,6}_lookup_inline, which is the only path that increments
lbm_to_counters. The backend's FIB load_balance stats_index
literally never ticks for LB-forwarded traffic, so the column
was always zero and misleading. docs/implementation/TODO
records the full investigation and the recommended upstream
path (new lb_as_stats_dump API message) for when we're ready
to carry that VPP patch.
- maglevc show vpp lb counters: plain-text tabular headers.
label() wraps strings in ANSI escapes (~11 bytes of overhead),
but tabwriter counts bytes, not rendered width — so a header
row with label()'d cells and data rows with plain cells drifts
column alignment on every row. color.go comment now spells
out the constraint: label() only works when column N is
wrapped identically in every row (key-value layouts are fine,
multi-column tables with header-only labelling are not).
SPA:
- stores/scope.ts is cookie-backed (maglev_scope, 1 year,
SameSite=Lax). App.tsx hydrates from the cookie then validates
against the fetched snapshots: a cookie referencing a maglevd
that no longer exists falls through to snaps[0] instead of
leaving the user on a ghost selection.
- components/Flash.tsx wraps props.value in createMemo. Solid's
on() fires its callback on every dep notification, not on
value change — source is right in solid-js/dist/solid.js:460,
no equality check. Without the memo, flipping scope between
two 'connected' maglevds (or any other cross-store reactive
re-eval that doesn't actually change the concrete string)
replays the animation every time. createMemo's default ===
dedupe fixes it in one place for every Flash consumer,
superseding the local createMemo workaround we'd added in
BackendRow earlier.
|
||
|
Pim van Pelt
|
1a1c48ef54 |
LB buckets column + health cascade; VPP dump fix; maglevc strictness
SPA (cmd/frontend/web): - New "lb buckets" column backed by a 1s-debounced GetVPPLBState fetch loop with leading+trailing edge coalesce. - Per-frontend health icon (✅/⚠️/❗/‼️/❓) in the Zippy header, gated by a settling flag that suppresses ‼️ until the next lb-state reconciliation after a backend transition or weight change. - In-place leaf merge on lb-state so stable bucket values (e.g. "0") don't retrigger the Flash animation on every refresh. - Zippy cards remember open state in a cookie, default closed on fresh load; fixed-width frontend-title-name + reserved icon slot so headers line up across all cards. - Clock-drift watchdog in sse.ts that forces a fresh EventSource on laptop-wake so the broker emits a resync instead of hanging on a dead half-open socket. Frontend service (cmd/frontend): - maglevClient.lbStateLoop, trigger on backend transitions + vpp-connect, best-effort fetch on refreshAll. - Admin handlers explicitly wake the lb-state loop after lifecycle ops and set-weight (the latter emits no transition event on the maglevd side, so the WatchEvents path wouldn't have caught it). - /favicon.ico served from embedded web/public IPng logo. VPP integration: - internal/vpp/lbstate.go: dumpASesForVIP drops Pfx from the dump request (setting it silently wipes IPv4 replies in the LB plugin) and filters results by prefix on the response side instead, which also demuxes multi-VIP-on-same-port cases correctly. maglevc: - Walk now returns the unconsumed token tail; dispatch and the question listener reject unknown commands with a targeted error instead of dumping the full command tree prefixed with garbage. - On '?', echo the current line (including the '?') before the help list so the output reads like birdc. Checker / prober: - internal/checker: ±10% jitter on NextInterval so probes across restart don't all fire on the same tick. - internal/prober: HTTP User-Agent now carries the build version and project URL. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com> |
||
|
Pim van Pelt
|
35643fd774 |
Rename maglev-frontend → maglevd-frontend; v0.9.1; API RX/TX pulse
Rename the web dashboard binary to maglevd-frontend and move it to /usr/sbin (it's a daemon and belongs with maglevd). The systemd unit name stays vpp-maglev-frontend.service since that prefix is the package name. Manpage, README, user-guide, and debian packaging all updated in lockstep; bump to 0.9.1 for the first real release. All frontend env vars are now prefixed MAGLEV_FRONTEND_ so a single /etc/default/vpp-maglev can be shared with maglevd without collisions. Every flag has an env equivalent for Docker use. MAGLEV_FRONTEND_USER and MAGLEV_FRONTEND_PASSWORD still gate the /admin surface. VPPInfoPanel now pulses "API: ↑↓" indicators in the zippy title whenever a vpp-api-send / vpp-api-recv log event arrives on the SSE stream for the scoped maglevd — 250ms blue flash, re-triggerable, with the two arrows tightly kerned via negative letter-spacing. |
||
|
Pim van Pelt
|
1191b3d994 |
Frontend aggregate state: SPA-side derive + checker fixes
The web UI showed the wrong up/down state for frontends whose pool
composition had been touched by a mix of runtime disable/enable and
weight changes: a frontend with every backend at effective_weight=0
would still display "up", while a sibling frontend with a serving
fallback backend would display "down". Two independent bugs, each
fixed on its own layer.
On the fast path (healthCheckEqual returns true), Reload did
`w.entry = b`, blindly replacing the runtime worker entry with the
fresh YAML record. YAML's default for Enabled is true, so any
backend the operator had runtime-disabled would have its Enabled
flag silently reset while the worker's backend.State stayed at
StateDisabled. Subsequent EnableBackend calls then early-returned
on `if w.entry.Enabled` and never transitioned the state machine
— the CLI reported "enabled, state is 'disabled'" and the backend
was permanently stuck.
Fix: preserve w.entry.Enabled across the fast-path replacement.
runtimeEnabled := w.entry.Enabled
w.entry = b
w.entry.Enabled = runtimeEnabled
Runtime operator state now outlives config reloads. On the worker-
restart path (different health check) the new worker is
structurally fresh and the YAML's Enabled is still authoritative.
Both methods used `w.entry.Enabled` as their idempotency check,
which meant a stuck `Enabled=true, State=disabled` combo couldn't
be repaired even after the Reload fix (existing bad state had to
survive the upgrade). Switched both methods to key on
`w.backend.State`:
- DisableBackend: if state == StateDisabled, sync the flag but
don't emit a redundant transition; otherwise do the full
state transition + flag flip + worker cancel.
- EnableBackend: if state != StateDisabled, sync the flag but
don't emit a redundant transition; otherwise do the full
transition + flag flip + probe-goroutine restart.
Either method will now unstick any inconsistency between the
flag and the state machine — future drift from a panic, a new
code path we haven't thought of, or existing already-stuck
backends from before this commit are all repaired on the next
enable/disable call.
Changing a backend's weight can flip a frontend between up and
down (e.g. zeroing the last non-zero-weighted backend in the
active pool), but SetFrontendPoolBackendWeight never called
updateFrontendState, so the checker's cached frontend state
would drift from reality until the next genuine backend
transition happened to trigger a recompute. The symptom was
"show frontends nginx-ip4-http" reporting up even with every
effective_weight=0.
Fix: call c.updateFrontendState(frontendName, fe) after the
weight mutation, under the same lock. The recompute emits a
FrontendEvent transition if the aggregate flipped, so any
WatchEvents consumer picks up the change live.
stores/state.ts recomputeEffectiveWeights is renamed and
extended to recomputeDerivedState, which now also writes
fe.state using the same rule as health.ComputeFrontendState:
unknown if no backends or all unknown, up if any effective
weight > 0, down otherwise. Called from every mutation path
(replaceAll, replaceSnapshot, applyBackendTransition,
applyConfiguredWeight) so the SPA is authoritative for *display*
state and doesn't inherit any staleness the server's cached
frontendStates map might have.
applyFrontendTransition is now a no-op for the state field —
the server's `to` value is no longer trusted because
recomputeDerivedState walks the local backends array on every
update and produces a fresh, correct answer. The reducer is kept
as a named function so sse.ts's dispatch table still has a
landing spot for "frontend" events (they still feed the
DebugPanel via pushEvent); the empty body is deliberate, not a
bug — a comment at the top spells it out.
|
||
|
Pim van Pelt
|
4347bb9b05 |
Bug fixes, config validation, SPA tightening, set-weight UI
This session covers three distinct arcs: correctness bug fixes in the
VPP sync path and frontend reducers, new config validation, and a
large polish pass on the web frontend (tighter layout, backend kebab
dialogs, live grouped-table, live config-reload re-sync).
- encap for a VIP is now derived from the backend address family,
not the VIP's. A v6 VIP with v4 backends is programmed as IP6_GRE4
(not the buggy IP6_GRE6), matching the VPP LB plugin's
requirement that encap reflects the tunnel inner family. desiredVIP
gained an Encap field populated in desiredFromFrontend.
- ActivePoolIndex now requires at least one backend in a pool to be
BOTH in StateUp AND pb.Weight>0 before the pool counts as active.
Previously a primary pool with every backend manually zeroed would
still win over a fallback with weight=100, so fallback traffic
never materialized. New TestActivePoolIndexWeightedFailover table
pins the rule in five subcases.
- SyncLBStateVIP gained a flushAddress parameter threaded through
reconcileVIP; it forces flush=true on the setASWeight call for a
specific backend regardless of the usual 0→N heuristic. Wires up
the explicit [flush] knob the CLI exposes.
- convertFrontend already enforced that backends within one frontend
share a family. New cross-frontend pass validateVIPFamilyConsistency
rejects configs where two frontends share a VIP address but carry
backends in different families — VPP's LB plugin requires every
VIP on a prefix to have the same encap type, so such a config
would fail at lb_add_del_vip_v2 time with VNET_API_ERROR_INVALID
_ARGUMENT (-73). Catching it at config load turns a silent
runtime failure into a clear startup error.
- Two new TestValidationErrors cases pin the behavior: mismatched
families reject, same-family frontends on one VIP address allowed.
- Proto adds `bool flush = 5` to SetWeightRequest. The RPC now
drives a VIP sync immediately after mutating config (fixing the
latent "weight change only takes effect at the next 30s periodic
reconcile" gap), passing flushAddress = backend IP when req.Flush
is true.
- maglevc grows an optional [flush] token: `set frontend F pool P
backend B weight N [flush]`. Implementation uses two Run closures
(runSetFrontendPoolBackendWeight and -Flush) because the tree
walker only puts slot tokens in args — literal keywords like
`flush` advance the node but don't appear in the arg list.
- docs/user-guide.md updated with the [flush] optional and a
three-paragraph explainer of the graceful-drain vs. flush
semantics at the VPP level.
- checker.ListFrontends now sorts alphabetically to match the
existing sort in ListBackends / ListHealthChecks — RPC responses
no longer shuffle VIPs per call. cmd/frontend/client.go also
sorts defensively in refreshAll so an old maglevd build renders
alphabetically too.
- backendFromProto was returning out.Transitions[n-1] as the
LastTransition, but maglevd stores (and the proto carries)
transitions newest-first, so [n-1] was actually the oldest.
Reverse on read, which normalizes the client's Transitions slice
to oldest-first and makes [n-1] genuinely the newest. LastTransition
now points at the actual latest transition record.
- applyBackendTransition (Go and TS) derives Enabled = state!="disabled"
so the two fields stay in lockstep — closed a drift window where
a recently re-enabled backend still rendered with a stuck
[disabled] tag. The tag was later removed entirely since state
and enabled carry the same information.
- Layout tightened substantially: "FRONTENDS" panel header removed,
zippy-summary and zippy-body paddings cut, backend-table row
padding dropped to 2px, per-pool <h3> removed. Pools now live in
a single consolidated table per frontend with a dedicated "pool"
column that shows the pool name only on the first row of each
group — classic grouped-table layout, maximally dense.
- Description moved inline into the Zippy summary as muted italic
text, freeing a vertical line per frontend card.
- formatVIPAddress() helper renders IPv6 VIPs as [addr]:port and
IPv4 as addr:port, matching RFC 3986 authority syntax.
- Pools with effective_weight=0 on every backend (standby
fallbacks, fully-drained primaries) render at opacity 0.35 on
their non-actions cells; the kebab column stays at full contrast
because its menu is still fully functional on standby backends.
- Config-reload propagation: a maglevd config-reload-done log
event triggers triggerConfigResync() on the frontend side —
refreshAll() runs off the event-dispatch goroutine, then a
BrowserEvent{Type:"resync"} is published through the broker.
writeEvent emits type="resync" as a named SSE frame so the
SPA's existing addEventListener("resync") handler picks it up
and calls fetchAllState → replaceAll.
- recomputeEffectiveWeights in stores/state.ts mirrors the
server-side health.EffectiveWeights logic so the SPA keeps
pool.effective_weight correct the moment a backend transitions,
without waiting for the 30s refresh. Fixed a nasty bug where
applyBackendEffectiveWeight wrote VIP-scoped vpp-lb-sync-as-*
event weights into every frontend sharing the backend,
corrupting frontends with different per-pool configured weights.
The old log-event reducer was removed; applyConfiguredWeight is
the narrower replacement used by the kebab set-weight flow.
- applyBackendTransition calls recomputeEffectiveWeights after
state updates so pool-failover transitions (primary ⇌ fallback)
reflect instantly in the UI.
- Confirmation dialogs via a new Modal primitive
(Portal-mounted to document.body, escape/click-outside close,
click-outside debounced on mousedown so mid-row-text-selection
drags don't dismiss).
- pause/resume/enable/disable each show a Modal with a consequence
paragraph explaining what hits live traffic ("will keep existing
flows", "will flush VPP's flow table", etc.). The disable commit
button is styled btn-danger red.
- set-weight action shows a Modal with a range slider (0-100,
seeded from the current configured weight, accent-colored live
numeric readout via <output>) plus a flush checkbox and a live-
swapping note/warn paragraph describing what will happen. On
commit, the SPA also updates its local store via
applyConfiguredWeight so the operator sees the new weight
immediately without waiting for the next refresh.
- ProbeHeartbeat is now state-aware: ▶ (play) at rest for up/
down/unknown backends, ⏸ (pause) for paused, ⏹ (stop) for
disabled/removed, ❤️ (heart) during an in-flight probe.
- Drop the probe-done event listener — fast probes (<10ms)
could fire probe-done in the same render tick as probe-start
and the heart would never visibly paint. Each probe-start now
runs a fixed 400ms scale-pop animation on a timer; subsequent
probe-start events reset the timer, so fast cadences produce a
continuous heart pulse.
- Fixed wrapper box (16x14 px, overflow hidden) so the row
doesn't jiggle when the glyph swaps between the narrow ▶/⏸/⏹
text glyphs and the wider ❤️ emoji.
- Brand wordmark changed from "maglev" to "vpp-maglev" and wrapped
in an <a> linking to https://git.ipng.ch/ipng/vpp-maglev. Logo
link changed to https://ipng.ch/. Both open in a new tab with
rel="noopener".
- .gitignore fix: `frontend`, `maglevc`, `maglevd` were matching
ANY file or directory with those names anywhere in the tree,
silently ignoring cmd/frontend and friends. Anchored with
leading slashes so only repo-root build artifacts match.
|
||
|
Pim van Pelt
|
25e9d79aba |
Frontend: live clocks, admin mode, backend actions; packaging polish
Builds on the maglev-frontend component introduced in
|
||
|
Pim van Pelt
|
284b4cc9a4 |
New maglev-frontend component; promote LB sync events to INFO
Introduces maglev-frontend, a responsive, real-time web dashboard for one
or more running maglevd instances. Source lives at cmd/frontend/; the
built binary is maglev-frontend. It is a single Go process with the
SolidJS SPA embedded via //go:embed — no runtime file dependencies.
Architecture
- One persistent gRPC connection per configured maglevd (-server A,B,C).
Each connection runs three background loops: a WatchEvents stream
subscribed at log_level=debug for live events, a 30s refresh loop as
a safety net for drift, and a 5s health loop that surfaces connection
drops quickly.
- In-process pub/sub broker with a 30s / 2000-event replay ring using
<epoch>-<seq> monotonic IDs. Short browser reconnects (nginx idle,
wifi flap, laptop wake) silently replay buffered events via the
EventSource Last-Event-ID header; longer outages or frontend restarts
fall through to a "resync" event that triggers a full state refetch.
- HTTP surface: /view/ (SPA), /view/api/state, /view/api/state/{name},
/view/api/maglevds, /view/api/version, /view/api/events (SSE),
/healthz, and an /admin/* placeholder returning 501 for a future
basic-auth mutation surface.
- SSE handler follows the full operational checklist: retry hint, 15s
: ping heartbeat, Flush after every write, r.Context().Done() teardown,
X-Accel-Buffering: no, and no gzip.
SolidJS SPA (cmd/frontend/web/, Vite + TypeScript)
- solid-js/store for a reactive per-maglevd state tree; reducers apply
backend transitions, maglevd-status flips, and resync refetches.
- Scope selector tabs for multi-maglevd support, per-maglevd frontend
cards with pool tables showing state, configured weight, effective
weight, and last-transition age.
- ProbeHeartbeat component turns a middle-dot into ❤️ on probe-start and
back on probe-done, driven by real log events; fixed-size wrapper so
the emoji swap doesn't jiggle the row.
- Flash wrapper animates any primitive on change (1s yellow fade via
Web Animations API, skipped on first mount). Wired into the state
badge, configured weight, and effective weight columns.
- DebugPanel: chronological rolling event tail with tail-style auto-
scroll, pause/resume, and scope/firehose filter. Syntactic highlight
for vpp-lb-sync-* events with fixed-order attribute formatting.
- Live effective_weight updates: vpp-lb-sync-as-added/removed/weight-
updated log events are routed through a reducer that walks the
snapshot's pool rows and sets effective_weight on every match
without waiting for the 30s refresh.
- Header shows build version + commit with build date in a tooltip,
fetched once from /view/api/version on mount.
- Prettier wired in as the web-side fixstyle; make fixstyle now tidies
both Go and web in one shot via a new fixstyle-web target.
Per-mutation VPP LB sync logging
- Promotes the addVIP/delVIP/addAS/delAS/setASWeight helpers from
slog.Debug to slog.Info and renames them from vpp-lbsync-* to
vpp-lb-sync-{vip-added,vip-removed,as-added,as-removed,as-weight-
updated}. Matching rename for vpp-lb-sync-start / -done / -error /
-vip-recreate. The Prometheus metric name (maglev_vpp_lbsync_total)
is left alone to preserve dashboards.
- setASWeight now takes the prior weight so the event can emit
from=X to=Y and the UI can show the delta.
- The vip field in every event is the bare address (no /32 or /128
mask), matching the CLI output style.
- Any listener on the gRPC WatchEvents stream — CLI watch events or
maglev-frontend — now sees every VIP/AS dataplane change in real
time without needing to raise the log level.
Build and tooling
- Makefile: maglev-frontend added to BINARIES; build / build-amd64 /
build-arm64 emit the binary alongside maglevd and maglevc. A new
maglev-frontend-web target rebuilds the SolidJS bundle via npm.
- web/dist/ is tracked so a bare `go build` keeps working for Go-only
contributors and CI.
- .gitignore skips cmd/frontend/web/node_modules/.
Stability fixes
- maglevd's WatchEvents synthetic replay events (from==to, at_unix_ns=0)
were corrupting the frontend's LastTransition cache with at=0,
rendering as "20555d ago" in the browser. Client now skips synthetic
events: the cache comes from refreshAll and doesn't need them.
- Frontends, Backends, and HealthChecks are now served in the order
returned by the corresponding List* RPC instead of Go map iteration
order, so reloads and refreshes keep the SPA stable.
|