Files
tailscale/tka/sync_test.go
T
Alex ChanandAlex Chan b5fb042501 tka/sync: add regression test for compacted nodes on forked chains
We previously identified sync failures that occur when a node falls behind
the remote, and compacts away most its local state. We fixed the underlying
issue in #19444, but that PR only tested the basic scenario where the
local chain is a direct ancestor of the remote chain.

This patch adds an explicit regression test for the case where a node is on
a fork (that is, its HEAD is not part of the remote's active chain).

Although #19444 happened to cover this case, other proposed patches did not
handle the forked state. Adding this test locks in the behaviour and prevents
future sync regressions in this area.

Also, add a shared helper for writing this sort of TKA sync test.

Updates tailscale/corp#40404

Change-Id: I78fdc6beaf71392edf11806197f126db48886f93
Signed-off-by: Alex Chan <alexc@tailscale.com>
2026-07-24 16:42:42 +01:00

604 lines
20 KiB
Go

// Copyright (c) Tailscale Inc & contributors
// SPDX-License-Identifier: BSD-3-Clause
package tka
import (
"bytes"
"fmt"
"strings"
"testing"
"time"
"github.com/google/go-cmp/cmp"
"tailscale.com/tstest"
"tailscale.com/util/must"
)
// getSyncOffer returns a SyncOffer for the given Chonk.
func getSyncOffer(t *testing.T, storage Chonk) SyncOffer {
t.Helper()
a, err := Open(storage)
if err != nil {
t.Fatal(err)
}
offer, err := a.SyncOffer(storage)
if err != nil {
t.Fatal(err)
}
return offer
}
func TestSyncOffer(t *testing.T) {
fakeState := &State{
Keys: []Key{{Kind: Key25519, Votes: 1}},
DisablementValues: [][]byte{bytes.Repeat([]byte{1}, 32)},
}
checkpointTemplate := optTemplate("checkpoint", AUM{MessageKind: AUMCheckpoint, State: fakeState})
// If we have a small chain with just a handful of AUMs, the SyncOffer
// contains the current HEAD and the first checkpoint.
t.Run("short-chain", func(t *testing.T) {
c := newTestchain(t, `A1 -> A2 -> A3 -> A4 -> A5`)
got := getSyncOffer(t, c.Chonk())
// A SyncOffer includes the first checkpoint.
want := SyncOffer{
Head: c.AUMHashes["A5"],
Ancestors: []AUMHash{
c.AUMHashes["A1"],
},
}
if diff := cmp.Diff(want, got); diff != "" {
t.Errorf("SyncOffer diff (-want, +got):\n%s", diff)
}
})
// If the chain contains multiple checkpoints, the SyncOffer includes
// all of them.
t.Run("chain-with-multiple-checkpoints", func(t *testing.T) {
c := newTestchain(t, `
A1 -> A2 -> A3 -> A4 -> A5 -> A6 -> A7 -> A8 -> A9 -> A10
A10 -> A11 -> A12 -> A13 -> A14 -> A15 -> A16 -> A17 -> A18
A18 -> A19 -> A20 -> A21 -> A22 -> A23 -> A24 -> A25 -> A26
A26 -> A27 -> A28 -> A29 -> A30 -> A31 -> A32 -> A33 -> A34
A34 -> A35 -> A36 -> A37 -> A38 -> A39 -> A40 -> A41 -> A42
A42 -> A43 -> A45 -> A46 -> A47 -> A48 -> A49 -> A50 -> A51
A51 -> A52 -> A53 -> A54 -> A55
A1.template = checkpoint
A11.template = checkpoint
A21.template = checkpoint
A31.template = checkpoint
A41.template = checkpoint
A51.template = checkpoint
`, checkpointTemplate)
got := getSyncOffer(t, c.Chonk())
// A SyncOffer includes the first checkpoint.
want := SyncOffer{
Head: c.AUMHashes["A55"],
Ancestors: []AUMHash{
c.AUMHashes["A51"],
c.AUMHashes["A41"],
c.AUMHashes["A31"],
c.AUMHashes["A21"],
c.AUMHashes["A11"],
c.AUMHashes["A1"],
},
}
if diff := cmp.Diff(want, got); diff != "" {
t.Errorf("SyncOffer diff (-want, +got):\n%s", diff)
}
})
// The size of a SyncOffer does not grow without bound as the number of AUMs increases.
t.Run("long-chain-size-is-bounded", func(t *testing.T) {
size := 1800
// Build a template string with a checkpoint every 50 AUMs.
var sb strings.Builder
sb.WriteString("A")
for i := range size {
sb.WriteString(fmt.Sprintf(" -> A%d", i))
}
for i := range size {
if i%50 == 0 {
sb.WriteString(fmt.Sprintf("\nA%d.template = checkpoint", i))
}
}
c := newTestchain(t, sb.String(), checkpointTemplate)
got := getSyncOffer(t, c.Chonk())
// We expect the SyncOffer to include:
//
// - the latest AUM as the HEAD
// - the checkpoints from the last 1000 AUMs (maxSyncHeadIntersectionIter)
// - the oldest AUM in storage
//
want := SyncOffer{
Head: c.AUMHashes["A1799"],
Ancestors: []AUMHash{
c.AUMHashes["A1750"], c.AUMHashes["A1700"], c.AUMHashes["A1650"],
c.AUMHashes["A1600"], c.AUMHashes["A1550"], c.AUMHashes["A1500"],
c.AUMHashes["A1450"], c.AUMHashes["A1400"], c.AUMHashes["A1350"],
c.AUMHashes["A1300"], c.AUMHashes["A1250"], c.AUMHashes["A1200"],
c.AUMHashes["A1150"], c.AUMHashes["A1100"], c.AUMHashes["A1050"],
c.AUMHashes["A1000"], c.AUMHashes["A950"], c.AUMHashes["A900"],
c.AUMHashes["A850"], c.AUMHashes["A800"], c.AUMHashes["A"],
},
}
if diff := cmp.Diff(want, got); diff != "" {
t.Errorf("SyncOffer diff (-want, +got):\n%s", diff)
}
})
}
func TestComputeSyncIntersection_FastForward(t *testing.T) {
// Node 1 has: A1 -> A2
// Node 2 has: A1 -> A2 -> A3 -> A4
c := newTestchain(t, `
A1 -> A2 -> A3 -> A4
`)
a1H, a2H := c.AUMHashes["A1"], c.AUMHashes["A2"]
chonk1 := c.ChonkWith("A1", "A2")
offer1 := getSyncOffer(t, chonk1)
chonk2 := c.Chonk() // All AUMs
offer2 := getSyncOffer(t, chonk2)
// Node 1 only knows about the first two nodes, so the head of n2 is
// alien to it.
t.Run("n1", func(t *testing.T) {
got, err := computeSyncIntersection(chonk1, offer1, offer2)
if err != nil {
t.Fatalf("computeSyncIntersection() failed: %v", err)
}
want := &intersection{
tailIntersection: &a1H,
}
if diff := cmp.Diff(want, got, cmp.AllowUnexported(intersection{})); diff != "" {
t.Errorf("intersection diff (-want, +got):\n%s", diff)
}
})
// Node 2 knows about the full chain, so it can see that the head of n1
// intersects with a subset of its chain (a Head Intersection).
t.Run("n2", func(t *testing.T) {
got, err := computeSyncIntersection(chonk2, offer2, offer1)
if err != nil {
t.Fatalf("computeSyncIntersection() failed: %v", err)
}
want := &intersection{
headIntersection: &a2H,
}
if diff := cmp.Diff(want, got, cmp.AllowUnexported(intersection{})); diff != "" {
t.Errorf("intersection diff (-want, +got):\n%s", diff)
}
})
}
func TestComputeSyncIntersection_ForkSmallDiff(t *testing.T) {
// The number of nodes in the chain is longer than ancestorSkipStart,
// so that during sync both nodes are able to find a common ancestor
// which was later than A1.
c := newTestchain(t, `
A1 -> A2 -> A3 -> A4 -> A5 -> A6 -> A7 -> A8 -> A9 -> A10
| -> F1
// Make F1 different to A9.
// hashSeed is chosen such that the hash is higher than A9.
F1.hashSeed = 7
`)
// Node 1 has: A1 -> A2 -> A3 -> A4 -> A5 -> A6 -> A7 -> A8 -> F1
// Node 2 has: A1 -> A2 -> A3 -> A4 -> A5 -> A6 -> A7 -> A8 -> A9 -> A10
f1H, a9H := c.AUMHashes["F1"], c.AUMHashes["A9"]
if bytes.Compare(f1H[:], a9H[:]) < 0 {
t.Fatal("failed assert: h(a9) > h(f1H)\nTweak hashSeed till this passes")
}
chonk1 := c.ChonkWith("A1", "A2", "A3", "A4", "A5", "A6", "A7", "A8", "F1")
offer1 := getSyncOffer(t, chonk1)
chonk2 := c.ChonkWith("A1", "A2", "A3", "A4", "A5", "A6", "A7", "A8", "A9", "A10")
offer2 := getSyncOffer(t, chonk2)
// Node 1 only knows about the first eight nodes, so the head of n2 is
// alien to it.
t.Run("n1", func(t *testing.T) {
// n2 has 10 nodes, so the first common ancestor is the genesis AUM
wantIntersection := c.AUMHashes["A1"]
got, err := computeSyncIntersection(chonk1, offer1, offer2)
if err != nil {
t.Fatalf("computeSyncIntersection() failed: %v", err)
}
want := &intersection{
tailIntersection: &wantIntersection,
}
if diff := cmp.Diff(want, got, cmp.AllowUnexported(intersection{})); diff != "" {
t.Errorf("intersection diff (-want, +got):\n%s", diff)
}
})
// Node 2 knows about the full chain but doesn't recognize the head.
t.Run("n2", func(t *testing.T) {
// n1 has 9 nodes, so the first common ancestor is the genesis AUM
wantIntersection := c.AUMHashes["A1"]
got, err := computeSyncIntersection(chonk2, offer2, offer1)
if err != nil {
t.Fatalf("computeSyncIntersection() failed: %v", err)
}
want := &intersection{
tailIntersection: &wantIntersection,
}
if diff := cmp.Diff(want, got, cmp.AllowUnexported(intersection{})); diff != "" {
t.Errorf("intersection diff (-want, +got):\n%s", diff)
}
})
}
func TestMissingAUMs_FastForward(t *testing.T) {
// Node 1 has: A1 -> A2
// Node 2 has: A1 -> A2 -> A3 -> A4
c := newTestchain(t, `
A1 -> A2 -> A3 -> A4
A1.hashSeed = 1
A2.hashSeed = 2
A3.hashSeed = 3
A4.hashSeed = 4
`)
chonk1 := c.ChonkWith("A1", "A2")
n1, err := Open(chonk1)
if err != nil {
t.Fatal(err)
}
offer1, err := n1.SyncOffer(chonk1)
if err != nil {
t.Fatal(err)
}
chonk2 := c.Chonk() // All AUMs
n2, err := Open(chonk2)
if err != nil {
t.Fatal(err)
}
offer2, err := n2.SyncOffer(chonk2)
if err != nil {
t.Fatal(err)
}
// Node 1 only knows about the first two nodes, so the head of n2 is
// alien to it. As such, it should send history from the newest ancestor,
// A1 (if the chain was longer there would be one in the middle).
t.Run("n1", func(t *testing.T) {
got, err := n1.MissingAUMs(chonk1, offer2)
if err != nil {
t.Fatalf("MissingAUMs() failed: %v", err)
}
// Both sides have A1, so the only AUM that n2 might not have is
// A2.
want := []AUM{c.AUMs["A2"]}
if diff := cmp.Diff(want, got); diff != "" {
t.Errorf("MissingAUMs diff (-want, +got):\n%s", diff)
}
})
// Node 2 knows about the full chain, so it can see that the head of n1
// intersects with a subset of its chain (a Head Intersection).
t.Run("n2", func(t *testing.T) {
got, err := n2.MissingAUMs(chonk2, offer1)
if err != nil {
t.Fatalf("MissingAUMs() failed: %v", err)
}
want := []AUM{
c.AUMs["A3"],
c.AUMs["A4"],
}
if diff := cmp.Diff(want, got); diff != "" {
t.Errorf("MissingAUMs diff (-want, +got):\n%s", diff)
}
})
}
func TestMissingAUMs_Fork(t *testing.T) {
// Node 1 has: A1 -> A2 -> A3 -> F1
// Node 2 has: A1 -> A2 -> A3 -> A4
c := newTestchain(t, `
A1 -> A2 -> A3 -> A4
| -> F1
A1.hashSeed = 1
A2.hashSeed = 2
A3.hashSeed = 3
A4.hashSeed = 4
`)
chonk1 := c.ChonkWith("A1", "A2", "A3", "F1")
n1, err := Open(chonk1)
if err != nil {
t.Fatal(err)
}
offer1, err := n1.SyncOffer(chonk1)
if err != nil {
t.Fatal(err)
}
chonk2 := c.ChonkWith("A1", "A2", "A3", "A4")
n2, err := Open(chonk2)
if err != nil {
t.Fatal(err)
}
offer2, err := n2.SyncOffer(chonk2)
if err != nil {
t.Fatal(err)
}
t.Run("n1", func(t *testing.T) {
got, err := n1.MissingAUMs(chonk1, offer2)
if err != nil {
t.Fatalf("MissingAUMs() failed: %v", err)
}
// Both sides have A1, so n1 will send everything it knows from
// there to head.
want := []AUM{
c.AUMs["A2"],
c.AUMs["A3"],
c.AUMs["F1"],
}
if diff := cmp.Diff(want, got); diff != "" {
t.Errorf("MissingAUMs diff (-want, +got):\n%s", diff)
}
})
t.Run("n2", func(t *testing.T) {
got, err := n2.MissingAUMs(chonk2, offer1)
if err != nil {
t.Fatalf("MissingAUMs() failed: %v", err)
}
// Both sides have A1, so n2 will send everything it knows from
// there to head.
want := []AUM{
c.AUMs["A2"],
c.AUMs["A3"],
c.AUMs["A4"],
}
if diff := cmp.Diff(want, got); diff != "" {
t.Errorf("MissingAUMs diff (-want, +got):\n%s", diff)
}
})
}
func TestSyncSimpleE2E(t *testing.T) {
pub, priv := testingKey25519(t, 1)
key := Key{Kind: Key25519, Public: pub, Votes: 2}
c := newTestchain(t, `
G1 -> L1 -> L2 -> L3
G1.template = genesis
`,
genesisTemplate(key),
optKey("key", key, priv),
optSignAllUsing("key"))
nodeStorage := ChonkMem()
node, err := Bootstrap(nodeStorage, c.AUMs["G1"])
if err != nil {
t.Fatalf("node Bootstrap() failed: %v", err)
}
controlStorage := c.Chonk()
control, err := Open(controlStorage)
if err != nil {
t.Fatalf("control Open() failed: %v", err)
}
// Control knows the full chain, node only knows the genesis. Let's see
// if they can sync.
nodeOffer, err := node.SyncOffer(nodeStorage)
if err != nil {
t.Fatal(err)
}
controlAUMs, err := control.MissingAUMs(controlStorage, nodeOffer)
if err != nil {
t.Fatalf("control.MissingAUMs(%v) failed: %v", nodeOffer, err)
}
if err := node.Inform(nodeStorage, controlAUMs); err != nil {
t.Fatalf("node.Inform(%v) failed: %v", controlAUMs, err)
}
if cHash, nHash := control.Head(), node.Head(); cHash != nHash {
t.Errorf("node & control are not synced: c=%x, n=%x", cHash, nHash)
}
}
// TestSyncFromFarBehind checks that nodes with compacted state can still find
// a common ancestor when the remote is significantly ahead.
//
// We simulate a node that has compacted its early history and is now ~500 AUMs
// behind the control plane, a distance that previously caused exponential sampling
// in SyncOffer to skip the node's entire local history.
//
// Regression test for http://go/corp/40404
func TestSyncFromFarBehind(t *testing.T) {
pub1, priv1 := testingKey25519(t, 1)
signer1 := signer25519(priv1)
key1 := Key{Kind: Key25519, Public: pub1, Votes: 2}
// Setup: persistentAuthority (control plane) vs compactingAuthority (client node).
state := State{
Keys: []Key{key1},
DisablementValues: [][]byte{DisablementKDF([]byte{1, 2, 3})},
}
persistentStorage, compactingStorage := ChonkMem(), ChonkMem()
persistentSize := func() int { return len(must.Get(persistentStorage.AllAUMs())) }
compactingSize := func() int { return len(must.Get(compactingStorage.AllAUMs())) }
// Backdate the clock on the compactingStorage so all AUMs will be old enough
// to be considered for compacting.
clock := tstest.NewClock(tstest.ClockOpts{
Start: time.Now().Add(-(CompactionDefaults.MinAge + 24*time.Hour)),
})
compactingStorage.SetClock(clock)
persistentAuthority, genesisAUM := must.Get2(Create(persistentStorage, state, signer1))
compactingAuthority := must.Get(Bootstrap(compactingStorage, genesisAUM))
// 1. Generate enough history to trigger checkpoints.
persistentNode := CreateSeedNode(t, persistentAuthority, persistentStorage)
compactingNode := CreateSeedNode(t, compactingAuthority, compactingStorage)
SeedAUMs(t, checkpointEvery*2, signer1, persistentNode, compactingNode)
t.Logf("genesis and first batch of AUMs: persistent = %d, compacting = %d", persistentSize(), compactingSize())
// 2. Compact the node state.
//
// It now has a different 'oldestAncestor' than the control plane.
beforeCompacting := compactingSize()
must.Do(compactingAuthority.Compact(compactingStorage, CompactionDefaults))
afterCompacting := compactingSize()
if beforeCompacting == afterCompacting {
t.Errorf("expected Compact to reduce the number of AUMs, but unchanged: size = %d", afterCompacting)
}
// 3. Advance the control plane far beyond the node.
//
// As of 2026-04-17, the largest TKA has ~750 AUMs.
//
// If you keep increasing this number, eventually the sync will fail because you
// hit the hard-coded limits on iteration during the sync process.
SeedAUMs(t, compactingSize()-persistentSize()+800, signer1, persistentNode)
t.Logf("post-compacting and extra AUMs: persistent = %d, compacting = %d", persistentSize(), compactingSize())
// 4. Verify Intersection.
// The node should find an intersection even with a 500-AUM gap.
persistentOffer := must.Get(persistentAuthority.SyncOffer(persistentStorage))
compactingOffer := must.Get(compactingAuthority.SyncOffer(compactingStorage))
if _, err := compactingAuthority.MissingAUMs(compactingStorage, persistentOffer); err != nil {
t.Errorf("node failed to find intersection with far-ahead control plane: %v", err)
}
// 5. Check that the persistent authority can find an intersection with the
// compacting authority, and has missing AUMs to send it.
missing, err := persistentAuthority.MissingAUMs(persistentStorage, compactingOffer)
if len(missing) == 0 {
t.Errorf("control plane did not find any missing AUMs for node")
}
if err != nil {
t.Errorf("control plane failed to find missing AUMs for node: %v", err)
}
}
// TestSyncFromFarBehindFork checks that nodes with compacted state that have
// also branched from the active chain can still find a common ancestor when
// the remote is significantly ahead of the inersection point.
//
// We simulate a node that has compacted its early history and is now ~500 AUMs
// behind the control plane, plus a few AUMs extra, a distance that previously
// caused exponential sampling in SyncOffer to skip the node's entire local history.
//
// Regression test for http://go/corp/40404
func TestSyncFromFarBehindFork(t *testing.T) {
// Set up two signing keys. They have a different number of votes, so if there's
// a fork, the chain with the winning key will take precedence.
majorityPub, majorityPriv := testingKey25519(t, 1)
losingPub, losingPriv := testingKey25519(t, 2)
winningSigner := signer25519(majorityPriv)
losingSigner := signer25519(losingPriv)
losingKey := Key{Kind: Key25519, Public: majorityPub, Votes: 1}
winningKey := Key{Kind: Key25519, Public: losingPub, Votes: 2}
// Setup: persistentAuthority (control plane) vs compactingAuthority (client node).
state := State{
Keys: []Key{losingKey, winningKey},
DisablementValues: [][]byte{DisablementKDF([]byte{1, 2, 3})},
}
persistentStorage, compactingStorage := ChonkMem(), ChonkMem()
persistentSize := func() int { return len(must.Get(persistentStorage.AllAUMs())) }
compactingSize := func() int { return len(must.Get(compactingStorage.AllAUMs())) }
// Backdate the clock on the compactingStorage so all AUMs will be old enough
// to be considered for compacting.
clock := tstest.NewClock(tstest.ClockOpts{
Start: time.Now().Add(-(CompactionDefaults.MinAge + 24*time.Hour)),
})
compactingStorage.SetClock(clock)
persistentAuthority, genesisAUM := must.Get2(Create(persistentStorage, state, winningSigner))
compactingAuthority := must.Get(Bootstrap(compactingStorage, genesisAUM))
// 1. Generate enough history to trigger checkpoints.
persistentNode := CreateSeedNode(t, persistentAuthority, persistentStorage)
compactingNode := CreateSeedNode(t, compactingAuthority, compactingStorage)
SeedAUMs(t, checkpointEvery*2, winningSigner, persistentNode, compactingNode)
t.Logf("genesis and first batch of AUMs: persistent = %d, compacting = %d", persistentSize(), compactingSize())
// 2. Compact the node state.
//
// It now has a different 'oldestAncestor' than the control plane.
beforeCompacting := compactingSize()
must.Do(compactingAuthority.Compact(compactingStorage, CompactionDefaults))
afterCompacting := compactingSize()
if beforeCompacting == afterCompacting {
t.Errorf("expected Compact to reduce the number of AUMs, but unchanged: size = %d", afterCompacting)
}
// 2. Advance the node state slightly beyond the control plane, using the
// losing signer.
SeedAUMs(t, 1, losingSigner, compactingNode)
// 3. Advance the control plane far beyond the node, using the winning signer.
//
// Now the node is forked from the control plane state, and the control plane's
// chain will win because its chain was signed by a key with more votes.
//
// As of 2026-04-17, the largest TKA has ~750 AUMs.
//
// If you keep increasing this number, eventually the sync will fail because you
// hit the hard-coded limits on iteration during the sync process.
SeedAUMs(t, compactingSize()-persistentSize()+800, winningSigner, persistentNode)
t.Logf("post-compacting and extra AUMs: persistent = %d, compacting = %d", persistentSize(), compactingSize())
// 4. Verify Intersection.
// The node should find an intersection even with a 500-AUM gap.
persistentOffer := must.Get(persistentAuthority.SyncOffer(persistentStorage))
compactingOffer := must.Get(compactingAuthority.SyncOffer(compactingStorage))
if _, err := compactingAuthority.MissingAUMs(compactingStorage, persistentOffer); err != nil {
t.Errorf("node failed to find intersection with far-ahead control plane: %v", err)
}
// 5. Check that the persistent authority can find an intersection with the
// compacting authority, and has missing AUMs to send it.
missing, err := persistentAuthority.MissingAUMs(persistentStorage, compactingOffer)
if len(missing) == 0 {
t.Errorf("control plane did not find any missing AUMs for node")
}
if err != nil {
t.Errorf("control plane failed to find missing AUMs for node: %v", err)
}
}