mixclient: Detect exited csppsolver processes

If the csppsolver process has unexpectedly exited after successful startup (due to crash, signal, oom-killer, etc.) unset the PR flag indicating root solving support for future mixes. Wait for other peers to publish roots (if any do), otherwise our peer will be (correctly) blamed for not publishing roots.
decred · Nov 8, 2024 · 0b7f593 · 0b7f593
1 parent 6447650
commit 0b7f593
Show file tree

Hide file tree

Showing 3 changed files with 26 additions and 9 deletions.
diff --git a/mixing/go.mod b/mixing/go.mod
@@ -3,7 +3,7 @@ module github.com/decred/dcrd/mixing
 go 1.18
 
 require (
-	decred.org/cspp/v2 v2.2.0
+	decred.org/cspp/v2 v2.3.0
 	github.com/companyzero/sntrup4591761 v0.0.0-20220309191932-9e0f3af2f07a
 	github.com/davecgh/go-spew v1.1.1
 	github.com/decred/dcrd/chaincfg/chainhash v1.0.4

diff --git a/mixing/go.sum b/mixing/go.sum
@@ -1,5 +1,5 @@
-decred.org/cspp/v2 v2.2.0 h1:VSOUC1w0Wo+QOGS0r1XO6TLnO16X67KuvpDmRRYyr08=
-decred.org/cspp/v2 v2.2.0/go.mod h1:9nO3bfvCheOPIFZw5f6sRQ42CjBFB5RKSaJ9Iq6G4MA=
+decred.org/cspp/v2 v2.3.0 h1:GC8emJnLbOVAkgBTHK/1wy6o/m0AVsN1r4m1ZnZZWjo=
+decred.org/cspp/v2 v2.3.0/go.mod h1:9nO3bfvCheOPIFZw5f6sRQ42CjBFB5RKSaJ9Iq6G4MA=
 github.com/agl/ed25519 v0.0.0-20170116200512-5312a6153412 h1:w1UutsfOrms1J05zt7ISrnJIXKzwaspym5BTKGx93EI=
 github.com/agl/ed25519 v0.0.0-20170116200512-5312a6153412/go.mod h1:WPjqKcmVOxf0XSf3YxCJs6N6AOSrOx3obionmG7T0y0=
 github.com/companyzero/sntrup4591761 v0.0.0-20220309191932-9e0f3af2f07a h1:clYxJ3Os0EQUKDDVU8M0oipllX0EkuFNBfhVQuIfyF0=

diff --git a/mixing/mixclient/client.go b/mixing/mixclient/client.go
@@ -16,6 +16,7 @@ import (
 	"runtime"
 	"sort"
 	"sync"
+	"sync/atomic"
 	"time"
 
 	"decred.org/cspp/v2/solverrpc"
@@ -264,6 +265,9 @@ type queueWork struct {
 
 // Client manages local mixing client sessions.
 type Client struct {
+	// atomics
+	atomicPRFlags uint32
+
 	wallet  Wallet
 	mixpool *mixpool.Pool
 
@@ -281,8 +285,7 @@ type Client struct {
 	blake256Hasher   hash.Hash
 	blake256HasherMu sync.Mutex
 
-	epoch   time.Duration
-	prFlags byte
+	epoch time.Duration
 
 	logger slog.Logger
 
@@ -300,14 +303,14 @@ func NewClient(w Wallet) *Client {
 
 	height, _ := w.BestBlock()
 	return &Client{
+		atomicPRFlags:  uint32(prFlags),
 		wallet:         w,
 		mixpool:        w.Mixpool(),
 		pairings:       make(map[string]*pairedSessions),
 		warming:        make(chan struct{}),
 		workQueue:      make(chan *queueWork, runtime.NumCPU()),
 		blake256Hasher: blake256.New(),
 		epoch:          w.Mixpool().Epoch(),
-		prFlags:        prFlags,
 		height:         height,
 	}
 }
@@ -718,10 +721,11 @@ func (c *Client) Dicemix(ctx context.Context, cj *CoinJoin) error {
 		return err
 	}
 
+	prFlags := byte(atomic.LoadUint32(&c.atomicPRFlags))
 	pr, err := wire.NewMsgMixPairReq(*p.id, cj.prExpiry, cj.mixValue,
 		string(mixing.ScriptClassP2PKHv0), cj.tx.Version,
 		cj.tx.LockTime, cj.mcount, cj.inputValue, cj.prUTXOs,
-		cj.change, c.prFlags, pairingFlags)
+		cj.change, prFlags, pairingFlags)
 	if err != nil {
 		return err
 	}
@@ -1694,7 +1698,8 @@ func (c *Client) run(ctx context.Context, ps *pairedSessions, madePairing *bool)
 }
 
 func (c *Client) solvesRoots() bool {
-	return c.prFlags&mixing.PRFlagCanSolveRoots != 0
+	prFlags := byte(atomic.LoadUint32(&c.atomicPRFlags))
+	return prFlags&mixing.PRFlagCanSolveRoots != 0
 }
 
 // roots returns the solutions to the slot reservation polynomial.  If the
@@ -1704,8 +1709,20 @@ func (c *Client) solvesRoots() bool {
 func (c *Client) roots(ctx context.Context, seenSRs []chainhash.Hash,
 	sesRun *sessionRun, a []*big.Int, F *big.Int, publishedRoots chan struct{}) ([]*big.Int, error) {
 
-	if c.solvesRoots() {
+	switch {
+	case c.solvesRoots():
 		roots, err := solverrpc.Roots(a, F)
+		if errors.Is(err, solverrpc.ErrSolverProcessExited) {
+			// Unset root-solving capability and future advertisement
+			prFlags := byte(atomic.LoadUint32(&c.atomicPRFlags))
+			prFlags &^= mixing.PRFlagCanSolveRoots
+			atomic.StoreUint32(&c.atomicPRFlags, uint32(prFlags))
+
+			// Wait for other peers to publish roots as a fallback
+			// in this run.  If we are the only solver, we will be
+			// blamed for not publishing any.
+			break
+		}
 		if err != nil || len(roots) != len(a)-1 {
 			c.forLocalPeers(ctx, sesRun, func(p *peer) error {
 				p.triggeredBlame = true