From a0b18e121ba706d0161d09bbe4043624f99f06a3 Mon Sep 17 00:00:00 2001
From: Keegan Carruthers-Smith <keegan.csmith@gmail.com>
Date: Fri, 27 Mar 2026 11:29:16 +0200
Subject: [PATCH] search: use structured logging for crashed shards

We recently had some crashes in production while experimenting with
ZOEKT_RE2_THRESHOLD_BYTES and found it hard to consume the stack traces
since they were over multiple lines.
---
 search/shards.go      | 25 +++++++++++++++++++++++--
 search/shards_test.go |  5 +++++
 2 files changed, 28 insertions(+), 2 deletions(-)

diff --git a/search/shards.go b/search/shards.go
index bee903f99..64668ba41 100644
--- a/search/shards.go
+++ b/search/shards.go
@@ -30,6 +30,7 @@ import (
 
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promauto"
+	sglog "github.com/sourcegraph/log"
 	"go.uber.org/atomic"
 	"golang.org/x/sync/semaphore"
 
@@ -41,6 +42,10 @@ import (
 )
 
 var (
+	shardRecoveryLogger = sync.OnceValue(func() sglog.Logger {
+		return sglog.Scoped("searchShards")
+	})
+
 	metricShardsLoaded = promauto.NewGauge(prometheus.GaugeOpts{
 		Name: "zoekt_shards_loaded",
 		Help: "The number of shards currently loaded",
@@ -927,12 +932,28 @@ func copyFiles(sr *zoekt.SearchResult) {
 	}
 }
 
+func logShardCrash(operation string, s zoekt.Searcher, recovered any, stack []byte) {
+	fields := []sglog.Field{
+		sglog.String("operation", operation),
+		sglog.String("shard", s.String()),
+		sglog.String("stacktrace", string(stack)),
+	}
+
+	if err, ok := recovered.(error); ok {
+		fields = append(fields, sglog.Error(err))
+	} else {
+		fields = append(fields, sglog.String("panic", fmt.Sprint(recovered)))
+	}
+
+	shardRecoveryLogger().Error("crashed shard", fields...)
+}
+
 func searchOneShard(ctx context.Context, s zoekt.Searcher, q query.Q, opts *zoekt.SearchOptions) (sr *zoekt.SearchResult, err error) {
 	metricSearchShardRunning.Inc()
 	defer func() {
 		metricSearchShardRunning.Dec()
 		if e := recover(); e != nil {
-			log.Printf("[ERROR] crashed shard: %s: %#v, %s", s, e, debug.Stack())
+			logShardCrash("search", s, e, debug.Stack())
 
 			if sr == nil {
 				sr = &zoekt.SearchResult{}
@@ -954,7 +975,7 @@ func listOneShard(ctx context.Context, s zoekt.Searcher, q query.Q, opts *zoekt.
 	defer func() {
 		metricListShardRunning.Dec()
 		if r := recover(); r != nil {
-			log.Printf("[ERROR] crashed shard: %s: %s, %s", s.String(), r, debug.Stack())
+			logShardCrash("list", s, r, debug.Stack())
 			sink <- shardListResult{
 				&zoekt.RepoList{Crashes: 1}, nil,
 			}
diff --git a/search/shards_test.go b/search/shards_test.go
index 668075b1f..ed1b7a733 100644
--- a/search/shards_test.go
+++ b/search/shards_test.go
@@ -36,6 +36,7 @@ import (
 	"github.com/google/go-cmp/cmp"
 	"github.com/google/go-cmp/cmp/cmpopts"
 	"github.com/grafana/regexp"
+	sglog "github.com/sourcegraph/log"
 
 	"github.com/sourcegraph/zoekt/index"
 
@@ -75,6 +76,10 @@ func TestCrashResilience(t *testing.T) {
 	log.SetOutput(out)
 	defer log.SetOutput(oldOut)
 
+	oldShardRecoveryLogger := shardRecoveryLogger
+	shardRecoveryLogger = sglog.NoOp
+	defer func() { shardRecoveryLogger = oldShardRecoveryLogger }()
+
 	ss := newShardedSearcher(2)
 	ss.ranked.Store([]*rankedShard{{Searcher: &crashSearcher{}}})