Disk buffer for batches VictoriaLogs cannot take, lossless Docker positions, non-blocking reverse DNS

- Batches that fail go to /data/spool (SPOOL_MAX_MB, 1 GiB by default) and
  are sent again oldest first; retries no longer block the store loop and
  follow the shutdown context.
- Docker and host logs wait for room in a full queue instead of being
  dropped; the Docker position only moves once a line is stored or spooled.
- Reverse DNS no longer holds up the syslog listeners, with an LRU cache
  and a cap on concurrent lookups.

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

rdns.go

@@ -1,6 +1,7 @@
 package main
 
 import (
+	"container/list"
 	"context"
 	"net"
 	"strings"
@@ -15,17 +16,26 @@ type ReverseDNS struct {
 	posTTL  time.Duration // cache duration of a found name
 	negTTL  time.Duration // cache duration of "no name"
 
+	lookups chan struct{} // limits the lookups running at once
+	waiters chan struct{} // limits the messages waiting for a lookup (Resolve)
+
 	mu    sync.Mutex
-	cache map[string]*rdnsEntry
+	cache map[string]*list.Element // ip -> element of lru
+	lru   *list.List               // *rdnsEntry, most recently used first
 }
 
 type rdnsEntry struct {
+	ip      string
 	name    string
 	expires time.Time
 	done    chan struct{} // closed once the lookup has finished
 }
 
-const rdnsMaxEntries = 10000
+const (
+	rdnsMaxEntries = 10000
+	rdnsMaxLookups = 64
+	rdnsMaxWaiters = 1024
+)
 
 // NewReverseDNS uses the system resolver, or `server` ("ip" or "ip:port") when set.
 func NewReverseDNS(enabled bool, server string) *ReverseDNS {
@@ -47,7 +57,10 @@ func NewReverseDNS(enabled bool, server string) *ReverseDNS {
 		r:       r,
 		posTTL:  time.Hour,
 		negTTL:  10 * time.Minute,
-		cache:   make(map[string]*rdnsEntry),
+		lookups: make(chan struct{}, rdnsMaxLookups),
+		waiters: make(chan struct{}, rdnsMaxWaiters),
+		cache:   make(map[string]*list.Element),
+		lru:     list.New(),
 	}
 }
 
@@ -60,6 +73,38 @@ func isClosed(ch chan struct{}) bool {
 	}
 }
 
+// entry returns the cache entry of ip, starting its lookup when it is missing
+// or expired, or nil when too many lookups are already running (a flood of
+// unknown addresses). The least recently used entry makes room for a new one.
+func (d *ReverseDNS) entry(ip string) *rdnsEntry {
+	d.mu.Lock()
+	defer d.mu.Unlock()
+	if el := d.cache[ip]; el != nil {
+		e := el.Value.(*rdnsEntry)
+		if !isClosed(e.done) || time.Now().Before(e.expires) {
+			d.lru.MoveToFront(el)
+			return e
+		}
+	}
+	select {
+	case d.lookups <- struct{}{}:
+	default:
+		return nil
+	}
+	if el := d.cache[ip]; el != nil {
+		d.lru.Remove(el)
+	}
+	for d.lru.Len() >= rdnsMaxEntries {
+		old := d.lru.Back()
+		d.lru.Remove(old)
+		delete(d.cache, old.Value.(*rdnsEntry).ip)
+	}
+	e := &rdnsEntry{ip: ip, done: make(chan struct{})}
+	d.cache[ip] = d.lru.PushFront(e)
+	go d.resolve(ip, e)
+	return e
+}
+
 // Lookup returns the name of ip, or "" when ip is not an IP address, has no
 // PTR record, or is not resolved within `wait`. A lookup that takes longer
 // keeps running in the background and fills the cache for later calls.
@@ -67,18 +112,10 @@ func (d *ReverseDNS) Lookup(ip string, wait time.Duration) string {
 	if !d.enabled || net.ParseIP(ip) == nil {
 		return ""
 	}
-	d.mu.Lock()
-	e := d.cache[ip]
-	if e == nil || (isClosed(e.done) && time.Now().After(e.expires)) {
-		if len(d.cache) >= rdnsMaxEntries {
-			d.cache = make(map[string]*rdnsEntry)
-		}
-		e = &rdnsEntry{done: make(chan struct{})}
-		d.cache[ip] = e
-		go d.resolve(ip, e)
+	e := d.entry(ip)
+	if e == nil {
+		return ""
 	}
-	d.mu.Unlock()
-
 	if !isClosed(e.done) {
 		timer := time.NewTimer(wait)
 		defer timer.Stop()
@@ -91,6 +128,43 @@ func (d *ReverseDNS) Lookup(ip string, wait time.Duration) string {
 	return e.name
 }
 
+// Resolve is Lookup without blocking the caller: fn gets the name (or "") at
+// once when it is known, otherwise from a goroutine after at most `wait`. The
+// syslog listeners use it so that a slow DNS server never delays the reading
+// of the next messages.
+func (d *ReverseDNS) Resolve(ip string, wait time.Duration, fn func(name string)) {
+	if !d.enabled || net.ParseIP(ip) == nil {
+		fn("")
+		return
+	}
+	e := d.entry(ip)
+	switch {
+	case e == nil:
+		fn("")
+		return
+	case isClosed(e.done):
+		fn(e.name)
+		return
+	}
+	select {
+	case d.waiters <- struct{}{}:
+	default:
+		fn("") // too many messages waiting already
+		return
+	}
+	go func() {
+		defer func() { <-d.waiters }()
+		timer := time.NewTimer(wait)
+		defer timer.Stop()
+		select {
+		case <-e.done:
+			fn(e.name)
+		case <-timer.C:
+			fn("")
+		}
+	}()
+}
+
 // LookupMany resolves several addresses in parallel; unresolved ones are absent.
 func (d *ReverseDNS) LookupMany(ips []string, wait time.Duration) map[string]string {
 	out := make(map[string]string)
@@ -112,6 +186,7 @@ func (d *ReverseDNS) LookupMany(ips []string, wait time.Duration) map[string]str
 }
 
 func (d *ReverseDNS) resolve(ip string, e *rdnsEntry) {
+	defer func() { <-d.lookups }()
 	ctx, cancel := context.WithTimeout(context.Background(), 3*time.Second)
 	defer cancel()
 	ttl := d.negTTL