Read only the event log tail for detached swarm agents

2 files · +40 -1

M packages/agent/swarm/event.go +33 -0


@@ -154,6 +154,39 @@ func ReadEventLog(path string) ([]Event, error) {
154	154		return readEvents(f)
155	155		}
156	156
	157	+	// ReadEventLogTail parses only the last maxBytes of the log. Startup
	158	+	// replay of detached agents keeps a bounded transcript anyway, so
	159	+	// reading multi-gigabyte logs in full only delays the first frame.
	160	+	// The first, possibly partial, line of the window is skipped.
	161	+	func ReadEventLogTail(path string, maxBytes int64) ([]Event, error) {
	162	+	f, err := os.Open(path)
	163	+	if err != nil {
	164	+	if os.IsNotExist(err) {
	165	+	return nil, nil
	166	+	}
	167	+	return nil, err
	168	+	}
	169	+	defer f.Close()
	170	+	fi, err := f.Stat()
	171	+	if err != nil {
	172	+	return nil, err
	173	+	}
	174	+	if maxBytes <= 0 \|\| fi.Size() <= maxBytes {
	175	+	return readEvents(f)
	176	+	}
	177	+	if _, err := f.Seek(fi.Size()-maxBytes, io.SeekStart); err != nil {
	178	+	return nil, err
	179	+	}
	180	+	br := bufio.NewReader(f)
	181	+	if _, err := br.ReadBytes('\n'); err != nil {
	182	+	if errors.Is(err, io.EOF) {
	183	+	return nil, nil
	184	+	}
	185	+	return nil, err
	186	+	}
	187	+	return readEvents(br)
	188	+	}
	189	+
157	190		func readEvents(r io.Reader) ([]Event, error) {
158	191		br := bufio.NewReader(r)
159	192		var out []Event


@@ -154,6 +154,39 @@ func ReadEventLog(path string) ([]Event, error) {
154	return readEvents(f)	154		return readEvents(f)
155	}	155		}
156		156
		157	+	// ReadEventLogTail parses only the last maxBytes of the log. Startup
		158	+	// replay of detached agents keeps a bounded transcript anyway, so
		159	+	// reading multi-gigabyte logs in full only delays the first frame.
		160	+	// The first, possibly partial, line of the window is skipped.
		161	+	func ReadEventLogTail(path string, maxBytes int64) ([]Event, error) {
		162	+	f, err := os.Open(path)
		163	+	if err != nil {
		164	+	if os.IsNotExist(err) {
		165	+	return nil, nil
		166	+	}
		167	+	return nil, err
		168	+	}
		169	+	defer f.Close()
		170	+	fi, err := f.Stat()
		171	+	if err != nil {
		172	+	return nil, err
		173	+	}
		174	+	if maxBytes <= 0 \|\| fi.Size() <= maxBytes {
		175	+	return readEvents(f)
		176	+	}
		177	+	if _, err := f.Seek(fi.Size()-maxBytes, io.SeekStart); err != nil {
		178	+	return nil, err
		179	+	}
		180	+	br := bufio.NewReader(f)
		181	+	if _, err := br.ReadBytes('\n'); err != nil {
		182	+	if errors.Is(err, io.EOF) {
		183	+	return nil, nil
		184	+	}
		185	+	return nil, err
		186	+	}
		187	+	return readEvents(br)
		188	+	}
		189	+
157	func readEvents(r io.Reader) ([]Event, error) {	190		func readEvents(r io.Reader) ([]Event, error) {
158	br := bufio.NewReader(r)	191		br := bufio.NewReader(r)
159	var out []Event	192		var out []Event

M packages/agent/swarm/persist.go +7 -1


@@ -180,6 +180,12 @@ func (f *Swarm) Reload() (loaded int, errs []error) {
180	180		return loaded, errs
181	181		}
182	182
	183	+	// detachedReplayBytes bounds how much of each event log is replayed for
	184	+	// a detached agent. The transcript keeps at most 2000 lines and the
	185	+	// terminal lifecycle event sits at the end of the log, so the tail is
	186	+	// enough; full logs are still read on demand by the dashboard.
	187	+	const detachedReplayBytes = 4 << 20
	188	+
183	189		// buildDetachedAgent constructs an Agent from a meta.json with no
184	190		// running Runner. The agent's transcript is populated from the tail
185	191		// of its event log so the dashboard immediately shows recent output;

@@ -223,7 +229,7 @@ func (f Swarm) buildDetachedAgent(m agentMeta) Agent {
223	229		// effort: a missing or unreadable log just leaves the agent
224	230		// detached with an empty transcript.
225	231		if a.EventLogPath != "" {
226		-	if evs, err := ReadEventLog(a.EventLogPath); err == nil {
	232	+	if evs, err := ReadEventLogTail(a.EventLogPath, detachedReplayBytes); err == nil {
227	233		replayEventsIntoAgent(a, evs)
228	234		}
229	235		}

 	return loaded, errs
 }
+// detachedReplayBytes bounds how much of each event log is replayed for
+// a detached agent. The transcript keeps at most 2000 lines and the
+// terminal lifecycle event sits at the end of the log, so the tail is
+// enough; full logs are still read on demand by the dashboard.
+const detachedReplayBytes = 4 << 20
 // buildDetachedAgent constructs an Agent from a meta.json with no
 // running Runner. The agent's transcript is populated from the tail
 // of its event log so the dashboard immediately shows recent output;
 	// effort: a missing or unreadable log just leaves the agent
 	// detached with an empty transcript.
 	if a.EventLogPath != "" {
-		if evs, err := ReadEventLog(a.EventLogPath); err == nil {
+		if evs, err := ReadEventLogTail(a.EventLogPath, detachedReplayBytes); err == nil {
 			replayEventsIntoAgent(a, evs)
 		}
 	}