Files
probo/pkg/agent/restore.go
Aurélien Sibiril d13c83c19f Snapshot MaxTurns in checkpoint and apply it on restore
MaxTurns is the only agent bound compared against a counter that is
serialised in the checkpoint (Turns). When config drifts between save
and restore -- typically because a deploy changed WithMaxTurns or a
different build of the agent is registered by name -- cp.Turns can
exceed agent.maxTurns on the resumed run, which previously surfaced
as a warning log and then a MaxTurnsExceededError on the first
iteration of the resumed coreLoop.

Capture MaxTurns in the new AgentConfig on every save, and on
restore clone the registry-resolved agent with WithMaxTurns applied
from the snapshot. The override flows through the outer Restore path
and through both inner-agent resolution sites in
restoreNestedSuspended and restoreAwaitingApproval, so nested
runs get the same treatment. Other loop bounds
(maxEmptyOutputRetries, maxToolDepth) reset per turn / per tool
depth and stay intentionally live so deploys can tune them without
invalidating in-flight checkpoints. Live references (tools, hooks,
LLM client, approval callbacks, guardrails) are not snapshotted for
the same reason.

With the snapshot in place, the "restored agent run has already
reached max turns" warning at the top of continueFromMessages is
structurally unreachable -- the live agent's bound is now the same
value cp.Turns was bounded by at save time -- and is removed.

Signed-off-by: Aurélien Sibiril <81782+aureliensibiril@users.noreply.github.com>
2026-04-26 22:58:21 +02:00

352 lines
9.5 KiB
Go

// Copyright (c) 2026 Probo Inc <hello@getprobo.com>.
//
// Permission to use, copy, modify, and/or distribute this software for any
// purpose with or without fee is hereby granted, provided that the above
// copyright notice and this permission notice appear in all copies.
//
// THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH
// REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
// AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT,
// INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
// LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR
// OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
// PERFORMANCE OF THIS SOFTWARE.
package agent
import (
"context"
"errors"
"fmt"
"sync"
"go.probo.inc/probo/pkg/llm"
)
// Restore continues a previously suspended or approval-interrupted agent run
// from its last persisted checkpoint. The registry must contain all agents
// that may have been active (including handoff targets).
func Restore(
ctx context.Context,
store Checkpointer,
runID string,
registry AgentRegistry,
) (*Result, error) {
cp, err := store.Load(ctx, runID)
if err != nil {
return nil, fmt.Errorf("cannot load checkpoint: %w", err)
}
if cp == nil {
return nil, fmt.Errorf("cannot restore: no checkpoint for run %s", runID)
}
agent, err := registry.Agent(cp.AgentName)
if err != nil {
return nil, fmt.Errorf("cannot resolve agent %q: %w", cp.AgentName, err)
}
agent = applyCheckpointConfig(agent, cp.Config)
return restoreCheckpoint(ctx, agent, cp, store, runID, registry)
}
// applyCheckpointConfig returns a clone of agent with the bounds from
// the checkpoint snapshot overriding the live values. Zero values in
// cfg fall through to the live agent so older checkpoints written
// before the Config field existed, or test-constructed Checkpoint
// literals that omit Config, still resume correctly.
func applyCheckpointConfig(agent *Agent, cfg AgentConfig) *Agent {
if cfg.MaxTurns <= 0 {
return agent
}
return agent.Clone(WithMaxTurns(cfg.MaxTurns))
}
func restoreCheckpoint(
ctx context.Context,
agent *Agent,
cp *Checkpoint,
store Checkpointer,
runID string,
registry AgentRegistry,
) (*Result, error) {
emitHook(agent, func(h RunHooks) { h.OnRunRestore(ctx, agent, cp) })
switch cp.Status {
case AgentStatusSuspended:
return restoreSuspended(ctx, agent, cp, store, runID, registry)
case AgentStatusAwaitingApproval:
return restoreAwaitingApproval(ctx, agent, cp, store, runID, registry)
default:
return nil, fmt.Errorf("cannot restore: unknown checkpoint status %q", cp.Status)
}
}
func restoreSuspended(
ctx context.Context,
agent *Agent,
cp *Checkpoint,
store Checkpointer,
runID string,
registry AgentRegistry,
) (*Result, error) {
if len(cp.InnerCheckpoints) > 0 {
return restoreNestedSuspended(ctx, agent, cp, store, runID, registry)
}
return continueFromMessages(ctx, agent, cp.Messages, cp, store, runID)
}
func continueFromMessages(
ctx context.Context,
agent *Agent,
messages []llm.Message,
cp *Checkpoint,
store Checkpointer,
runID string,
) (*Result, error) {
messagesCopy := make([]llm.Message, len(messages))
copy(messagesCopy, messages)
return coreLoop(
ctx,
agent,
messagesCopy,
runOpts{
callLLM: blockingCallLLM,
onEvent: noopEvent,
skipInputGuardrails: true,
skipSessionLoad: true,
initialUsage: cp.Usage,
initialTurns: cp.Turns,
checkpointer: store,
runID: runID,
toolUsedInRun: cp.ToolUsedInRun,
},
)
}
func restoreNestedSuspended(
ctx context.Context,
agent *Agent,
cp *Checkpoint,
store Checkpointer,
runID string,
registry AgentRegistry,
) (*Result, error) {
type nestedRestoreEntry struct {
toolCall llm.ToolCall
originalCheckpoint *Checkpoint
suspendedCheckpoint *Checkpoint
result ToolResult
completed bool
err error
}
completedByID := make(map[string]ToolResult, len(cp.CompletedCalls))
for _, cc := range cp.CompletedCalls {
completedByID[cc.ToolCallID] = cc.Result
}
entries := make([]nestedRestoreEntry, len(cp.AllToolCalls))
var wg sync.WaitGroup
for i, tc := range cp.AllToolCalls {
entries[i].toolCall = tc
result, ok := completedByID[tc.ID]
if ok {
entries[i].result = result
entries[i].completed = true
continue
}
innerCP, ok := cp.InnerCheckpoints[tc.ID]
if !ok {
entries[i].err = fmt.Errorf("cannot restore nested tool call %q: missing inner checkpoint", tc.ID)
continue
}
entries[i].originalCheckpoint = innerCP
innerAgent, err := registry.Agent(innerCP.AgentName)
if err != nil {
entries[i].err = fmt.Errorf("cannot resolve inner agent %q: %w", innerCP.AgentName, err)
continue
}
innerAgent = applyCheckpointConfig(innerAgent, innerCP.Config)
wg.Add(1)
go func(i int, tc llm.ToolCall, innerAgent *Agent, innerCP *Checkpoint) {
defer wg.Done()
result, err := restoreCheckpoint(ctx, innerAgent, innerCP, nil, "", registry)
if err != nil {
if se, ok := errors.AsType[*SuspendedError](err); ok {
if se.Checkpoint == nil {
entries[i].err = fmt.Errorf("cannot restore nested tool call %q: missing suspension checkpoint", tc.ID)
return
}
entries[i].suspendedCheckpoint = se.Checkpoint
return
}
entries[i].err = fmt.Errorf("cannot restore nested tool call %q: %w", tc.ID, err)
return
}
entries[i].result = ToolResult{Content: result.FinalMessage().Text()}
entries[i].completed = true
}(i, tc, innerAgent, innerCP)
}
wg.Wait()
messages := make([]llm.Message, len(cp.Messages))
copy(messages, cp.Messages)
completedCalls := make([]CompletedCall, 0, len(cp.AllToolCalls))
remainingInner := make(map[string]*Checkpoint)
var restoreErr error
for _, entry := range entries {
switch {
case entry.err != nil:
if entry.originalCheckpoint != nil {
remainingInner[entry.toolCall.ID] = entry.originalCheckpoint
}
if restoreErr == nil {
restoreErr = entry.err
}
continue
case entry.suspendedCheckpoint != nil:
remainingInner[entry.toolCall.ID] = entry.suspendedCheckpoint
continue
case !entry.completed:
if restoreErr == nil {
restoreErr = fmt.Errorf("cannot restore nested tool call %q: no result", entry.toolCall.ID)
}
continue
}
completedCalls = append(
completedCalls,
CompletedCall{
ToolCallID: entry.toolCall.ID,
Result: entry.result,
},
)
messages = append(
messages,
llm.Message{
Role: llm.RoleTool,
ToolCallID: entry.toolCall.ID,
Parts: []llm.Part{llm.TextPart{Text: entry.result.Content}},
},
)
}
saveProgress := func() (*Checkpoint, error) {
next := *cp
next.InnerCheckpoints = remainingInner
next.CompletedCalls = completedCalls
if store != nil && runID != "" {
if err := store.Save(ctx, runID, &next); err != nil {
return nil, fmt.Errorf("cannot save nested restore progress: %w", err)
}
}
return &next, nil
}
if restoreErr != nil {
if _, err := saveProgress(); err != nil {
return nil, errors.Join(restoreErr, err)
}
return nil, restoreErr
}
if len(remainingInner) > 0 {
next, err := saveProgress()
if err != nil {
return nil, err
}
return nil, &SuspendedError{RunID: runID, Checkpoint: next}
}
return continueFromMessages(ctx, agent, messages, cp, store, runID)
}
func restoreAwaitingApproval(
ctx context.Context,
agent *Agent,
cp *Checkpoint,
store Checkpointer,
runID string,
registry AgentRegistry,
) (*Result, error) {
// Reconstruct an InterruptedError from the checkpoint.
ie := &InterruptedError{
ToolCalls: cp.PendingToolCalls,
PendingApprovals: cp.PendingApprovals,
Agent: agent,
Messages: cp.Messages,
Usage: cp.Usage,
Turns: cp.Turns,
}
// Reconstruct outerState if this was a nested interruption.
if len(cp.InnerCheckpoints) > 0 {
if len(cp.InnerCheckpoints) > 1 {
return nil, fmt.Errorf("cannot restore approval checkpoint: expected one inner checkpoint, got %d", len(cp.InnerCheckpoints))
}
for toolCallID, innerCP := range cp.InnerCheckpoints {
innerAgent, err := registry.Agent(innerCP.AgentName)
if err != nil {
return nil, fmt.Errorf("cannot resolve inner agent %q: %w", innerCP.AgentName, err)
}
innerAgent = applyCheckpointConfig(innerAgent, innerCP.Config)
innerIE := &InterruptedError{
ToolCalls: innerCP.PendingToolCalls,
PendingApprovals: innerCP.PendingApprovals,
Agent: innerAgent,
Messages: innerCP.Messages,
Usage: innerCP.Usage,
Turns: innerCP.Turns,
}
ie.Agent = innerAgent
ie.Messages = innerCP.Messages
ie.Usage = innerCP.Usage
ie.Turns = innerCP.Turns
ie.ToolCalls = innerCP.PendingToolCalls
ie.PendingApprovals = innerCP.PendingApprovals
ie.outerState = &outerLoopState{
agent: agent,
messages: cp.Messages,
usage: cp.Usage,
turns: cp.Turns,
allToolCalls: cp.AllToolCalls,
toolCallID: toolCallID,
completedCalls: cp.CompletedCalls,
innerInterrupt: innerIE,
}
break
}
}
if len(cp.ApprovalInput) > 0 {
return resumeWithOpts(
ctx,
ie,
ResumeInput{Approvals: cp.ApprovalInput},
runOpts{
callLLM: blockingCallLLM,
onEvent: noopEvent,
checkpointer: store,
runID: runID,
toolUsedInRun: cp.ToolUsedInRun,
},
)
}
return nil, ie
}