agent/pkg/runner/queue.go
Virgil e825550a90 fix(ax): continue AX naming cleanup
Co-Authored-By: Virgil <virgil@lethean.io>
2026-03-30 21:30:49 +00:00

351 lines
9.4 KiB
Go

// SPDX-License-Identifier: EUPL-1.2
package runner
import (
"strconv"
"time"
"dappco.re/go/agent/pkg/agentic"
core "dappco.re/go/core"
"gopkg.in/yaml.v3"
)
// DispatchConfig mirrors the `dispatch:` block in `agents.yaml`.
//
// config := runner.DispatchConfig{
// DefaultAgent: "codex", DefaultTemplate: "coding", WorkspaceRoot: "/srv/core/workspace",
// }
type DispatchConfig struct {
DefaultAgent string `yaml:"default_agent"`
DefaultTemplate string `yaml:"default_template"`
WorkspaceRoot string `yaml:"workspace_root"`
}
// RateConfig mirrors one agent pool under `rates:` in `agents.yaml`.
//
// rate := runner.RateConfig{
// ResetUTC: "06:00", DailyLimit: 200, SustainedDelay: 120, BurstWindow: 2, BurstDelay: 300,
// }
type RateConfig struct {
ResetUTC string `yaml:"reset_utc"`
DailyLimit int `yaml:"daily_limit"`
MinDelay int `yaml:"min_delay"`
SustainedDelay int `yaml:"sustained_delay"`
BurstWindow int `yaml:"burst_window"`
BurstDelay int `yaml:"burst_delay"`
}
// ConcurrencyLimit supports both flat (int) and nested (map with total + per-model) formats.
//
// claude: 1 → Total=1, Models=nil
// codex: → Total=5, Models={"gpt-5.4": 1}
// total: 5
// gpt-5.4: 1
type ConcurrencyLimit struct {
Total int
Models map[string]int
}
// UnmarshalYAML handles both int and map forms for concurrency limits.
//
// var limit ConcurrencyLimit
// _ = yaml.Unmarshal([]byte("total: 5\ngpt-5.4: 1\n"), &limit)
func (c *ConcurrencyLimit) UnmarshalYAML(value *yaml.Node) error {
var n int
if err := value.Decode(&n); err == nil {
c.Total = n
return nil
}
var m map[string]int
if err := value.Decode(&m); err != nil {
return err
}
c.Total = m["total"]
c.Models = make(map[string]int)
for k, v := range m {
if k != "total" {
c.Models[k] = v
}
}
return nil
}
// AgentsConfig mirrors the full `agents.yaml` file.
//
// config := runner.AgentsConfig{
// Version: 1,
// Dispatch: runner.DispatchConfig{DefaultAgent: "codex", DefaultTemplate: "coding"},
// }
type AgentsConfig struct {
Version int `yaml:"version"`
Dispatch DispatchConfig `yaml:"dispatch"`
Concurrency map[string]ConcurrencyLimit `yaml:"concurrency"`
Rates map[string]RateConfig `yaml:"rates"`
}
// loadAgentsConfig reads `agents.yaml` from the Core root.
//
// config := s.loadAgentsConfig()
// core.Println(config.Dispatch.DefaultAgent)
func (s *Service) loadAgentsConfig() *AgentsConfig {
paths := []string{
core.JoinPath(CoreRoot(), "agents.yaml"),
}
for _, path := range paths {
readResult := fs.Read(path)
if !readResult.OK {
continue
}
var config AgentsConfig
if err := yaml.Unmarshal([]byte(readResult.Value.(string)), &config); err != nil {
continue
}
return &config
}
return &AgentsConfig{
Dispatch: DispatchConfig{
DefaultAgent: "claude",
DefaultTemplate: "coding",
},
Concurrency: map[string]ConcurrencyLimit{
"claude": {Total: 1},
"gemini": {Total: 3},
},
}
}
// canDispatchAgent checks both pool-level and per-model concurrency limits.
//
// if !s.canDispatchAgent("codex") { /* queue it */ }
func (s *Service) canDispatchAgent(agent string) (bool, string) {
var concurrency map[string]ConcurrencyLimit
if s.ServiceRuntime != nil {
configurationResult := s.Core().Config().Get("agents.concurrency")
if configurationResult.OK {
concurrency, _ = configurationResult.Value.(map[string]ConcurrencyLimit)
}
}
if concurrency == nil {
config := s.loadAgentsConfig()
concurrency = config.Concurrency
}
base := baseAgent(agent)
limit, ok := concurrency[base]
if !ok || limit.Total <= 0 {
return true, ""
}
running := s.countRunningByAgent(base)
if running >= limit.Total {
return false, core.Sprintf("total %d/%d", running, limit.Total)
}
if limit.Models != nil {
model := modelVariant(agent)
if model != "" {
modelRunning := s.countRunningByModel(agent)
if modelLimit, has := limit.Models[model]; has && modelLimit > 0 {
if modelRunning >= modelLimit {
return false, core.Sprintf("model %s %d/%d", model, modelRunning, modelLimit)
}
}
}
}
return true, ""
}
// countRunningByAgent counts running workspaces using the in-memory Registry.
//
// n := s.countRunningByAgent("codex")
func (s *Service) countRunningByAgent(agent string) int {
var runtime *core.Core
if s.ServiceRuntime != nil {
runtime = s.Core()
}
count := 0
s.workspaces.Each(func(_ string, workspaceStatus *WorkspaceStatus) {
if workspaceStatus.Status != "running" || baseAgent(workspaceStatus.Agent) != agent {
return
}
switch {
case workspaceStatus.PID < 0:
count++
case workspaceStatus.PID > 0 && agentic.ProcessAlive(runtime, "", workspaceStatus.PID):
count++
}
})
return count
}
// countRunningByModel counts running workspaces for a specific `agent:model`.
//
// n := s.countRunningByModel("codex:gpt-5.4")
func (s *Service) countRunningByModel(agent string) int {
var runtime *core.Core
if s.ServiceRuntime != nil {
runtime = s.Core()
}
count := 0
s.workspaces.Each(func(_ string, workspaceStatus *WorkspaceStatus) {
if workspaceStatus.Status != "running" || workspaceStatus.Agent != agent {
return
}
switch {
case workspaceStatus.PID < 0:
count++
case workspaceStatus.PID > 0 && agentic.ProcessAlive(runtime, "", workspaceStatus.PID):
count++
}
})
return count
}
// drainQueue fills any free concurrency slots from queued workspaces.
//
// s.drainQueue()
func (s *Service) drainQueue() {
if s.frozen {
return
}
s.drainMu.Lock()
defer s.drainMu.Unlock()
for s.drainOne() {
// keep filling slots
}
}
func (s *Service) drainOne() bool {
for _, statusPath := range agentic.WorkspaceStatusPaths() {
workspaceDir := core.PathDir(statusPath)
statusResult := ReadStatusResult(workspaceDir)
if !statusResult.OK {
continue
}
workspaceStatus, ok := statusResult.Value.(*WorkspaceStatus)
if !ok || workspaceStatus == nil || workspaceStatus.Status != "queued" {
continue
}
if can, _ := s.canDispatchAgent(workspaceStatus.Agent); !can {
continue
}
pool := baseAgent(workspaceStatus.Agent)
if until, ok := s.backoff[pool]; ok && time.Now().Before(until) {
continue
}
delay := s.delayForAgent(workspaceStatus.Agent)
if delay > 0 {
time.Sleep(delay)
}
if can, _ := s.canDispatchAgent(workspaceStatus.Agent); !can {
continue
}
// Ask agentic to spawn — runner owns the gate,
// agentic owns the actual process launch.
// Workspace name is relative path from workspace root (e.g. "core/go-ai/dev")
workspaceName := agentic.WorkspaceName(workspaceDir)
core.Info("drainOne: found queued workspace", "workspace", workspaceName, "agent", workspaceStatus.Agent)
// Spawn directly — agentic is a Core service, use ServiceFor to get it
if s.ServiceRuntime == nil {
continue
}
type spawner interface {
SpawnFromQueue(agent, prompt, workspaceDir string) core.Result
}
agenticService, ok := core.ServiceFor[spawner](s.Core(), "agentic")
if !ok {
core.Error("drainOne: agentic service not found")
continue
}
prompt := core.Concat("TASK: ", workspaceStatus.Task, "\n\nResume from where you left off. Read CODEX.md for conventions. Commit when done.")
spawnResult := agenticService.SpawnFromQueue(workspaceStatus.Agent, prompt, workspaceDir)
if !spawnResult.OK {
core.Error("drainOne: spawn failed", "workspace", workspaceName, "reason", core.Sprint(spawnResult.Value))
continue
}
pid, ok := spawnResult.Value.(int)
if !ok {
core.Error("drainOne: spawn returned non-int pid", "workspace", workspaceName)
continue
}
// Only mark running AFTER successful spawn
workspaceStatus.Status = "running"
workspaceStatus.PID = pid
workspaceStatus.Runs++
if writeResult := WriteStatus(workspaceDir, workspaceStatus); !writeResult.OK {
core.Error("drainOne: failed to write workspace status", "workspace", workspaceName, "reason", core.Sprint(writeResult.Value))
continue
}
s.TrackWorkspace(workspaceName, workspaceStatus)
core.Info("drainOne: spawned", "pid", pid, "workspace", workspaceName)
return true
}
return false
}
func (s *Service) delayForAgent(agent string) time.Duration {
var rates map[string]RateConfig
if s.ServiceRuntime != nil {
rates, _ = s.Core().Config().Get("agents.rates").Value.(map[string]RateConfig)
}
if rates == nil {
config := s.loadAgentsConfig()
rates = config.Rates
}
base := baseAgent(agent)
rate, ok := rates[base]
if !ok || rate.SustainedDelay == 0 {
return 0
}
resetHour, resetMin := 6, 0
parts := core.Split(rate.ResetUTC, ":")
if len(parts) >= 2 {
if hour, err := strconv.Atoi(core.Trim(parts[0])); err == nil {
resetHour = hour
}
if min, err := strconv.Atoi(core.Trim(parts[1])); err == nil {
resetMin = min
}
}
now := time.Now().UTC()
resetToday := time.Date(now.Year(), now.Month(), now.Day(), resetHour, resetMin, 0, 0, time.UTC)
if now.Before(resetToday) {
resetToday = resetToday.AddDate(0, 0, -1)
}
nextReset := resetToday.AddDate(0, 0, 1)
hoursUntilReset := nextReset.Sub(now).Hours()
if rate.BurstWindow > 0 && hoursUntilReset <= float64(rate.BurstWindow) {
return time.Duration(rate.BurstDelay) * time.Second
}
return time.Duration(rate.SustainedDelay) * time.Second
}
// --- Helpers ---
func baseAgent(agent string) string {
return core.SplitN(agent, ":", 2)[0]
}
func modelVariant(agent string) string {
parts := core.SplitN(agent, ":", 2)
if len(parts) < 2 {
return ""
}
return parts[1]
}