Files
skillshare/internal/server/handler_check_stream.go
T
Willie 5dd83a8412 fix(server): guard metadata store access against concurrent requests
Every API request reloads the skills and agents metadata stores and
swaps the server's pointers under s.mu, but several handlers read or
wrote the stores outside that lock: listing, skill detail, check, the
update stream, and the hash refresh after setting targets. Concurrent
requests could race on the pointer or on the stores' maps, which the
race detector reported hundreds of times under load and which can
crash the server with a concurrent map access.

Store access now always happens under s.mu. Reads return a copy of
the entry so it can be used after unlocking, and target handlers hash
files outside the lock and only write the result under it.
2026-09-26 21:12:57 +08:00

239 lines
5.8 KiB
Go

package server
import (
"net/http"
"path/filepath"
"strings"
"sync"
"sync/atomic"
"time"
"skillshare/internal/check"
"skillshare/internal/git"
"skillshare/internal/install"
)
// handleCheckStream serves an SSE endpoint that streams check progress in real time.
// Events:
// - "discovering" → {"phase":"..."} immediately on connect
// - "start" → {"total": N, "repos": R, "sources": S} after discovery (N = work units)
// - "progress" → {"checked": N} every 200ms
// - "done" → {"tracked_repos":…,"skills":…} final payload (same shape as GET /api/check)
//
// "total" counts actual work units (repos + remote URL groups), NOT individual skills.
// This ensures the progress bar advances evenly — one tick per network call.
func (s *Server) handleCheckStream(w http.ResponseWriter, r *http.Request) {
safeSend, ok := initSSE(w)
if !ok {
return
}
ctx := r.Context()
// Snapshot config under RLock, then release before slow I/O.
s.mu.RLock()
sourceDir := s.skillsSource()
s.mu.RUnlock()
// Immediate feedback before the potentially slow discovery walk.
safeSend("discovering", map[string]string{"phase": "scanning source directory"})
repos, _ := install.GetTrackedRepos(sourceDir)
skills, _ := install.GetUpdatableSkills(sourceDir)
// --- Pre-process: group skills by URL (fast, local only) ---
urlGroups := make(map[urlBranchGroup][]skillWithMetaEntry)
var localResults []skillCheckResult
for _, skill := range skills {
entry := s.skillEntry(skill)
if entry == nil || entry.RepoURL == "" {
localResults = append(localResults, skillCheckResult{
Name: skill,
Status: "local",
})
continue
}
key := urlBranchGroup{url: entry.RepoURL, branch: entry.Branch}
urlGroups[key] = append(urlGroups[key], skillWithMetaEntry{
name: skill,
entry: entry,
})
}
// Total = repos + URL groups (the actual network-bound work units).
total := len(repos) + len(urlGroups)
safeSend("start", map[string]any{
"total": total,
"repos": len(repos),
"sources": len(urlGroups),
})
// Atomic counter + ticker for progress events
var checked atomic.Int64
done := make(chan struct{})
var wg sync.WaitGroup
wg.Add(1)
go func() {
defer wg.Done()
ticker := time.NewTicker(200 * time.Millisecond)
defer ticker.Stop()
for {
select {
case <-ticker.C:
safeSend("progress", map[string]int64{"checked": checked.Load()})
case <-done:
return
case <-ctx.Done():
return
}
}
}()
// --- Phase 1: Check tracked repos (1 work unit per repo) ---
var repoResults []repoCheckResult
for _, repo := range repos {
select {
case <-ctx.Done():
close(done)
wg.Wait()
return
default:
}
repoPath := filepath.Join(sourceDir, repo)
result := repoCheckResult{Name: repo}
if isDirty, _ := git.IsDirty(repoPath); isDirty {
result.Status = "dirty"
result.Message = "has uncommitted changes"
} else if behind, err := git.GetBehindCount(repoPath); err != nil {
result.Status = "error"
result.Message = err.Error()
} else if behind == 0 {
result.Status = "up_to_date"
} else {
result.Status = "behind"
result.Behind = behind
}
repoResults = append(repoResults, result)
checked.Add(1)
}
// --- Phase 2: Check skills by URL group (1 work unit per URL) ---
skillResults := append([]skillCheckResult{}, localResults...)
for key, group := range urlGroups {
select {
case <-ctx.Done():
close(done)
wg.Wait()
return
default:
}
remoteHash, err := key.remoteHash()
if err != nil {
for _, sw := range group {
r := skillCheckResult{
Name: sw.name,
Source: sw.entry.Source,
Version: sw.entry.Version,
Status: "error",
}
if !sw.entry.InstalledAt.IsZero() {
r.InstalledAt = sw.entry.InstalledAt.Format("2006-01-02")
}
skillResults = append(skillResults, r)
}
checked.Add(1)
continue
}
// Fast path: all commit hashes match
allMatch := true
for _, sw := range group {
if sw.entry.Version != remoteHash {
allMatch = false
break
}
}
if allMatch {
for _, sw := range group {
r := skillCheckResult{
Name: sw.name,
Source: sw.entry.Source,
Version: sw.entry.Version,
Status: "up_to_date",
}
if !sw.entry.InstalledAt.IsZero() {
r.InstalledAt = sw.entry.InstalledAt.Format("2006-01-02")
}
skillResults = append(skillResults, r)
}
checked.Add(1)
continue
}
// Slow path: tree hash comparison
var hasTreeHash bool
for _, sw := range group {
if sw.entry.TreeHash != "" && sw.entry.Subdir != "" {
hasTreeHash = true
break
}
}
var remoteTreeHashes map[string]string
if hasTreeHash {
remoteTreeHashes = check.FetchRemoteTreeHashesForRef(key.url, key.branch)
}
for _, sw := range group {
r := skillCheckResult{
Name: sw.name,
Source: sw.entry.Source,
Version: sw.entry.Version,
}
if !sw.entry.InstalledAt.IsZero() {
r.InstalledAt = sw.entry.InstalledAt.Format("2006-01-02")
}
if sw.entry.Version == remoteHash {
r.Status = "up_to_date"
} else if sw.entry.TreeHash != "" && sw.entry.Subdir != "" && remoteTreeHashes != nil {
normalizedSubdir := strings.TrimPrefix(sw.entry.Subdir, "/")
if rh, ok := remoteTreeHashes[normalizedSubdir]; ok && sw.entry.TreeHash == rh {
r.Status = "up_to_date"
} else {
r.Status = "update_available"
}
} else {
r.Status = "update_available"
}
skillResults = append(skillResults, r)
}
checked.Add(1)
}
// Stop ticker
close(done)
wg.Wait()
if repoResults == nil {
repoResults = []repoCheckResult{}
}
if skillResults == nil {
skillResults = []skillCheckResult{}
}
safeSend("done", map[string]any{
"tracked_repos": repoResults,
"skills": skillResults,
})
}