health.go
| 1 | package handler |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "maps" |
| 7 | "net/http" |
| 8 | "slices" |
| 9 | "strings" |
| 10 | "sync" |
| 11 | "time" |
| 12 | |
| 13 | "vidarchive/internal/util" |
| 14 | ) |
| 15 | |
| 16 | // healthProbeTimeout caps the external version lookups so the endpoint always |
| 17 | // answers promptly, even when a tool is wedged. |
| 18 | const healthProbeTimeout = 3 * time.Second |
| 19 | |
| 20 | // toolCacheTTL bounds how stale the reported versions may be. /healthz is |
| 21 | // public and each lookup forks a process, so without a cache anyone who can |
| 22 | // reach the port costs the host three processes per request. |
| 23 | const toolCacheTTL = time.Minute |
| 24 | |
| 25 | // toolRetryTTL is the shorter TTL used when a probe timed out. A timeout says |
| 26 | // less about the tool than a real answer, so it is worth re-probing sooner. It |
| 27 | // is still cached, though: not caching it hands an unauthenticated caller three |
| 28 | // forks per request for as long as the tool stays wedged. |
| 29 | const toolRetryTTL = 10 * time.Second |
| 30 | |
| 31 | // toolCache holds the last tool version probe. The mutex is held across the |
| 32 | // probe itself, so a burst of requests collapses into one set of forks. ttl |
| 33 | // varies per result; see toolRetryTTL. |
| 34 | type toolCache struct { |
| 35 | mu sync.Mutex |
| 36 | at time.Time |
| 37 | ttl time.Duration |
| 38 | results map[string]string |
| 39 | } |
| 40 | |
| 41 | type healthResponse struct { |
| 42 | Status string `json:"status"` |
| 43 | Dependencies map[string]string `json:"dependencies"` |
| 44 | } |
| 45 | |
| 46 | // Health reports whether the app can reach its database and external tools. It |
| 47 | // returns 503 when the database is unreachable, so a container healthcheck or |
| 48 | // orchestrator can act on it. A missing yt-dlp/ffmpeg is reported but does not |
| 49 | // fail the check: the UI still works, only downloads are affected. |
| 50 | func (h *Handler) Health(w http.ResponseWriter, r *http.Request) { |
| 51 | // Bound the whole probe: a health check that can hang is worse than useless, |
| 52 | // since an orchestrator reads a stuck request as "still starting". |
| 53 | ctx, cancel := context.WithTimeout(r.Context(), healthProbeTimeout) |
| 54 | defer cancel() |
| 55 | |
| 56 | resp := healthResponse{Status: "ok", Dependencies: make(map[string]string, 4)} |
| 57 | |
| 58 | // The database is probed first because it alone decides the status code, and |
| 59 | // it must get the budget before a wedged tool can spend it — otherwise a slow |
| 60 | // yt-dlp reports the database as unreachable and the container is restarted |
| 61 | // for the wrong reason. |
| 62 | status := http.StatusOK |
| 63 | if err := h.downloadSvc.Ping(ctx); err != nil { |
| 64 | resp.Status = "unhealthy" |
| 65 | resp.Dependencies["database"] = "unreachable" |
| 66 | status = http.StatusServiceUnavailable |
| 67 | } else { |
| 68 | resp.Dependencies["database"] = "ok" |
| 69 | } |
| 70 | |
| 71 | for tool, version := range h.toolVersions(ctx) { |
| 72 | resp.Dependencies[tool] = version |
| 73 | } |
| 74 | |
| 75 | w.Header().Set("Content-Type", "application/json") |
| 76 | w.WriteHeader(status) |
| 77 | _ = json.NewEncoder(w).Encode(resp) |
| 78 | } |
| 79 | |
| 80 | // toolVersions returns the external tools' versions, re-probing them at most |
| 81 | // once per toolCacheTTL. |
| 82 | func (h *Handler) toolVersions(ctx context.Context) map[string]string { |
| 83 | h.tools.mu.Lock() |
| 84 | defer h.tools.mu.Unlock() |
| 85 | |
| 86 | if h.tools.results != nil && time.Since(h.tools.at) < h.tools.ttl { |
| 87 | return h.tools.results |
| 88 | } |
| 89 | |
| 90 | results := map[string]string{ |
| 91 | "yt-dlp": toolVersion(ctx, h.cfg.YTDLPPath, "--version"), |
| 92 | "ffmpeg": toolVersion(ctx, h.cfg.FFmpegPath, "-version"), |
| 93 | "ffprobe": toolVersion(ctx, h.cfg.FFprobePath, "-version"), |
| 94 | } |
| 95 | ttl := toolCacheTTL |
| 96 | if slices.Contains(slices.Collect(maps.Values(results)), "timed out") { |
| 97 | ttl = toolRetryTTL |
| 98 | } |
| 99 | h.tools.results, h.tools.at, h.tools.ttl = results, time.Now(), ttl |
| 100 | return results |
| 101 | } |
| 102 | |
| 103 | // toolVersion returns the first line of the tool's version output. It reports |
| 104 | // "not installed" when the binary is missing and "timed out" when it doesn't |
| 105 | // answer within ctx. |
| 106 | func toolVersion(ctx context.Context, path, versionArg string) string { |
| 107 | out, err := util.KillableCommand(ctx, path, versionArg).Output() |
| 108 | if err != nil { |
| 109 | if ctx.Err() != nil { |
| 110 | return "timed out" |
| 111 | } |
| 112 | return "not installed" |
| 113 | } |
| 114 | line, _, _ := strings.Cut(strings.TrimSpace(string(out)), "\n") |
| 115 | // ffmpeg prints "ffmpeg version N-x ..."; keep the version token only. |
| 116 | if fields := strings.Fields(line); len(fields) >= 3 && fields[1] == "version" { |
| 117 | return fields[2] |
| 118 | } |
| 119 | return line |
| 120 | } |
| 121 |