@@ -26,7 +26,12 @@ type WorkspaceAppHealthReporter func(ctx context.Context)
2626
2727// NewWorkspaceAppHealthReporter creates a WorkspaceAppHealthReporter that reports app health to coderd.
2828func NewWorkspaceAppHealthReporter (logger slog.Logger ,apps []codersdk.WorkspaceApp ,postWorkspaceAgentAppHealth PostWorkspaceAgentAppHealth )WorkspaceAppHealthReporter {
29+ logger = logger .Named ("apphealth" )
30+
2931runHealthcheckLoop := func (ctx context.Context )error {
32+ ctx ,cancel := context .WithCancel (ctx )
33+ defer cancel ()
34+
3035// no need to run this loop if no apps for this workspace.
3136if len (apps )== 0 {
3237return nil
@@ -87,6 +92,7 @@ func NewWorkspaceAppHealthReporter(logger slog.Logger, apps []codersdk.Workspace
8792return nil
8893}()
8994if err != nil {
95+ nowUnhealthy := false
9096mu .Lock ()
9197if failures [app .ID ]< int (app .Healthcheck .Threshold ) {
9298// increment the failure count and keep status the same.
@@ -96,14 +102,17 @@ func NewWorkspaceAppHealthReporter(logger slog.Logger, apps []codersdk.Workspace
96102// set to unhealthy if we hit the failure threshold.
97103// we stop incrementing at the threshold to prevent the failure value from increasing forever.
98104health [app .ID ]= codersdk .WorkspaceAppHealthUnhealthy
105+ nowUnhealthy = true
99106}
100107mu .Unlock ()
108+ logger .Debug (ctx ,"error checking app health" ,slog .F ("id" ,app .ID .String ()),slog .F ("slug" ,app .Slug ),slog .F ("now_unhealthy" ,nowUnhealthy ),slog .Error (err ))
101109}else {
102110mu .Lock ()
103111// we only need one successful health check to be considered healthy.
104112health [app .ID ]= codersdk .WorkspaceAppHealthHealthy
105113failures [app .ID ]= 0
106114mu .Unlock ()
115+ logger .Debug (ctx ,"workspace app healthy" ,slog .F ("id" ,app .ID .String ()),slog .F ("slug" ,app .Slug ))
107116}
108117
109118t .Reset (time .Duration (app .Healthcheck .Interval )* time .Second )
@@ -137,7 +146,9 @@ func NewWorkspaceAppHealthReporter(logger slog.Logger, apps []codersdk.Workspace
137146Healths :lastHealth ,
138147})
139148if err != nil {
140- logger .Error (ctx ,"failed to report workspace app stat" ,slog .Error (err ))
149+ logger .Error (ctx ,"failed to report workspace app health" ,slog .Error (err ))
150+ }else {
151+ logger .Debug (ctx ,"sent workspace app health" ,slog .F ("health" ,lastHealth ))
141152}
142153}
143154}