Network Robustness: Allow final job update to come in "late" (after job was completed), in case server dropped offline then came back later.

Author: jhuckaby

htdocs/js/pages/Job.class.js

@@ -3263,6 +3263,7 @@ Page.Job = class Job extends Page.PageUtils {
 
 			case 'job_completed':
 				Debug.trace('job', "Job has completed, refreshing page");
+				app.cacheBust = hires_time_now();
 				Nav.refresh();
 			break;
 

lib/api/file.js

@@ -101,13 +101,14 @@ class FileManagement {
 				if (!self.requireValidUser(session, user, callback)) return;
 			}
 
-			var job = self.activeJobs[params.id];
-			if (!job) return self.doError('job', "Job not found: " + params.id, callback);
+			// JH 2026-02-28 Skipping this check, as files may come in late, after job has died
+			// var job = self.activeJobs[params.id];
+			// if (!job) return self.doError('job', "Job not found: " + params.id, callback);
 
 			var file = args.files.file1;
 			if (!file) return self.doError('job', "File upload data not found", callback);
 
-			var storage_key_prefix = 'files/jobs/' + job.id + '/' + Tools.generateUniqueBase64(32);
+			var storage_key_prefix = 'files/jobs/' + params.id + '/' + Tools.generateUniqueBase64(32);
 			var temp_file = file.path;
 			var filename = Path.basename(file.name).replace(/[^\w\-\.]+/g, '_').toLowerCase();
 			var storage_key = storage_key_prefix + '/' + filename;

lib/job.js

@@ -1016,6 +1016,44 @@ class Jobs {
 		return slims;
 	}
 
+	updateJobLate(updates) {
+		// perform late update on job that may have already completed
+		var self = this;
+		
+		this.logJob(5, "Attempting late update on completed job: " + updates.id, this.debugLevel(9) ? updates : false );
+		
+		this.unbase.update( 'jobs', updates.id, function(job) {
+			// perform updates here or bail out (inside unbase lock block)
+			if (job.code == 'abort') {
+				updates.code = 'warning';
+				updates.description = 'Job was aborted, but received late update after completion.';
+			}
+			
+			// append to job meta log
+			job.activity.push({ 
+				id: Tools.generateShortID('m'),
+				epoch: Tools.timeNow(), 
+				server: 'm:' + self.hostID,
+				msg: "WARNING: Received final job update post-completion."
+			});
+			updates.activity = job.activity;
+			
+			return updates;
+		}, 
+		function(err, job) {
+			// done with update (and unlocked)
+			if (err && (err === "ABORT")) return; // update was aborted and callback was handled
+			
+			if (err) {
+				self.logError('job', "Failed to update job late: " + updates.id + ": " + err);
+				return;
+			}
+			
+			// refresh anyone sitting on job details page
+			self.doPageBroadcast( 'Job?id=' + updates.id, 'job_completed', {} );
+		} ); // unbase.update
+	}
+	
 	updateJobData(socket, data) {
 		// receive job update from satellite
 		// may contain multiple jobs, and apply updates as shallow merges to each
@@ -1035,8 +1073,9 @@ class Jobs {
 			var job = self.activeJobs[job_id];
 			var info = self.jobDetails[job_id];
 			if (!job || !info) {
-				// should never happen, sanity check
-				self.logJob(9, "Job not found, updates not applied: " + job_id, updates);
+				// can theoretically happen if a job timed out and the server came back
+				if (updates.state == 'complete') self.updateJobLate(updates);
+				else self.logJob(9, "Job not found, updates not applied: " + job_id, updates);
 				return;
 			}
 			if (job.state == 'complete') {