dcp-worker 4.2.2 → 4.2.4

package/bin/dcp-evaluator-manager

@@ -38,15 +38,16 @@ var seq = 0;
 var ptsDir = '/dev/pts';
 var lastConnectionMs = 0;
 
+const defaultSocketDisplayLimit = Number(process.env.DCP_EM_RAW_BYTES || '93');
 const defaultPrefix = '/usr';
 const daemonConfig = {
   net:  new URL('tcpip://localhost:9000/'),
   proc: process.argv[0],
-  argv: [ require.resolve('./dcp-evaluator-start'), '-s', '--prefix', `${path.resolve(defaultPrefix)}/` ],
-  pidfile: pidfile.getDefaultPidFilename,
+  argv: [ require.resolve('./dcp-evaluator-start'), '-s', `--prefix=${path.resolve(defaultPrefix)}/` ],
+  pidfile: pidfile.getDefaultPidFileName(),
   limits: {
     loadavgType: 'short',
-    randomLoadSamples: new Array(5).fill(0.5),
+    maxLoad: Math.max(2, os.cpus().length * 2),
   },
   session: {
     idlePoll:    3,  /* s between checks*/
@@ -78,24 +79,29 @@ Where:
  -p specifies the port to listen on for dcpsaw: worker connections (${daemonConfig.net.port})
  -H specifies the address to listen on for dcpsaw: worker connections (${daemonConfig.net.port})
  --prefix specifies the directory where the DCP evaluator package was installed
- --pidfile specifies an alternate location for the pid file (currently ${daemonConfig.pidfile})
+ --pidfile specifies an alternate location for the pid file;
+           currently ${daemonConfig.pidfile}
  --pts specifies the location of the system's pseudo terminal (${ptsDir})
  --disable-monitor=session disables monitoring of ssh/telnet/etc sessions
  --disable-monitor=screensaver disables monitoring of the screensaver
  --disable-monitor=signal disables monitoring of the signal
  --signal=name,mode changes which signal to monitor and its behaviour. Toggle
-   mode turns the monitor on/off, quick mode kills active evaluators but
-   more can spawn right away. Currently ${daemonConfig.signal.which},${daemonConfig.signal.mode}.
- -a means always run the evaluator, even when the monitors say the machine is busy
+   mode turns the monitor on/off, quick mode kills active evaluators but more
+   can spawn right away. Currently ${daemonConfig.signal.which},${daemonConfig.signal.mode}.
+ -a means always run the evaluator, even if the monitors say the machine is busy
  --rate specifies the minimum number of seconds between connections
- --max-load specifies the point at which machine load prevents connections
- --loadavg-type determines how load is measured; random generates a pseudo-random
-   load between 0 and 1; currently ${daemonConfig.limits.loadavgType}.
+ --max-load specifies the point at which machine load prevents connections.
+   A value of zero disables this limit; the default value is ${daemonConfig.limits.maxLoad}.
+ --loadavg-type determines how load is measured; 'short', 'medium', 'long',
+                'random'. Random generates a load between 0 and 1 (${daemonConfig.limits.loadavgType}).
+
  -L only apply the limits when the monitors say the machine is busy
 
 Environment:
-  DEBUG      - enables debugging, eg DEBUG='dcp-evaluator-*,-dcp-evaluator-manager:network'
-  DCP_CONFIG - sets location of dcp-worker-config file
+  DEBUG            - enables debugging, eg
+                     DEBUG='dcp-evaluator-*,-dcp-evaluator-manager:network'
+  DCP_CONFIG       - sets location of dcp-worker-config file
+  DCP_EM_RAW_BYTES - number of bytes to display in :network debugging
 `);
 
   process.exit(0);
@@ -194,15 +200,18 @@ function handleConnection (socket, config)
 
     switch(daemonConfig.limits.loadavgType)
     {
+      default:
+        console.warn(`Warning: invalid loadavg type ${daemonConfig.limits.loadavgType}`);
+        /* fallthrough */
       case 'short':
-        loadavg = os.loadavg[0];
+        loadavg = os.loadavg()[0];
         break;
       case 'medium':
       case 'med':
-        loadavg = os.loadavg[1];
+        loadavg = os.loadavg()[1];
         break;
       case 'long':
-        loadavg = os.loadavg[2];
+        loadavg = os.loadavg()[2];
         break;
       case 'random': /* random number in range [0,1) that usually changes slowly and averages 0.5 */
       {
@@ -214,7 +223,7 @@ function handleConnection (socket, config)
       }
     }
 
-    if (daemonConfig.limits.maxLoad < loadavg)
+    if (daemonConfig.limits.maxLoad && daemonConfig.limits.maxLoad < loadavg)
     {
       debug('dcp-evaluator-manager')(`New connection; closing due to ${daemonConfig.limits.loadavgType} system load ${loadavg} (${os.loadavg().join(', ')})`);
       cleanup();
@@ -267,7 +276,7 @@ function handleConnection (socket, config)
   });
 
   child.stdout.on('data', function (data) {
-    debug('dcp-evaluator-manager:network')('<', child.id, bufToDisplayStr(data), 93);
+    debug('dcp-evaluator-manager:network')('<', child.id, bufToDisplayStr(data));
     if (socket)
       socket.write(data);
   });
@@ -277,7 +286,7 @@ function handleConnection (socket, config)
   });
 
   socket.on('data', function (data) {
-    debug('dcp-evaluator-manager:network')('>', child.id, bufToDisplayStr(data, 93));
+    debug('dcp-evaluator-manager:network')('>', child.id, bufToDisplayStr(data));
 
     try
     {
@@ -297,6 +306,8 @@ function handleConnection (socket, config)
 function bufToDisplayStr (buf, limit)
 {
   const str = buf.toString('utf-8').replace(/\n/, '\u2424').toString('utf-8');
+  if (typeof limit === 'undefined')
+    limit = defaultSocketDisplayLimit;
 
   if (typeof limit === 'number' && str.length > limit)
     return str.substr(0, limit) + '\u2026';
@@ -589,6 +600,7 @@ async function main()
       case 'R':
       {
         daemonConfig.limits.randomLoad = true;
+        daemonConfig.limits.randomLoadSamples = new Array(5).fill(0.5);
         break;
       }
 

package/bin/dcp-worker

@@ -687,8 +687,15 @@ async function main()
       console.error('Unrecoverable', error.code ? `${error.name} (${error.code})` : error.name,
                     '- shutting down the worker:', error.message);
     }
+
+    /* Let the supervisor try to finish slices for up to three minutes after a single unrecoverable error. */
     worker.stop(Boolean(handleWorkerError.count));
     handleWorkerError.count = 1 + (handleWorkerError.count || 0);
+    if (handleWorkerError.count === 1)
+    {
+      setTimeout(() => worker && worker.stop(true), 180e3);
+      setTimeout(() => { throw new Error('Shutdown timeout after unrecoverable error') }, 185e3);
+    }
   });
 
   if (pidfileName && require('../lib/pidfile').write(pidfileName) !== true)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dcp-worker",
-  "version": "4.2.2",
+  "version": "4.2.4",
   "description": "Node.js Worker for Distributive Compute Platform",
   "main": "bin/dcp-worker",
   "keywords": [
@@ -39,7 +39,7 @@
     "blessed": "0.1.81",
     "blessed-contrib": "4.11.0",
     "chalk": "^4.1.0",
-    "dcp-client": "^5.1.3",
+    "dcp-client": "^5.1.5",
     "kvin": "^1.2.7",
     "posix-getopt": "^1.2.1",
     "semver": "^7.3.8",