From 4f010c912f014d4db22cc5b1e4df290fd32f29c8 Mon Sep 17 00:00:00 2001
From: Marco Vogt <marco.vogt@unibas.ch>
Date: Mon, 20 Jun 2022 13:26:19 +0200
Subject: [PATCH] Do not cancel monitoring if unable to fetch status from
 chronos control

---
 .../dmi/dbis/chronos/agent/AbstractChronosAgent.java  | 11 +++++++++--
 .../dmi/dbis/chronos/agent/ChronosHttpClient.java     |  2 --
 2 files changed, 9 insertions(+), 4 deletions(-)

diff --git a/src/main/java/ch/unibas/dmi/dbis/chronos/agent/AbstractChronosAgent.java b/src/main/java/ch/unibas/dmi/dbis/chronos/agent/AbstractChronosAgent.java
index e7a83b0..7f6c244 100644
--- a/src/main/java/ch/unibas/dmi/dbis/chronos/agent/AbstractChronosAgent.java
+++ b/src/main/java/ch/unibas/dmi/dbis/chronos/agent/AbstractChronosAgent.java
@@ -45,6 +45,7 @@ of this software and associated documentation files (the "Software"), to deal
 import java.util.TimerTask;
 import java.util.concurrent.ConcurrentHashMap;
 import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicInteger;
 import lombok.extern.slf4j.Slf4j;
 import net.lingala.zip4j.ZipFile;
 import net.lingala.zip4j.model.ZipParameters;
@@ -678,6 +679,8 @@ private class AbortedMonitorTask extends TimerTask {
 
             private final ChronosJob observable;
 
+            private final AtomicInteger getStatusErrorCounter = new AtomicInteger( 0 );
+
 
             /**
              *
@@ -710,9 +713,13 @@ public void run() {
                         // Quietly cancelObservation and de-register this observable since it is already finished
                         cancelAndRemoveObservable();
                     }
+                    getStatusErrorCounter.set( 0 );
                 } catch ( NoSuchElementException | ChronosException | IOException ex ) {
-                    log.warn( "getStatus for \"" + this.observable + "\" failed. Canceling monitoring!", ex );
-                    cancelAndRemoveObservable();
+                    getStatusErrorCounter.incrementAndGet();
+                    if ( getStatusErrorCounter.get() % 10 == 0 ) {
+                        log.warn( "Unable to get status of \"" + this.observable + "\" since quite a while.", ex );
+                    }
+                    //cancelAndRemoveObservable();
                 } catch ( InterruptedException ex ) {
                     log.warn( "We have been interrupted!", ex );
                     cancelAndRemoveObservable();
diff --git a/src/main/java/ch/unibas/dmi/dbis/chronos/agent/ChronosHttpClient.java b/src/main/java/ch/unibas/dmi/dbis/chronos/agent/ChronosHttpClient.java
index d8b2788..44c20a9 100644
--- a/src/main/java/ch/unibas/dmi/dbis/chronos/agent/ChronosHttpClient.java
+++ b/src/main/java/ch/unibas/dmi/dbis/chronos/agent/ChronosHttpClient.java
@@ -567,7 +567,6 @@ private boolean notifyChronos( final ChronosJob job, final Properties uploadConf
         Exception lastException;
 
         do {
-
             try {
                 // NOTICE: intentional return!
                 return this.doNotifyChronos( job, uploadConfiguration, parameters );
@@ -579,7 +578,6 @@ private boolean notifyChronos( final ChronosJob job, final Properties uploadConf
             if ( norifyChronosAttempt < maxAttempts ) {
                 Thread.sleep( failedAttemptSleepTimeMillis );
             }
-
         } while ( norifyChronosAttempt++ < maxAttempts );
 
         throw new IOException( "The maximum number of attempts reached. See above WARNINGs for details. The last exception was: ", lastException );