Remove unfinished usage job entries of the host (apache#10848)

Author: sureshanaparti

engine/schema/src/main/java/com/cloud/usage/dao/UsageJobDao.java

@@ -37,4 +37,6 @@ public interface UsageJobDao extends GenericDao<UsageJobVO, Long> {
     UsageJobVO isOwner(String hostname, int pid);
 
     void updateJobSuccess(Long jobId, long startMillis, long endMillis, long execTime, boolean success);
+
+    void removeLastOpenJobsOwned(String hostname, int pid);
 }

engine/schema/src/main/java/com/cloud/usage/dao/UsageJobDaoImpl.java

@@ -22,6 +22,7 @@
 import java.util.List;
 
 
+import org.apache.commons.collections.CollectionUtils;
 import org.springframework.stereotype.Component;
 
 import com.cloud.usage.UsageJobVO;
@@ -114,7 +115,7 @@ public Long checkHeartbeat(String hostname, int pid, int aggregationDuration) {
     public UsageJobVO isOwner(String hostname, int pid) {
         TransactionLegacy txn = TransactionLegacy.open(TransactionLegacy.USAGE_DB);
         try {
-            if ((hostname == null) || (pid <= 0)) {
+            if (hostname == null || pid <= 0) {
                 return null;
             }
 
@@ -174,7 +175,7 @@ public UsageJobVO getNextImmediateJob() {
         SearchCriteria<UsageJobVO> sc = createSearchCriteria();
         sc.addAnd("endMillis", SearchCriteria.Op.EQ, Long.valueOf(0));
         sc.addAnd("jobType", SearchCriteria.Op.EQ, Integer.valueOf(UsageJobVO.JOB_TYPE_SINGLE));
-        sc.addAnd("scheduled", SearchCriteria.Op.EQ, Integer.valueOf(0));
+        sc.addAnd("scheduled", SearchCriteria.Op.EQ, Integer.valueOf(UsageJobVO.JOB_NOT_SCHEDULED));
         List<UsageJobVO> jobs = search(sc, filter);
 
         if ((jobs == null) || jobs.isEmpty()) {
@@ -194,4 +195,36 @@ public Date getLastHeartbeat() {
         }
         return jobs.get(0).getHeartbeat();
     }
+
+    private List<UsageJobVO> getLastOpenJobsOwned(String hostname, int pid) {
+        SearchCriteria<UsageJobVO> sc = createSearchCriteria();
+        sc.addAnd("endMillis", SearchCriteria.Op.EQ, Long.valueOf(0));
+        sc.addAnd("host", SearchCriteria.Op.EQ, hostname);
+        if (pid > 0) {
+            sc.addAnd("pid", SearchCriteria.Op.EQ, Integer.valueOf(pid));
+        }
+        return listBy(sc);
+    }
+
+    @Override
+    public void removeLastOpenJobsOwned(String hostname, int pid) {
+        if (hostname == null) {
+            return;
+        }
+
+        TransactionLegacy txn = TransactionLegacy.open(TransactionLegacy.USAGE_DB);
+        try {
+            List<UsageJobVO> jobs = getLastOpenJobsOwned(hostname, pid);
+            if (CollectionUtils.isNotEmpty(jobs)) {
+                s_logger.info(String.format("Found %s opens job, to remove", jobs.size()));
+                for (UsageJobVO job : jobs) {
+                    s_logger.debug(String.format("Removing job - id: %d, pid: %d, job type: %d, scheduled: %d, heartbeat: %s",
+                            job.getId(), job.getPid(), job.getJobType(), job.getScheduled(), job.getHeartbeat()));
+                    remove(job.getId());
+                }
+            }
+        } finally {
+            txn.close();
+        }
+    }
 }

usage/src/main/java/com/cloud/usage/UsageManagerImpl.java

@@ -324,6 +324,9 @@ public boolean start() {
             logger.info("Starting Usage Manager");
         }
 
+        _usageJobDao.removeLastOpenJobsOwned(_hostname, 0);
+        Runtime.getRuntime().addShutdownHook(new AbandonJob());
+
         // use the configured exec time and aggregation duration for scheduling the job
         _scheduledFuture =
                 _executor.scheduleAtFixedRate(this, _jobExecTime.getTimeInMillis() - System.currentTimeMillis(), _aggregationDuration * 60 * 1000, TimeUnit.MILLISECONDS);
@@ -336,7 +339,6 @@ public boolean start() {
             _sanity = _sanityExecutor.scheduleAtFixedRate(new SanityCheck(), 1, _sanityCheckInterval, TimeUnit.DAYS);
         }
 
-        Runtime.getRuntime().addShutdownHook(new AbandonJob());
         TransactionLegacy usageTxn = TransactionLegacy.open(TransactionLegacy.USAGE_DB);
         try {
             if (_heartbeatLock.lock(3)) { // 3 second timeout
@@ -2263,17 +2265,17 @@ protected void runInContext() {
                         // the aggregation range away from executing the next job
                         long now = System.currentTimeMillis();
                         long timeToJob = _jobExecTime.getTimeInMillis() - now;
-                        long timeSinceJob = 0;
+                        long timeSinceLastSuccessJob = 0;
                         long aggregationDurationMillis = _aggregationDuration * 60L * 1000L;
                         long lastSuccess = _usageJobDao.getLastJobSuccessDateMillis();
                         if (lastSuccess > 0) {
-                            timeSinceJob = now - lastSuccess;
+                            timeSinceLastSuccessJob = now - lastSuccess;
                         }
 
-                        if ((timeSinceJob > 0) && (timeSinceJob > (aggregationDurationMillis - 100))) {
+                        if ((timeSinceLastSuccessJob > 0) && (timeSinceLastSuccessJob > (aggregationDurationMillis - 100))) {
                             if (timeToJob > (aggregationDurationMillis / 2)) {
                                 if (logger.isDebugEnabled()) {
-                                    logger.debug("it's been " + timeSinceJob + " ms since last usage job and " + timeToJob +
+                                    logger.debug("it's been " + timeSinceLastSuccessJob + " ms since last usage job and " + timeToJob +
                                             " ms until next job, scheduling an immediate job to catch up (aggregation duration is " + _aggregationDuration + " minutes)");
                                 }
                                 scheduleParse();
@@ -2360,17 +2362,12 @@ protected void runInContext() {
             }
         }
     }
+
     private class AbandonJob extends Thread {
         @Override
         public void run() {
-            logger.info("exitting Usage Manager");
-            deleteOpenjob();
-        }
-        private void deleteOpenjob() {
-            UsageJobVO job = _usageJobDao.isOwner(_hostname, _pid);
-            if (job != null) {
-                _usageJobDao.remove(job.getId());
-            }
+            logger.info("exiting Usage Manager");
+            _usageJobDao.removeLastOpenJobsOwned(_hostname, _pid);
         }
     }
 }