Correct a problem where the maximum concurrent storage jobs counter gets out of sync...

[bacula/bacula] / bacula / src / dird / jobq.c

diff --git a/bacula/src/dird/jobq.c b/bacula/src/dird/jobq.c
index 768d6b34c8816a24a5b5e43e7dcaf29c49d24a90..c4a4c8cb8b53b922c6a3f87ea20759d5422a373c 100644 (file)
--- a/bacula/src/dird/jobq.c
+++ b/bacula/src/dird/jobq.c
@@ -1,22 +1,3 @@
-/*
- * Bacula job queue routines.
- *
- *  This code consists of three queues, the waiting_jobs
- *  queue, where jobs are initially queued, the ready_jobs
- *  queue, where jobs are placed when all the resources are
- *  allocated and they can immediately be run, and the
- *  running queue where jobs are placed when they are
- *  running.
- *
- *  Kern Sibbald, July MMIII
- *
- *   Version $Id$
- *
- *  This code was adapted from the Bacula workq, which was
- *    adapted from "Programming with POSIX Threads", by
- *    David R. Butenhof
- *
- */
 /*
    Bacula® - The Network Backup Solution
 
@@ -44,6 +25,25 @@
    (FSFE), Fiduciary Program, Sumatrastrasse 25, 8006 Zürich,
    Switzerland, email:ftf@fsfeurope.org.
 */
+/*
+ * Bacula job queue routines.
+ *
+ *  This code consists of three queues, the waiting_jobs
+ *  queue, where jobs are initially queued, the ready_jobs
+ *  queue, where jobs are placed when all the resources are
+ *  allocated and they can immediately be run, and the
+ *  running queue where jobs are placed when they are
+ *  running.
+ *
+ *  Kern Sibbald, July MMIII
+ *
+ *   Version $Id$
+ *
+ *  This code was adapted from the Bacula workq, which was
+ *    adapted from "Programming with POSIX Threads", by
+ *    David R. Butenhof
+ *
+ */
 
 #include "bacula.h"
 #include "dird.h"
@@ -453,6 +453,7 @@ void *jobq_server(void *arg)
             }
          }
          jq->running_jobs->append(je);
+         set_jcr_in_tsd(jcr);
          Dmsg1(2300, "Took jobid=%d from ready and appended to run\n", jcr->JobId);
 
          /* Release job queue lock */
@@ -682,14 +683,13 @@ static bool acquire_resources(JCR *jcr)
    jcr->acquired_resource_locks = false;
    if (jcr->rstore) {
       Dmsg1(200, "Rstore=%s\n", jcr->rstore->name());
-      if (jcr->rstore->NumConcurrentJobs == 0 &&
-          jcr->rstore->NumConcurrentJobs < jcr->rstore->MaxConcurrentJobs) {
-         /* Simple case, first job */
+      /*
+       * Let only one Restore/Verify job run at a time regardless
+       *  of MaxConcurrentjobs.
+       */
+      if (jcr->rstore->NumConcurrentJobs == 0) {
          jcr->rstore->NumConcurrentJobs = 1;
          Dmsg0(200, "Set rncj=1\n");
-      } else if (jcr->rstore->NumConcurrentJobs < jcr->rstore->MaxConcurrentJobs) {
-         jcr->rstore->NumConcurrentJobs++;
-         Dmsg1(200, "Inc rncj=%d\n", jcr->rstore->NumConcurrentJobs);
       } else {
          Dmsg1(200, "Fail rncj=%d\n", jcr->rstore->NumConcurrentJobs);
          set_jcr_job_status(jcr, JS_WaitStoreRes);
@@ -700,7 +700,7 @@ static bool acquire_resources(JCR *jcr)
    if (jcr->wstore) {
       Dmsg1(200, "Wstore=%s\n", jcr->wstore->name());
       if (jcr->rstore == jcr->wstore) {           /* deadlock */
-         jcr->rstore->NumConcurrentJobs--;        /* back out rstore */
+         jcr->rstore->NumConcurrentJobs = 0;      /* back out rstore */
          Jmsg(jcr, M_FATAL, 0, _("Job canceled. Attempt to read and write same device.\n"
             "    Read storage \"%s\" (From %s) -- Write storage \"%s\" (From %s)\n"), 
             jcr->rstore->name(), jcr->rstore_source, jcr->wstore->name(), jcr->wstore_source);
@@ -716,7 +716,7 @@ static bool acquire_resources(JCR *jcr)
          jcr->wstore->NumConcurrentJobs++;
          Dmsg1(200, "Inc wncj=%d\n", jcr->wstore->NumConcurrentJobs);
       } else if (jcr->rstore) {
-         jcr->rstore->NumConcurrentJobs--;        /* back out rstore */
+         jcr->rstore->NumConcurrentJobs = 0;      /* back out rstore */
          Dmsg1(200, "Fail wncj=%d\n", jcr->wstore->NumConcurrentJobs);
          skip_this_jcr = true;
       } else {
@@ -738,7 +738,7 @@ static bool acquire_resources(JCR *jcr)
          Dmsg1(200, "Dec wncj=%d\n", jcr->wstore->NumConcurrentJobs);
       }
       if (jcr->rstore) {
-         jcr->rstore->NumConcurrentJobs--;  
+         jcr->rstore->NumConcurrentJobs = 0;
          Dmsg1(200, "Dec rncj=%d\n", jcr->rstore->NumConcurrentJobs);
       }
       set_jcr_job_status(jcr, JS_WaitClientRes);
@@ -753,7 +753,7 @@ static bool acquire_resources(JCR *jcr)
          Dmsg1(200, "Dec wncj=%d\n", jcr->wstore->NumConcurrentJobs);
       }
       if (jcr->rstore) {
-         jcr->rstore->NumConcurrentJobs--;
+         jcr->rstore->NumConcurrentJobs = 0;
          Dmsg1(200, "Dec rncj=%d\n", jcr->rstore->NumConcurrentJobs);
       }
       jcr->client->NumConcurrentJobs--;