exchange

commit ba69ba689f60213f8c8a2433fd84d6f3319afd84
parent 26a20dafcb36935e4d2eb362c1da3ee066fd4899
Author: Christian Grothoff <grothoff@gnunet.org>
Date:   Mon, 17 Aug 2026 15:15:19 +0200

re-try on serialization failure

Diffstat:
M
src/auditor/taler-helper-auditor-wire-debit.c
 | 
52
++++++++++++++++++++++++++++++++++++++++++++++++----

1 file changed, 48 insertions(+), 4 deletions(-)
diff --git a/src/auditor/taler-helper-auditor-wire-debit.c b/src/auditor/taler-helper-auditor-wire-debit.c
@@ -206,11 +206,23 @@ struct ReserveClosure
 static struct GNUNET_CONTAINER_MultiHashMap *reserve_closures;
 
 /**
+ * How often do we retry the transaction after a serialization
+ * failure before we give up?
+ */
+#define MAX_RETRIES 3
+
+/**
  * Return value from main().
  */
 static int global_ret;
 
 /**
+ * Number of serialization failures we have retried since the last
+ * successfully committed transaction.
+ */
+static unsigned int retries;
+
+/**
  * State of the current database transaction with
  * the auditor DB.
  */
@@ -599,21 +611,49 @@ commit (enum GNUNET_DB_QueryStatus qs)
     goto handle_db_error;
   GNUNET_log (GNUNET_ERROR_TYPE_INFO,
               "Transaction concluded!\n");
+  retries = 0;
   if (1 == test_mode)
     GNUNET_SCHEDULER_shutdown ();
   return;
 handle_db_error:
   TALER_AUDITORDB_rollback (TALER_ARL_adb);
-  for (unsigned int max_retries = 3; max_retries>0; max_retries--)
+  if (GNUNET_DB_STATUS_SOFT_ERROR == qs)
   {
-    if (GNUNET_DB_STATUS_HARD_ERROR == qs)
-      break;
+    if (MAX_RETRIES < ++retries)
+    {
+      GNUNET_log (GNUNET_ERROR_TYPE_ERROR,
+                  "Failed to commit the transaction after %u retries, terminating\n",
+                  (unsigned int) MAX_RETRIES);
+      global_ret = EXIT_FAILURE;
+      GNUNET_SCHEDULER_shutdown ();
+      return;
+    }
     GNUNET_log (GNUNET_ERROR_TYPE_INFO,
                 "Serialization issue, trying again\n");
-    qs = begin_transaction ();
+    /* Any bank request still in flight belongs to the attempt we are
+       abandoning.  #process_debits() would leave it alone and start
+       nothing, so the restarted audit would never make progress. */
+    for (struct WireAccount *wa = wa_head;
+         NULL != wa;
+         wa = wa->next)
+    {
+      if (NULL != wa->dhh)
+      {
+        TALER_BANK_debit_history_cancel (wa->dhh);
+        wa->dhh = NULL;
+      }
+      if (NULL != wa->dhh_task)
+      {
+        GNUNET_SCHEDULER_cancel (wa->dhh_task);
+        wa->dhh_task = NULL;
+      }
+    }
+    if (GNUNET_DB_STATUS_HARD_ERROR != begin_transaction ())
+      return; /* the retry continues asynchronously from here */
   }
   GNUNET_log (GNUNET_ERROR_TYPE_ERROR,
               "Hard database error, terminating\n");
+  global_ret = EXIT_FAILURE;
   GNUNET_SCHEDULER_shutdown ();
 }
 
@@ -1662,6 +1702,10 @@ begin_transaction (void)
        NULL != wa;
        wa = wa->next)
   {
+    /* we are called again after a serialization failure, so the labels
+       may already have been computed */
+    GNUNET_free (wa->label_wire_out_serial_id);
+    GNUNET_free (wa->label_wire_off_out);
     GNUNET_asprintf (&wa->label_wire_out_serial_id,
                      "wire-%s-%s",
                      wa->ai->section_name,