Re: [Tarantool-patches] [PATCH 8/8] replication: write and read CONFIRM entries

[Serge Petrenko <sergepetrenko@tarantool.org>]

09.06.2020 15:20, Serge Petrenko пишет:
> Make txn_limbo write a CONFIRM entry as soon as a batch of entries
> receive their acks. CONFIRM entry is written to WAL and later replicated
> to all the replicas.
>
> Now replicas put synchronous transactions into txn_limbo and wait for
> corresponding confirmation entries to arrive and end up in their WAL
> before committing the transactions.
>
Added a new commit:

commit 88884031e6ea8ca836228432b400ca1557afafa6
Author: Serge Petrenko <sergepetrenko@tarantool.org>
Date:   Fri Jun 19 18:21:11 2020 +0300

     txn: rework synchronous tx on_rollback trigger

     Instead of allocating the trigger on heap, which would lead to use 
after
     free, allocate the trigger together with the txn.
     Also, clear the trigger when asynchronous journal write succeeds.

     [TO BE SQUASHED INTO THE PREVIOUS COMMIT]

diff --git a/src/box/txn.c b/src/box/txn.c
index 4f787db79..16856da0d 100644
--- a/src/box/txn.c
+++ b/src/box/txn.c
@@ -238,6 +238,7 @@ txn_begin(void)
      /* fiber_on_yield is initialized by engine on demand */
      trigger_create(&txn->fiber_on_stop, txn_on_stop, NULL, NULL);
      trigger_add(&fiber()->on_stop, &txn->fiber_on_stop);
+    trigger_create(&txn->on_write_failure, NULL, NULL, NULL);
      /*
       * By default all transactions may yield.
       * It's a responsibility of an engine to disable yields
@@ -458,6 +459,12 @@ txn_complete(struct txn *txn)
                           stop_tm - txn->start_tm);
          }
      } else {
+        /*
+         * Async write succeeded. Clear the trigger which
+         * would remove the corresponding txn_limbo entry
+         * in case of failure.
+         */
+        trigger_clear(&txn->on_write_failure);
          /*
           * Complete is called on every WAL operation
           * authored by this transaction. And it not always
@@ -681,10 +688,10 @@ txn_commit_async(struct txn *txn)
       * Set a trigger to abort waiting for confirm on WAL write
       * failure.
       */
+    trigger_create(&txn->on_write_failure, txn_limbo_on_rollback,
+               limbo_entry, NULL);
      if (is_sync) {
-        struct trigger trig;
-        trigger_create(&trig, txn_limbo_on_rollback, limbo_entry, NULL);
-        txn_on_rollback(txn, &trig);
+        txn_on_rollback(txn, &txn->on_write_failure);
      }
      return 0;
  }
diff --git a/src/box/txn.h b/src/box/txn.h
index e7705bb48..9efd6fd0d 100644
--- a/src/box/txn.h
+++ b/src/box/txn.h
@@ -228,6 +228,14 @@ struct txn {
       * in case a fiber stops (all engines).
       */
      struct trigger fiber_on_stop;
+    /**
+     * An on_rollback trigger for synchronous transactions
+     * removing the txn_limbo entry which would wait for
+     * confirmation otherwise.
+     * Is issued on asynchronous write failure and is cleared
+     * on write success.
+     */
+    struct trigger on_write_failure;
      /** Commit and rollback triggers. */
      struct rlist on_commit, on_rollback;
      /**