[Tarantool-patches] [PATCH 4/7] txn_limbo: persist the latest effective promote in snapshot

[Serge Petrenko]

15.06.2021 23:59, Vladislav Shpilevoy пишет:
> Danke schön für der Patch!
>
> See 3 comments below.
>
>> diff --git a/src/box/memtx_engine.c b/src/box/memtx_engine.c
>> index 6c4982b9f..3713d39d0 100644
>> --- a/src/box/memtx_engine.c
>> +++ b/src/box/memtx_engine.c
>> @@ -225,6 +226,22 @@ memtx_engine_recover_raft(const struct xrow_header *row)
>>   	return 0;
>>   }
>>   
>> +static int
>> +memtx_engine_recover_promote(const struct xrow_header *row)
> 1. Maybe call it recover_synchro instead of promote? Even in the
> body PROMOTE is not mentioned anyhow except for the assertion.
> What if we ever add more members to the synchro state?

Ok.

>> +{
>> +	assert(row->type == IPROTO_PROMOTE);
>> +	struct synchro_request req;
>> +	if (xrow_decode_synchro(row, &req) != 0)
>> +		return -1;
>> +	/*
>> +	 * Origin id cannot be deduced from row.replica_id in a checkpoint,
>> +	 * because all it's rows have a zero replica_id.
>> +	 */
>> +	req.origin_id = req.replica_id;
>> +	txn_limbo_process(&txn_limbo, &req);
>> +	return 0;
>> +}
>> @@ -655,6 +676,17 @@ finish:
>>   	return rc;
>>   }
>>   
>> +static int
>> +checkpoint_write_promote(struct xlog *l, const struct synchro_request *req)
>> +{
>> +	struct xrow_header row;
>> +	char body[XROW_SYNCHRO_BODY_LEN_MAX];
>> +	xrow_encode_synchro(&row, body, req);
>> +	if (checkpoint_write_row(l, &row) != 0)
>> +		return -1;
>> +	return 0;
> 2. Could also make 'return checkpoint_write_row(...);'
> without branching + that might activate the tail call
> optimization.

Sure, thanks for noticing!

>> diff --git a/src/box/txn_limbo.h b/src/box/txn_limbo.h
>> index e409ac657..84a19bb40 100644
>> --- a/src/box/txn_limbo.h
>> +++ b/src/box/txn_limbo.h
>> @@ -311,6 +311,12 @@ txn_limbo_process(struct txn_limbo *limbo, const struct synchro_request *req);
>>   int
>>   txn_limbo_wait_confirm(struct txn_limbo *limbo);
>>   
>> +/**
>> + * Persist limbo state to a given synchro request.
>> + */
>> +void
>> +txn_limbo_checkpoint(struct txn_limbo *limbo, struct synchro_request *req);
> 3. Lets make the limbo a const pointer. It does not change, does it?

Ok. Please, find the incremental diff below.

==================================

diff --git a/src/box/memtx_engine.c b/src/box/memtx_engine.c
index a2ba17855..a2cfb2615 100644
--- a/src/box/memtx_engine.c
+++ b/src/box/memtx_engine.c
@@ -227,7 +227,7 @@ memtx_engine_recover_raft(const struct xrow_header *row)
  }

  static int
-memtx_engine_recover_promote(const struct xrow_header *row)
+memtx_engine_recover_synchro(const struct xrow_header *row)
  {
      assert(row->type == IPROTO_PROMOTE);
      struct synchro_request req;
@@ -235,7 +235,7 @@ memtx_engine_recover_promote(const struct 
xrow_header *row)
          return -1;
      /*
       * Origin id cannot be deduced from row.replica_id in a checkpoint,
-     * because all it's rows have a zero replica_id.
+     * because all its rows have a zero replica_id.
       */
      req.origin_id = req.replica_id;
      txn_limbo_process(&txn_limbo, &req);
@@ -251,7 +251,7 @@ memtx_engine_recover_snapshot_row(struct 
memtx_engine *memtx,
          if (row->type == IPROTO_RAFT)
              return memtx_engine_recover_raft(row);
          if (row->type == IPROTO_PROMOTE)
-            return memtx_engine_recover_promote(row);
+            return memtx_engine_recover_synchro(row);
          diag_set(ClientError, ER_UNKNOWN_REQUEST_TYPE,
               (uint32_t) row->type);
          return -1;
@@ -561,7 +561,7 @@ struct checkpoint {
      struct vclock vclock;
      struct xdir dir;
      struct raft_request raft;
-    struct synchro_request promote;
+    struct synchro_request synchro_state;
      /**
       * Do nothing, just touch the snapshot file - the
       * checkpoint already exists.
@@ -587,7 +587,7 @@ checkpoint_new(const char *snap_dirname, uint64_t 
snap_io_rate_limit)
      xdir_create(&ckpt->dir, snap_dirname, SNAP, &INSTANCE_UUID, &opts);
      vclock_create(&ckpt->vclock);
      box_raft_checkpoint_local(&ckpt->raft);
-    txn_limbo_checkpoint(&txn_limbo, &ckpt->promote);
+    txn_limbo_checkpoint(&txn_limbo, &ckpt->synchro_state);
      ckpt->touch = false;
      return ckpt;
  }
@@ -677,14 +677,12 @@ finish:
  }

  static int
-checkpoint_write_promote(struct xlog *l, const struct synchro_request *req)
+checkpoint_write_synchro(struct xlog *l, const struct synchro_request *req)
  {
      struct xrow_header row;
      char body[XROW_SYNCHRO_BODY_LEN_MAX];
      xrow_encode_synchro(&row, body, req);
-    if (checkpoint_write_row(l, &row) != 0)
-        return -1;
-    return 0;
+    return checkpoint_write_row(l, &row);
  }

  static int
@@ -724,7 +722,7 @@ checkpoint_f(va_list ap)
      }
      if (checkpoint_write_raft(&snap, &ckpt->raft) != 0)
          goto fail;
-    if (checkpoint_write_promote(&snap, &ckpt->promote) != 0)
+    if (checkpoint_write_synchro(&snap, &ckpt->synchro_state) != 0)
          goto fail;
      if (xlog_flush(&snap) < 0)
          goto fail;
diff --git a/src/box/txn_limbo.c b/src/box/txn_limbo.c
index 57f105ee3..c4ff5cc34 100644
--- a/src/box/txn_limbo.c
+++ b/src/box/txn_limbo.c
@@ -297,7 +297,8 @@ complete:
  }

  void
-txn_limbo_checkpoint(struct txn_limbo *limbo, struct synchro_request *req)
+txn_limbo_checkpoint(const struct txn_limbo *limbo,
+             struct synchro_request *req)
  {
      req->type = IPROTO_PROMOTE;
      req->replica_id = limbo->owner_id;
diff --git a/src/box/txn_limbo.h b/src/box/txn_limbo.h
index 84a19bb40..f5079d577 100644
--- a/src/box/txn_limbo.h
+++ b/src/box/txn_limbo.h
@@ -315,7 +315,8 @@ txn_limbo_wait_confirm(struct txn_limbo *limbo);
   * Persist limbo state to a given synchro request.
   */
  void
-txn_limbo_checkpoint(struct txn_limbo *limbo, struct synchro_request *req);
+txn_limbo_checkpoint(const struct txn_limbo *limbo,
+             struct synchro_request *req);

  /**
   * Write a PROMOTE request, which has the same effect as CONFIRM(@a 
lsn) and

==================================

-- 
Serge Petrenko