Uploaded image for project: 'Core Server'
  1. Core Server
  2. SERVER-41369

Txns with writes that span multiple PSA shards(disabled majority read concern) - return transient txn error

    • Fully Compatible
    • v4.2
    • Sharding 2019-06-17, Sharding 2019-07-01, Sharding 2019-07-15, Sharding 2019-07-29

      If read concern majority is disabled, transactions with writes that span multiple shards should error.

      But, when I have a txn that spans multiple shards (with rc majority disabled), I get a TransientTransactionError instead of a "'prepareTransaction' is not supported with 'enableMajorityReadConcern=false'".

      Steps to reproduce:

      1. Start a sharded cluster with 2 PSA shards (enableMajorityReadConcern=false for both shards)
      2. Connect mongo shell to the mongos and run:
        session = db.getMongo().startSession( { readPreference: { mode: "primary" } } );
        coll1 = session.getDatabase("mydb1").foo;
        coll2 = session.getDatabase("mydb2").bar;
        test = session.getDatabase("test").students2;
        
        // Using readconcern local and write concern majority
        session.startTransaction( { readConcern: { level: "local" }, writeConcern: { w: "majority" } } );
        
        try {
            coll1.insertOne( { abc: 1 } );        // ShardB
            coll2.insertOne( { xyz: 999 } );      // ShardB
            test.insertOne( { x: 1 })             // ShardA
        } catch (error) {
            // Abort transaction on error
            session.abortTransaction();
            throw error;
        }
        
         // Commit the transaction using write concern set at transaction start
        session.commitTransaction();
        
        session.endSession();
        
        

        This returns:

        MongoDB Enterprise mongos> session.commitTransaction();
        2019-05-29T13:24:28.121-0400 E  QUERY    [js] uncaught exception: Error: command failed: {
        	"errorLabels" : [
        		"TransientTransactionError"
        	],
        	"ok" : 0,
        	"errmsg" : "Transaction was aborted",
        	"code" : 251,
        	"codeName" : "NoSuchTransaction",
        	"operationTime" : Timestamp(1559150668, 4),
        	"$clusterTime" : {
        		"clusterTime" : Timestamp(1559150668, 4),
        		"signature" : {
        			"hash" : BinData(0,"jPzNbm4zF7knG/Lw+fOM/+PuNKw="),
        			"keyId" : NumberLong("6696440883090292753")
        		}
        	},
        	"recoveryToken" : {
        		"recoveryShardId" : "shardB"
        	}
        } 
        

        I would have expected this to error once and stop immediately (similar to what the txn does if using rc snapshot with rc majority disabled)

      Am attaching the log of shardB primary (had turned the verbosity level to 4)

      From the log, the txnCoordinator sends prepare txn to shardB again even after the coordinator received the assertion (which is the error I would expect to see bubble up and stop everything).

      2019-05-29T13:50:11.266-0400 D3 TXN      [TransactionCoordinator] Coordinator going to send command { prepareTransaction: 1, lsid: { id: UUID("003f3817-1a36-45c2-a439-d91dd9fffa16"), uid: BinData(0, DE9C6A9004CD518295FECA13EEAA8A134CC2D0115BD29073D69938C577121EC2) }, txnNumber: 0, autocommit: false, writeConcern: { w: "majority" } } to  shard shardA
      2019-05-29T13:50:11.268-0400 D3 TXN      [TransactionCoordinator] Coordinator going to send command { prepareTransaction: 1, lsid: { id: UUID("003f3817-1a36-45c2-a439-d91dd9fffa16"), uid: BinData(0, DE9C6A9004CD518295FECA13EEAA8A134CC2D0115BD29073D69938C577121EC2) }, txnNumber: 0, autocommit: false, writeConcern: { w: "majority" } } to  local  shard shardB
      2019-05-29T13:50:11.269-0400 D2 COMMAND  [TransactionCoordinator] run command admin.$cmd { prepareTransaction: 1, lsid: { id: UUID("003f3817-1a36-45c2-a439-d91dd9fffa16"), uid: BinData(0, DE9C6A9004CD518295FECA13EEAA8A134CC2D0115BD29073D69938C577121EC2) }, txnNumber: 0, autocommit: false, writeConcern: { w: "majority" }, $db: "admin" }
      2019-05-29T13:50:11.271-0400 D2 COMMAND  [TransactionCoordinator] command: prepareTransaction
      2019-05-29T13:50:11.272-0400 I  TXN      [TransactionCoordinator] transaction parameters:{ lsid: { id: UUID("003f3817-1a36-45c2-a439-d91dd9fffa16"), uid: BinData(0, DE9C6A9004CD518295FECA13EEAA8A134CC2D0115BD29073D69938C577121EC2) }, txnNumber: 0, autocommit: false, readConcern: { level: "snapshot" } }, readTimestamp:Timestamp(0, 0), ninserted:2 keysInserted:2 terminationCause:aborted timeActiveMicros:162 timeInactiveMicros:2089046 numYields:0 locks:{ ReplicationStateTransition: { acquireCount: { w: 4 } }, Global: { acquireCount: { r: 2, w: 1 } }, Database: { acquireCount: { r: 1, w: 2 } }, Collection: { acquireCount: { r: 1, w: 2 } }, Mutex: { acquireCount: { r: 9 } } } storage:{} wasPrepared:0, 2089ms
      2019-05-29T13:50:11.273-0400 D1 COMMAND  [TransactionCoordinator] assertion while executing command 'prepareTransaction' on database 'admin' with arguments '{ prepareTransaction: 1, lsid: { id: UUID("003f3817-1a36-45c2-a439-d91dd9fffa16"), uid: BinData(0, DE9C6A9004CD518295FECA13EEAA8A134CC2D0115BD29073D69938C577121EC2) }, txnNumber: 0, autocommit: false, writeConcern: { w: "majority" }, $db: "admin" }': Location50993: 'prepareTransaction' is not supported with 'enableMajorityReadConcern=false'
      2019-05-29T13:50:11.274-0400 I  COMMAND  [TransactionCoordinator] command admin.$cmd command: prepareTransaction { prepareTransaction: 1, lsid: { id: UUID("003f3817-1a36-45c2-a439-d91dd9fffa16"), uid: BinData(0, DE9C6A9004CD518295FECA13EEAA8A134CC2D0115BD29073D69938C577121EC2) }, txnNumber: 0, autocommit: false, writeConcern: { w: "majority" }, $db: "admin" } numYields:0 ok:0 errMsg:"'prepareTransaction' is not supported with 'enableMajorityReadConcern=false'" errName:Location50993 errCode:50993 reslen:460 locks:{ ReplicationStateTransition: { acquireCount: { w: 4 } }, Global: { acquireCount: { r: 2, w: 1 } }, Database: { acquireCount: { r: 1, w: 2 } }, Collection: { acquireCount: { r: 1, w: 2 } }, Mutex: { acquireCount: { r: 9 } } } flowControl:{ acquireCount: 1 } storage:{} protocol:op_msg 1ms
      2019-05-29T13:50:11.275-0400 D3 TXN      [TransactionCoordinator] Coordinator shard received Location50993: 'prepareTransaction' is not supported with 'enableMajorityReadConcern=false' from shard shardB for { prepareTransaction: 1, lsid: { id: UUID("003f3817-1a36-45c2-a439-d91dd9fffa16"), uid: BinData(0, DE9C6A9004CD518295FECA13EEAA8A134CC2D0115BD29073D69938C577121EC2) }, txnNumber: 0, autocommit: false, writeConcern: { w: "majority" } }
      2019-05-29T13:50:11.276-0400 D3 TXN      [TransactionCoordinator] Coordinator going to send command { prepareTransaction: 1, lsid: { id: UUID("003f3817-1a36-45c2-a439-d91dd9fffa16"), uid: BinData(0, DE9C6A9004CD518295FECA13EEAA8A134CC2D0115BD29073D69938C577121EC2) }, txnNumber: 0, autocommit: false, writeConcern: { w: "majority" } } to  local  shard shardB
      2019-05-29T13:50:11.277-0400 D2 COMMAND  [TransactionCoordinator] run command admin.$cmd { prepareTransaction: 1, lsid: { id: UUID("003f3817-1a36-45c2-a439-d91dd9fffa16"), uid: BinData(0, DE9C6A9004CD518295FECA13EEAA8A134CC2D0115BD29073D69938C577121EC2) }, txnNumber: 0, autocommit: false, writeConcern: { w: "majority" }, $db: "admin" }
      2019-05-29T13:50:11.277-0400 D2 COMMAND  [TransactionCoordinator] command: prepareTransaction
      
      

            Assignee:
            kaloian.manassiev@mongodb.com Kaloian Manassiev
            Reporter:
            kay.kim@mongodb.com Kay Kim (Inactive)
            Votes:
            0 Vote for this issue
            Watchers:
            13 Start watching this issue

              Created:
              Updated:
              Resolved: