• Make transaction participants do the following upon receiving a startTransaction statement with a txnRetryCounter:
  • If the FCV is latest and the feature flag is enabled, attach supportTxnRetryCounter:true to the response to the startTransaction statement. After that, the transaction must continue to use that txnRetryCounter regardless of FCV or feature flag changes. 
  • Otherwise, attach supportTxnRetryCounter:false to the response for the statement if txnRetryCounter=0, or throw a TxnRetryCounterNotSupported error if txnRetryCounter>0.

• Remove InvalidOptions uassert for txnRetryCounter from initializeOperationSessionInfo().
• Remove the feature flag checks in op_observer_impl.cpp and transaction_coordinator_util.cpp, and make it such that txnRetryCounter is written to config.transactions and config.transaction_coordinators entries and transaction oplog entries if and only if it is initialized.

Then:

• Make mongos use the flag to determine if txnRetryCounter should be attached in future transaction statements.
• Add TxnRetryCounterNotSupported to the list of transient transaction errors so that mongos attaches the TransientTransactionError label to the response to the command in the transaction that was internally retried with a non-zero txnRetryCounter.