I was just reading the Coin Metrics Newsletter #53, and it included a section about the recent mempool congestion. After mentioning transactions that got evicted due to the network's queue of unconfirmed transactions exceeding the default mempool limit, it mentions about transactions expiring due to the 14 day limit:
Secondly, transactions residing in the mempool for over two weeks expired. By default, Bitcoin Core nodes remove transactions from their mempool if no miner found transaction fees to be attractive enough to include them in a block over the last 336 hours (two weeks).
In total 1,627 transactions expired between May 25th, and May 30th. Only 35% of these resided in the mempool for two weeks. The remaining 65% likely spent unconfirmed parent transactions and became invalid as their parents expired.
In the last sentence (highlight added), the newsletter describes that transactions chained from expired transactions would get dropped from the mempool as well. Is that an accurate description of the mempool behavior?
If yes, let's say
txA was in the mempool for two weeks and got dropped, and
txB was a CPFP transaction spending an output of
txA which therefore got invalidated and dropped as well.
What would happen if the original sender rebroadcast
txB? Given that
txB inputs would appear to not exist without the context of
txA, would the peers of the sender request the preceding transaction
txA or just reject
txB? Would then the two transactions repropagate as a unit?