corruption due to lack of memory is not handled #19066

My node ran out of memory while processing a blocktxn:-

2020-05-24 01:52:51.790 recv blocktxn 00000000000000000003a7888f845933fb26c15993deb6092c44b43ef96fcc20 (631487) age=54s indexes=1036 size=1450359 peer=128 2020-05-24 01:52:51.867 made block 00000000000000000003a7888f845933fb26c15993deb6092c44b43ef96fcc20 size=1511809 txs: mempool=142+3 req=1036 2020-05-24 01:52:52.429 ProcessNewBlock: Calling ActivateBestChain() 2020-05-24 01:57:09.964

---

EXCEPTION: St9bad_alloc
std::bad_alloc
bitcoin in ProcessMessages()

2020-05-24 01:57:10.004 ProcessMessages(blocktxn, 1450359 bytes) FAILED peer=128

but now any subsequent blocks are being marked as invalid:-

2020-05-24 02:51:32.291 ERROR: ConnectBlock(): inputs missing/spent 2020-05-24 02:51:32.291 InvalidChainFound: invalid block=00000000000000000006cad2fa2e041afac8973d4ebbe606296c516f08e29c80 height=631488 log2_work=91.973518 date=2020-05-24 02:04:06 2020-05-24 02:51:32.291 InvalidChainFound: current best=00000000000000000003a7888f845933fb26c15993deb6092c44b43ef96fcc20 height=631487 log2_work=91.973498 date=2020-05-24 01:51:56 2020-05-24 02:51:32.291 ERROR: ConnectTip(): ConnectBlock 00000000000000000006cad2fa2e041afac8973d4ebbe606296c516f08e29c80 failed.

reconsiderblock does not allow it to fix the apparent corruption (using the latest master)

This is strange. I'm not sure how this can happen.

To prevent this from happening the new handler is overridden to prevent bad_alloc from ever being raised here: https://github.com/bitcoin/bitcoin/blob/master/src/init.cpp#L905 This was introduced in #9856 (first in v0.15.0).

@laanwj the corruption happened on a version prior to #9856 (v13.99) but I ran the latest master hoping it would be able to recover, and it didn't.

v13.99 is EOL and no longer supported