State-provisioning must flush parentchain import queue and set last imported parentchain header

[clangenb]

I have yet to conclude my analysis, but I think I found the cause of the failing integration tests in #1337.

As noted in #1337 (comment), w2 imports some missing parentchain blocks before importing the sidechain block. However, the imported blocks are [0...=n], where n is the layer_one_head denoted in the to be imported sidechain block.

This means, we import blocks that are already included in the provisioned sidechain state, and re-apply them. My current guess is that by chance this did not lead to state transitions pre polkadot-0.9.42 such that we did not note that bug before.

Logs observed on #1337 with the following log-levels:

'itp_stf_state_handler=debug,'
'its_consensus_common=debug,'
'its_consensus_aura=trace,'
'itc_parentchain_block_importer=debug,'

[Attempting to import sidechain block (number: 36, hash: 0xaf8bb6dbed80fd8cef2471b0e8d263c567b1285be4591097137c9a2944d98501, parentchain hash: 0x56fc141a2895a1cc2c738193cc13714dfe8ee5913b03ed889cdc0134df72b3ef)
[2023-06-15T14:32:14Z [DEBUG its_consensus_aura::block_importer] Peeking parentchain header
[2023-06-15T14:32:14Z [DEBUG its_consensus_aura::block_importer] sidechain block parentchain head: 0x56fc…b3ef
[2023-06-15T14:32:14Z [DEBUG its_consensus_aura::block_importer] last imported head: 0xc0f2…4314, number: 0, parenthash: 0x0000…0000
[2023-06-15T14:32:14Z [WARN  its_consensus_common::peer_block_sync] Got ancestry mismatch error upon block import. Attempting to fetch missing blocks from peer
[2023-06-15T14:32:14Z [INFO  its_consensus_common::peer_block_sync] Initiating fetch blocks from peer, last imported block hash: 0xb529cac5cec61d2e5281457c5ebd2edc350f03d1bd0914016a89e9454474f4db, until block hash: 0xaf8bb6dbed80fd8cef2471b0e8d263c567b1285be4591097137c9a2944d98501
[2023-06-15T14:32:14Z [INFO  its_consensus_common::peer_block_sync] Fetched 20 blocks from peer to import
[2023-06-15T14:32:14Z [DEBUG its_consensus_common::block_import] Attempting to import sidechain block (number: 16, hash: 0x95f38e6d6f1ff7d3a237562721828df06bced3e3ec8d14ad3db6ab2824814059, parentchain hash: 0x598897bfe33c9fdf5d9d3e9386121f7ec6476b7b6f3374b51469a1c3485a6c86)
[2023-06-15T14:32:14Z [DEBUG its_consensus_aura::block_importer] Peeking parentchain header
[2023-06-15T14:32:14Z [DEBUG its_consensus_aura::block_importer] sidechain block parentchain head: 0x5988…6c86
[2023-06-15T14:32:14Z [DEBUG its_consensus_aura::block_importer] last imported head: 0xc0f2…4314, number: 0, parenthash: 0x0000…0000
[2023-06-15T14:32:14Z [DEBUG itc_parentchain_block_importer::block_importer]Import blocks to light-client!
[2023-06-15T14:32:14Z [DEBUG itp_stf_state_handler::state_handler] Writing state
[2023-06-15T14:32:14Z [INFO  itc_parentchain_block_importer::block_importer] Successfully imported parentchain block (number: 1, hash: 0xfa79…7982)
[2023-06-15T14:32:14Z [DEBUG itp_stf_state_handler::state_handler] Writing state
[2023-06-15T14:32:14Z [INFO  itc_parentchain_block_importer::block_importer] Successfully imported parentchain block (number: 2, hash: 0xbe75…c85e)
[2023-06-15T14:32:14Z [DEBUG itp_stf_state_handler::state_handler] Writing state
[2023-06-15T14:32:14Z [INFO  itc_parentchain_block_importer::block_importer] Successfully imported parentchain block (number: 3, hash: 0x5776…8966)
[2023-06-15T14:32:14Z [DEBUG itp_stf_state_handler::state_handler] Writing state
[2023-06-15T14:32:14Z [INFO  itc_parentchain_block_importer::block_importer] Successfully imported parentchain block (number: 4, hash: 0x5988…6c86)
[2023-06-15T14:32:14Z [INFO  its_consensus_common::block_import] Applying state diff for block 16 of size 219 bytes
[2023-06-15T14:32:14Z [ERROR its_consensus_common::peer_block_sync] Failed to import sidechain block that was fetched from peer: Other(InvalidAprioriHash)
[2023-06-15T14:32:14Z enclave_runtime::error] Returning error Consensus(Other(InvalidAprioriHash)) as sgx unexpected.

[clangenb]

To be precise, I see two potential causes for the bug #1337:

1. The parentchain block imports did not lead to any additional state transition pre-polkadot-0.9.42, hence these unwanted imports were noops, which did not change our apriori hash.
2. 0.9.42 bricked some mechanism that already implemented what I suggest here, but I don't see yet where this mechanism could be in our codebase.

TBH, to me both seem unlikely, but I deem the first one more plausible.

[clangenb]

To be precise, I see two potential causes for the bug #1337:

1. The parentchain block imports did not lead to any additional state transition pre-polkadot-0.9.42, hence these unwanted imports were noops, which did not change our apriori hash.

2. 0.9.42 bricked some mechanism that already implemented what I suggest here, but I don't see yet where this mechanism could be in our codebase.

TBH, to me both seem unlikely, but I deem the first one more plausible.

It is definitely 1.

[clangenb]

The parentchain import queue is not populated at that step, as the parentchain light-client is not even initialized. Hence, the solution will be to provision the current light-client state together with the state-key and the state. Therefore this is superseded by #1370