[mkoura]

Summary

Make the target protocol version of node upgrade testing configurable, so the upgrade run can also hard fork from Conway PV11 to Dijkstra PV12, e.g. for feature branches.

Target protocol version

• New upgrade_prot_ver input of the upgrade workflow (11 or 12, default 11), passed to the runner as UPGRADE_PROT_VER.
• runner/node_upgrade_pytest.sh takes the target version from UPGRADE_PROT_VER. It defaults to BASE_PROT_VER, so nightly upgrade runs keep the 11 -> 11 behavior (no hard fork).
• COMMAND_ERA follows the era the cluster is in. It switches to the target era after the hard fork unless the user sets it.
• Step2 and step3 add DijkstraGenesisFile / DijkstraGenesisHash to the configs, reusing the original Dijkstra genesis when there is one.
• The step2 and step3 dry runs use the base protocol version. Configs generated for PV12 would contain TestDijkstraHardForkAtEpoch = 0 while the cluster is still in Conway.
• ExperimentalProtocolsEnabled / ExperimentalHardForksEnabled are enabled when the node version is lower than the target protocol version (e.g. PV12 with node < 12.0.0). test_hardfork uses the same rule when deciding to skip.
• After step3's hard fork stage, the runner checks that the cluster really is at the target protocol version, so a skipped test_hardfork doesn't go unnoticed.

BLS keys after the hard fork

The cluster pools are registered in Conway, so they have no BLS keys after the hard fork to Dijkstra. The new test_register_pools_bls_keys runs in step3 when the target era is Dijkstra:

• generates a BLS key pair for each pool and re-registers the pool with it, keeping the other pool parameters
• puts the keys where the cardonnay start scripts pick them up and restarts the nodes
• checks that every pool has its BLS key registered from the next epoch

reregister_cluster_pool moved from test_bls_rotation.py to tests_dijkstra/bls.py so the upgrade test can reuse it. A new allow_xfail arg turns off the xfail on the known VRFKeyHashAlreadyRegistered ledger issue (ledger#6102). That issue doesn't apply after a hard fork, because the Conway -> Dijkstra translation records the pools' VRF key hashes.

Docs

agent_docs/tag_testing.md now describes how to set the target protocol version.