hyperledger-labs / fabric-smart-client / 40378131460

92%

Build Type

push

github

Committed by web-flow

Commit Message

fix(committer): catch up on config another replica committed before the first block (#2265)

Replicas of one FSC node share the vault, but each keeps its own membership service and orderer list. On start, ReloadConfigTransactions loads the Valid config transactions from the vault. Block delivery starts later and, until it receives its first block, resumes from the vault's last non-Busy transaction. If another replica commits a config block and later blocks in between, the resume point lies past that config block, so the restarting replica never receives it and keeps the previous configuration until the next config block.

Before committing its first block, the committer now applies every Valid config transaction in the vault from the sequence its membership service holds onwards. Replicas commit blocks in order, so every config block up to the resume point is Valid by then. The held configuration is applied again, which is idempotent, so a retry after a failed orderer update completes it. The catch-up runs inside the block retry, so a transient vault failure is retried like any other commit failure.

ReloadConfigTransactions and the catch-up share one loop, loadConfigs, which starts at a given sequence. The comment on the Busy case now describes replicas, and a read error no longer reports "failed setting configtx state".

Fixes #2263

Signed-off-by: Marcus Brandenburger <bur@zurich.ibm.com>

Coverage Stats

15 of 18 new or added lines in 2 files covered. (83.33%)

8 existing lines in 4 files now uncovered.

30036 of 32886 relevant lines covered (91.33%)

1210.86 hits per line