bedrock-kv / bedrock / 8d0bb980c046858373ddf0c36d5bbe2198c69125-PR-326

82%

develop: 82%

Build Type

Pull #326

github

Committed by jallum

Commit Message

bedrock-w10: Recover from the durable bootstrap, not the coordinator's copy

A coordinator loaded prior_core_state from the bootstrap once, at boot, and
afterwards learned only what its own directors notified it. A follower that
later won leadership handed its Director that boot-time value: nil if it
booted before any bootstrap existed, so recovery initialized a brand new
cluster and overwrote the real bootstrap; or a retired log set, so recovery
waited forever for logs that no longer exist.

The Director now reads the bootstrap itself when recovery starts, as FDB's
recovery reads cstate first. Only an absent record means fresh; a failed
read leaves the Director starting and retrying once per second (FDB's
MASTER_SPIN_DELAY) rather than initializing. Without object storage there
is no durable record, so the coordinator's copy - now only what its own
directors notified it - still stands there. The coordinator no longer
loads a core state at boot, and its director-launch trace no longer
reports one.

Pull Request Pull Request #326: bedrock-w10: recover from the durable bootstrap, not the coordinator's copy

Coverage Stats

17 of 25 new or added lines in 7 files covered. (68.0%)

2 existing lines in 2 files now uncovered.

7207 of 8812 relevant lines covered (81.79%)

1168.37 hits per line