mirror of
https://github.com/GreptimeTeam/greptimedb.git
synced 2026-11-18 08:44:07 +00:00
* feat(log-store): fence object store WAL I/O and recovery Signed-off-by: WenyXu <wenymedia@gmail.com> * feat(datanode): bind object store WAL I/O to node leases Signed-off-by: WenyXu <wenymedia@gmail.com> * fix(log-store): scope lease fencing to pre-enablement support Signed-off-by: WenyXu <wenymedia@gmail.com> * feat(datanode): wire node lease fencing behind a cluster WAL gate Signed-off-by: WenyXu <wenymedia@gmail.com> * refactor(wal): clarify fixed object-store directory generation Use a deployment-neutral name for the directory generation shared by standalone and distributed WAL stores. Keep the value and persisted prefixes unchanged. Signed-off-by: WenyXu <wenymedia@gmail.com> * fix(log-store): refine node lease expiry and retry handling Report stop failures only for discarded backlog, preserve retry counts for unissued creates, and request route refresh on lease expiry. Install I/O permits once, clarify shutdown semantics, and log permission transitions. Add deterministic regressions for stopped probes and lazy delete polling. Signed-off-by: WenyXu <wenymedia@gmail.com> * fix(log-store): observe WAL I/O permission availability Report unavailable and available permission transitions at info level without treating the first grant as an expired lease. Track unavailable actors with a gauge that is cleared on recovery and exit, and cover repeated checks, both ack modes, and stop/drop cleanup. Signed-off-by: WenyXu <wenymedia@gmail.com> * docs(log-store): document WAL permission availability gauge Describe first-grant, idle, recovery and exit accounting in the RFC and align metric help terminology with the other object store logstore metrics. Signed-off-by: WenyXu <wenymedia@gmail.com> --------- Signed-off-by: WenyXu <wenymedia@gmail.com>