Improve master startup time

Description

Currently the bulk of startup time is spent catching up the journal for each master. This is done sequentially for each master, taking at least 5 seconds per master (FileSystemMaster, MetaMaster, BlockMaster, MetricsMaster). There is no dependency between the journal replays, so we could significantly improve startup time by catching each master up in parallel.

Environment

None

Status

Assignee

Andrew Audibert

Reporter

Andrew Audibert

Labels

None

Components

Fix versions

Affects versions

1.8.1

Priority

Major
Configure