crmstore

Replication history

Scheduled test writes and local confirmations, recorded with the browser closed.

Page JSON

Loading observers…

Confirmation time by store

Elapsed time from this observer starting a test write until it reads that write from each local store. All durations use this observer’s monotonic clock.

Observation log

Raft sends writes from the elected leader to its followers. These are observer → store checks through that leader, not an A → B → C relay.

Started / changesLeader / write

Each VM checks every 10 seconds, keeps up to 7 days / 65,000 samples in its own observer.db, and waits up to 5 seconds for a probe. Confirmation times include forwarding, quorum commit, read requests and polling (25 ms while waiting). They are observed upper bounds, not pure network or exact apply latency. These tiny synthetic writes do not create contacts.

Failures stay in the log and are excluded from latency percentiles, never treated as zero. “Read failed” means the observer could not read that node; it does not prove the machine stopped. Gaps may mean this observer was down. Choose a surviving observer to inspect another node’s outage. History survives service restarts but is local to each VM and is not in CRM backups.