Built to keep working when a part fails
Database fallback, automatic sync-back, a warm mirror and a documented two-server failover design. The aim is that data written during an outage is not lost.
Five layers
Database fallback
The platform starts against your PostgreSQL primary. If it is unreachable after several retries, ThirdEYE switches to a local database and keeps ingesting, investigating and responding instead of stopping.
AvailableAutomatic sync-back
When the primary returns, a sync engine copies everything written during the outage back into it, without duplicates and keeping the newer edit, then moves the live connection over with no restart.
AvailableWarm-standby mirror
While the primary is healthy, a periodic mirror keeps a local copy of every table, so the fallback starts from recent data rather than empty.
Partly builtTwo-server PostgreSQL HA
A documented design using Patroni and etcd: automatic detection of a failed primary, promotion of the standby, and safe rejoin of the old primary, with a connection pooler routing to the current leader.
ConfigurableModels and tools degrade, not fail
Analysis walks a six-step fallback chain down to built-in rules, and the MCP gateway is optional, so a provider or gateway outage never stops alert handling.
What is tested
- The mirror copies, updates and deletes rows between PostgreSQL-style primary and the local copy.
- Recovery sync takes the newer edit when the same record changed on both sides.
- Fallback mode, mirror status and a manual "run mirror now" action are exposed in the platform.
Honest limits
- "No lost data" is the design goal, not a guarantee. The mirror runs on an interval, so the newest writes before a total loss of the primary may be missing from the local copy.
- The two-server Patroni design is documented configuration. It has not been validated on a two-machine failover test in the build environment, so test it on your own servers first.
- Standard PostgreSQL replication is asynchronous unless you configure it otherwise, which affects what a failover can lose.
Plan your failover with us.
Tell us your uptime and data-loss targets and we will say what the platform covers and what you need to add.