The two levels
| Construct | What it is | Protects against |
|---|---|---|
| Availability domain (AD) | One or more data centres with independent power, cooling and network | Data centre failure |
| Fault domain (FD) | A grouping of hardware within an AD (three per AD) | Rack/hardware failure and planned maintenance on one FD |
Many OCI regions (including Mumbai and Hyderabad) are single-AD. In those regions, HA within the region means spreading across fault domains, and region-level DR becomes more important.
Placement rules
Multi-AD region: put HA pairs in different ADs (and FDs). Example: web-1 in AD-1, web-2 in AD-2, web-3 in AD-3.
Single-AD region: put HA members in different FDs: FAULT-DOMAIN-1, -2, -3.
oci compute instance list --compartment-id <cmp> \
--query "data[].{name:\"display-name\",ad:\"availability-domain\",fd:\"fault-domain\"}" --output table
If you don't choose an FD, OCI picks one, and two "HA" nodes can end up in the same FD. Always set it explicitly.
Service scope cheat sheet
| Service | Scope | HA note |
|---|---|---|
| VCN, subnets (regional), route tables | Regional | Use regional subnets so resources can go in any AD |
| Load Balancer, Network Load Balancer | Regional, highly available | Backends across FDs/ADs |
| Compute instance | AD + FD | Pair across FDs/ADs |
| Block volume | AD | Attach only within the same AD; use replication/backups for others |
| Object Storage | Regional | Already redundant |
| Base DB RAC | AD, nodes in different FDs | See the RAC post |
| Autonomous Database | AD (with infrastructure redundancy) | Autonomous Data Guard for AD/region |
| File Storage | AD | Snapshots/replication for other ADs |
Maintenance
OCI performs infrastructure maintenance one fault domain at a time. With members in different FDs, a maintenance event affects at most one of them, and live migration usually avoids even that reboot.
Design check
- Every tier has 2+ members in different FDs (or ADs).
- Load balancer health checks on the application path.
- Shared state outside the instances (database, Object Storage, cache).
- For single-AD regions: a DR region plan, because an AD outage is a region outage.
Tip
Ask "what happens if FAULT-DOMAIN-2 goes away?" for each tier. If any answer is "the app stops", that tier isn't HA yet.