1. Failures

<aside> 💡 Fail-Stop failures: falure ⇒ stop computer

</aside>

Replication can’t solve problems like:

• Logic bugs
• Configuration errors
• Malicious errors

And may solve problems like:

• Earth quake

2. Challenge

1. Has primary actually failed?
   • Can’t tell diff between network partition and computer fail
   • May cause split-brain system
2. How do we keep primary / backup in sync
   1. Apply all changes in the right order
   2. Deal with non-determinism
3. Fail over

3. Two Approaches

1. State transfer ⇒ Send snapshots to the backup
2. Replicated State Machine ⇒ Only send operations to the backup

<aside> 💡 Level of operations to replicate

• Application-level
  • File append, write
• Machine level ⇒ transparent!
  • Then application doesn’t need to be modified at all
  • Use virtual machines </aside>