Database Administrator
High availability and disaster recovery questions
138 questions on High availability and disaster recovery for database administrators, easiest first.
Easy 39
- Full vs incremental backupeasy
- Logical vs physical backupeasy
- Proving backups workeasy
- RPO and RTOeasy
- Why add a read replicaeasy
- Downtime allowed by 99.9%easy
- Error budget for a 30-day montheasy
- Replicas after a bad DROP TABLEeasy
- The 3-2-1 backup ruleeasy
- Spotting a single point of failureeasy
- Active-passive vs active-activeeasy
- Making a PostgreSQL server a standbyeasy
- Copying data for a new replicaeasy
- Failures a 3-node quorum surviveseasy
- Why keep a delayed replicaeasy
- Turning a standby into the primaryeasy
- RDS backup retention limiteasy
- Where an RDS point-in-time restore goeseasy
- What a pilot light setup keeps runningeasy
- Planning the way backeasy
- Planned switchover vs failovereasy
- Worst-case data loss from backupseasy
- What an RPO of zero requireseasy
- Adding replicas to fix slow writeseasy
- Partitioning vs shardingeasy
- Why run DR game dayseasy
- Eleven nines of durabilityeasy
- Removing a server without dropping userseasy
- Keeping a shop usable during DB failovereasy
- Rebuilding the stack in a new regioneasy
- Restoring one dropped tableeasy
- Kafka topic replication for HAeasy
- Stopping calls to a failing dependencyeasy
- Standby fell too far behindeasy
- Servers flapping in and out of a pooleasy
- What hot standby meanseasy
- Seeing connected standbyseasy
- Grandfather-father-son retentioneasy
- Consistent mysqldump without lockseasy
Medium 55
- What PITR needsmedium
- Risk of asynchronous replicationmedium
- Measuring replica lagmedium
- Reports cancelled on the replicamedium
- Is pg_dump consistent?medium
- Restoring a Percona XtraBackupmedium
- Row vs statement binlog formatmedium
- Multi-AZ vs read replica on RDSmedium
- Terraform plan wants to replace the DBmedium
- Availability of components in seriesmedium
- Two redundant copies of a servicemedium
- Availability from MTBF and MTTRmedium
- Four or five members for etcdmedium
- Placing the tie-breakermedium
- What fencing protects againstmedium
- Sync replication that survives one replica lossmedium
- Reading your own write on a replicamedium
- Too many replicas on one primarymedium
- Lossless semi-sync in MySQLmedium
- Why GTIDs ease MySQL failovermedium
- Stopping writes to a MySQL replicamedium
- MySQL replica lags under heavy writesmedium
- Replica lag spike after a batch jobmedium
- Users see old data after savingmedium
- Connection string that finds the primarymedium
- Routing to the Patroni leadermedium
- How Aurora storage survives an AZ lossmedium
- Lowest-effort cross-region DR for RDSmedium
- Backups that ransomware cannot deletemedium
- Restoring to just before a bad deploymedium
- New timeline after promotionmedium
- Checking WAL archiving healthmedium
- InnoDB Cluster losing two of threemedium
- Fast new MySQL replicamedium
- MySQL point-in-time restoremedium
- Kafka settings that do not lose acked writesmedium
- Writes lost after Redis failovermedium
- Choosing during a network partitionmedium
- Primary on the minority side of a partitionmedium
- Last-writer-wins drops a newer updatemedium
- Commit latency with a distant sync replicamedium
- DR when data must stay in the UAEmedium
- DR for Indian payment datamedium
- Static stability during a zone failuremedium
- Limiting the blast radiusmedium
- Database overwhelmed right after failovermedium
- Sending reads to a lagging replicamedium
- Unique IDs on many shardsmedium
- Moving one big tenant to its own shardmedium
- Sync standby placed in the same zonemedium
- Shorter app impact during RDS failovermedium
- Which MySQL replica to promotemedium
- Surprise bill from replica readsmedium
- Roles missing after a restoremedium
- Making pg_restore use 8 coresmedium
Hard 44
- Choosing a shard keyhard
- Avoiding split-brainhard
- Disk filling with WALhard
- Commits hang with sync replicationhard
- Logical replication stopped after a migrationhard
- pg_wal growing without limithard
- Errant GTIDs block failoverhard
- Primary demoted when etcd is unreachablehard
- Rejoining the old primaryhard
- Restore takes longer than the RTOhard
- DR restore blocked by an encryption keyhard
- A WAL archive with silent gapshard
- Low traffic makes WAL archiving laghard
- Files needed for a restorehard
- Native incremental backups in PostgreSQL 17hard
- Speeding up point-in-time recoveryhard
- Snapshots of data and WAL volumeshard
- Keys moved when adding a shardhard
- All writes hit one shardhard
- How long Patroni waits before failoverhard
- Replica skipped during failoverhard
- CDC breaks after every failoverhard
- Kubernetes keeps killing the databasehard
- All replicas in one zonehard
- Capacity to survive losing a zonehard
- A transfer between two shardshard
- Joining a small table on every shardhard
- One celebrity account overloads a shardhard
- A DB blip removes every app serverhard
- Running PostgreSQL on Kuberneteshard
- Why Patroni uses a watchdoghard
- Aurora Global Database in a regional outagehard
- Multi-AZ instance vs Multi-AZ clusterhard
- One replica always lagshard
- The drill took 3 hours, not 30 minuteshard
- Choosing logical or physical replicationhard
- Routing users to the nearest healthy regionhard
- How long a 3 TB restore takeshard
- Starting a chaos experiment safelyhard
- Keeping WAL for replicas without slotshard
- Standby failures under ANY 2 of 4hard
- A base backup that will not starthard
- etcd split across two regionshard
- Standbys after pg_upgrade --linkhard