Database Specialty - Aurora Flashcards

Question

Fault injection queries – replica failure ALTER SYSTEM SIMULATE percentage_of_failure PERCENT READ REPLICA FAILURE [ TO ALL | TO "replica name" ] FOR INTERVAL quantity { YEAR | QUARTER | MONTH | WEEK | DAY | HOUR | MINUTE | SECOND };

Answer 1

* percentage_of_failure = % of requests to block * TO ALL / TO = simulate failure of all or a specific replica * quantity = duration of replica failure

Answer 2

* percentage_of_failure = % of the disk to mark as faulting * DISK index = simulate failure of a specific logical block of data * NODE index = simulate failure of a specific storage node * quantity = duration of disk failure

Answer 3

* percentage_of_failure = % of the disk to mark as congested * DISK index / NODE index = simulate failure of a specific disk or node * minimum / maximum = min and max amount of congestion delay in milliseconds (a random number between the two will be used) * quantity = duration of disk congestion

Answer 4

* Use CCM (apg_ccm_enabled=1) * Use cluster / reader / custom endpoint (instead of instance endpoints) * Cleanup / re-establish connections, if using instance endpoints * Add automatic retry capability to the application * Aggressively set TCP keepalives (=low values) - Quickly closes active connections if client is no longer able to connect to the DB * Reduce Java DNS Cache timeout value (low value for DNS TTL) * Reader endpoint cycles through available readers. If a client caches DNS info, requests might go to the old instance until DNS cache times out * Use separate connection objects for long and short running queries * Use Aurora PostgreSQL connection string with multiple hosts

Answer 5

* Use a list of hosts in your JDBC connection string * Or maintain a file containing cluster endpoints (reader / writer) * Your application can read this file to populate the host section of the connection string * JDBC connection driver will loop through all nodes on this list to find a valid connection * Set connection parameters (in red) aggressively so your app doesn’t wait too long on any host * Alternatively, you can also maintain a file containing instance endpoints * When you add/remove nodes, you must update this file

Answer 6

* Replication between clusters = can have more than 15 read replicas * Replication * between two Aurora MySQL DB clusters in different regions (Cross-Region Replication) * between two Aurora MySQL DB clusters in same region * between RDS MySQL DB instance and an Aurora MySQL DB cluster

Answer 7

* enable binary logging (binlog_format parameter) * then, create a cross-region read replica in another region * you can promote the replica to a standalone DB cluster (typically, for DR purposes)

Answer 8

* enable binary logging (binlog_format parameter) on source * then, replicate using a snapshot of the replication master

Answer 9

* By creating an Aurora read replica of RDS DB instance * Typically used for migration to Aurora rather than ongoing replication * To migrate, stop the writes on master. After replication lag is zero, promote the Aurora replica as a standalone Aurora DB cluster

Answer 10

* Give Aurora MySQL access to Lambda by setting DB cluster parameter aws_default_lambda_role = IAM role ARN * Option 1 – Using mysql.lambda_async procedure (deprecated) * Wrap calls in a stored procedure and call through triggers or application code * Option 2 – Using native functions lambda_sync and lambda_async * User must have INVOKE LAMBDA privilege * GRANT INVOKE LAMBDA ON *.* TO user@host

Answer 11

* Use SQL statements * LOAD DATA FROM S3, or * LOAD XML FROM S3 * Must give the Aurora cluster access to S3 by setting DB cluster parameter * aurora_load_from_s3_role = IAM role ARN, or * aws_default_s3_role = IAM role ARN * User must have LOAD FROM S3 privilege * GRANT LOAD FROM S3 ON *.* TO user@host

Answer 12

* Stopping an RDS DB * Can stop an RDS instance only if it does not have a replica * Cannot stop an RDS replica * Cannot stop RDS for SQL Server DB instance if it’s in Multi-AZ * Stopping an Aurora DB * Can stop the cluster, not the individual instances * Cannot manually stop Aurora Serverless * Cannot stop Aurora Multi-Master or Aurora Global DB cluster * Cannot stop a cluster if it uses parallel query * Cannot delete a stopped cluster without starting if first * If you don't manually start your DB instance/Cluster after seven days, it will be automatically started

Answer 13

* Maximum Connections in RDS or Aurora is controlled via parameter groups * Each DB engine has a specified formula for the default max connections value * You can override this value using a custom parameter group * Ideally, you’d want to scale the instanceto get higher max connections